From fde510315baf84e33f689f12c5d21297b5321470 Mon Sep 17 00:00:00 2001 From: David McCullough Date: Thu, 7 Jun 2001 12:08:54 +0000 Subject: The m68k-elf compiler chokes on this code when compiling for PIC as compile_regex is one big function (relative function calls further than cpu32 can do). The solution was to re-order the code a little to reduce the size of these relative calls. So the total sum of the changes is: * Move compile_regex to the end of the file * make store_op1 an inline Unfortunately CVS diff doesn't show this and makes it look like the whole file has been severely hacked. It hasn't. --- libc/misc/regex/regex.c | 6423 +++++++++++++++++++++++------------------------ 1 file changed, 3211 insertions(+), 3212 deletions(-) (limited to 'libc/misc/regex/regex.c') diff --git a/libc/misc/regex/regex.c b/libc/misc/regex/regex.c index d14595dfd..350535fa1 100644 --- a/libc/misc/regex/regex.c +++ b/libc/misc/regex/regex.c @@ -1832,3898 +1832,3897 @@ int num_regs; #endif /* not MATCH_MAY_ALLOCATE */ -static boolean group_in_compile_stack _RE_ARGS((compile_stack_type - compile_stack, +/* Subroutines for `regex_compile'. */ - regnum_t regnum)); +/* Store OP at LOC followed by two-byte integer parameter ARG. */ -/* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. - Returns one of error codes defined in `regex.h', or zero for success. +static inline void store_op1(op, loc, arg) +re_opcode_t op; +unsigned char *loc; +int arg; +{ + *loc = (unsigned char) op; + STORE_NUMBER(loc + 1, arg); +} - Assumes the `allocated' (and perhaps `buffer') and `translate' - fields are set in BUFP on entry. - If it succeeds, results are put in BUFP (if it returns an error, the - contents of BUFP are undefined): - `buffer' is the compiled pattern; - `syntax' is set to SYNTAX; - `used' is set to the length of the compiled pattern; - `fastmap_accurate' is zero; - `re_nsub' is the number of subexpressions in PATTERN; - `not_bol' and `not_eol' are zero; +/* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ - The `fastmap' and `newline_anchor' fields are neither - examined nor set. */ +static void store_op2(op, loc, arg1, arg2) +re_opcode_t op; +unsigned char *loc; +int arg1, arg2; +{ + *loc = (unsigned char) op; + STORE_NUMBER(loc + 1, arg1); + STORE_NUMBER(loc + 3, arg2); +} -/* Return, freeing storage we allocated. */ -#define FREE_STACK_RETURN(value) \ - return (free (compile_stack.stack), value) -static reg_errcode_t regex_compile(pattern, size, syntax, bufp) -const char *pattern; -size_t size; -reg_syntax_t syntax; -struct re_pattern_buffer *bufp; +/* Copy the bytes from LOC to END to open up three bytes of space at LOC + for OP followed by two-byte integer parameter ARG. */ + +static void insert_op1(op, loc, arg, end) +re_opcode_t op; +unsigned char *loc; +int arg; +unsigned char *end; { - /* We fetch characters from PATTERN here. Even though PATTERN is - `char *' (i.e., signed), we declare these variables as unsigned, so - they can be reliably used as array indices. */ - register unsigned char c, c1; + register unsigned char *pfrom = end; + register unsigned char *pto = end + 3; - /* A random temporary spot in PATTERN. */ - const char *p1; + while (pfrom != loc) + *--pto = *--pfrom; - /* Points to the end of the buffer, where we should append. */ - register unsigned char *b; + store_op1(op, loc, arg); +} - /* Keeps track of unclosed groups. */ - compile_stack_type compile_stack; - /* Points to the current (ending) position in the pattern. */ - const char *p = pattern; - const char *pend = pattern + size; +/* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ - /* How to translate the characters in the pattern. */ - RE_TRANSLATE_TYPE translate = bufp->translate; +static void insert_op2(op, loc, arg1, arg2, end) +re_opcode_t op; +unsigned char *loc; +int arg1, arg2; +unsigned char *end; +{ + register unsigned char *pfrom = end; + register unsigned char *pto = end + 5; - /* Address of the count-byte of the most recently inserted `exactn' - command. This makes it possible to tell if a new exact-match - character can be added to that command or if the character requires - a new `exactn' command. */ - unsigned char *pending_exact = 0; + while (pfrom != loc) + *--pto = *--pfrom; - /* Address of start of the most recently finished expression. - This tells, e.g., postfix * where to find the start of its - operand. Reset at the beginning of groups and alternatives. */ - unsigned char *laststart = 0; + store_op2(op, loc, arg1, arg2); +} - /* Address of beginning of regexp, or inside of last group. */ - unsigned char *begalt; - /* Place in the uncompiled pattern (i.e., the {) to - which to go back if the interval is invalid. */ - const char *beg_interval; +/* P points to just after a ^ in PATTERN. Return true if that ^ comes + after an alternative or a begin-subexpression. We assume there is at + least one character before the ^. */ - /* Address of the place where a forward jump should go to the end of - the containing expression. Each alternative of an `or' -- except the - last -- ends with a forward jump of this sort. */ - unsigned char *fixup_alt_jump = 0; +static boolean at_begline_loc_p(pattern, p, syntax) +const char *pattern, *p; +reg_syntax_t syntax; +{ + const char *prev = p - 2; + boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; - /* Counts open-groups as they are encountered. Remembered for the - matching close-group on the compile stack, so the same register - number is put in the stop_memory as the start_memory. */ - regnum_t regnum = 0; + return + /* After a subexpression? */ + (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) + /* After an alternative? */ + || (*prev == '|' + && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); +} -#ifdef DEBUG - DEBUG_PRINT1("\nCompiling pattern: "); - if (debug) { - unsigned debug_count; - for (debug_count = 0; debug_count < size; debug_count++) - putchar(pattern[debug_count]); - putchar('\n'); - } -#endif /* DEBUG */ +/* The dual of at_begline_loc_p. This one is for $. We assume there is + at least one character after the $, i.e., `P < PEND'. */ - /* Initialize the compile stack. */ - compile_stack.stack = - TALLOC(INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); - if (compile_stack.stack == NULL) - return REG_ESPACE; +static boolean at_endline_loc_p(p, pend, syntax) +const char *p, *pend; +reg_syntax_t syntax; +{ + const char *next = p; + boolean next_backslash = *next == '\\'; + const char *next_next = p + 1 < pend ? p + 1 : 0; - compile_stack.size = INIT_COMPILE_STACK_SIZE; - compile_stack.avail = 0; + return + /* Before a subexpression? */ + (syntax & RE_NO_BK_PARENS ? *next == ')' + : next_backslash && next_next && *next_next == ')') + /* Before an alternative? */ + || (syntax & RE_NO_BK_VBAR ? *next == '|' + : next_backslash && next_next && *next_next == '|'); +} - /* Initialize the pattern buffer. */ - bufp->syntax = syntax; - bufp->fastmap_accurate = 0; - bufp->not_bol = bufp->not_eol = 0; - /* Set `used' to zero, so that if we return an error, the pattern - printer (for debugging) will think there's no pattern. We reset it - at the end. */ - bufp->used = 0; +/* Returns true if REGNUM is in one of COMPILE_STACK's elements and + false if it's not. */ - /* Always count groups, whether or not bufp->no_sub is set. */ - bufp->re_nsub = 0; +static boolean group_in_compile_stack _RE_ARGS((compile_stack_type + compile_stack, + regnum_t regnum)); -#if !defined emacs && !defined SYNTAX_TABLE - /* Initialize the syntax table. */ - init_syntax_once(); -#endif +static boolean group_in_compile_stack(compile_stack, regnum) +compile_stack_type compile_stack; +regnum_t regnum; +{ + int this_element; - if (bufp->allocated == 0) { - if (bufp->buffer) { /* If zero allocated, but buffer is non-null, try to realloc - enough space. This loses if buffer's address is bogus, but - that is the user's responsibility. */ - RETALLOC(bufp->buffer, INIT_BUF_SIZE, unsigned char); - } else { /* Caller did not allocate a buffer. Do it for them. */ - bufp->buffer = TALLOC(INIT_BUF_SIZE, unsigned char); - } - if (!bufp->buffer) - FREE_STACK_RETURN(REG_ESPACE); + for (this_element = compile_stack.avail - 1; + this_element >= 0; this_element--) + if (compile_stack.stack[this_element].regnum == regnum) + return true; - bufp->allocated = INIT_BUF_SIZE; - } + return false; +} - begalt = b = bufp->buffer; - /* Loop through the uncompiled pattern until we're at the end. */ - while (p != pend) { - PATFETCH(c); +/* Read the ending character of a range (in a bracket expression) from the + uncompiled pattern *P_PTR (which ends at PEND). We assume the + starting character is in `P[-2]'. (`P[-1]' is the character `-'.) + Then we set the translation of all bits between the starting and + ending characters (inclusive) in the compiled pattern B. - switch (c) { - case '^': - { - if ( /* If at start of pattern, it's an operator. */ - p == pattern + 1 - /* If context independent, it's an operator. */ - || syntax & RE_CONTEXT_INDEP_ANCHORS - /* Otherwise, depends on what's come before. */ - || at_begline_loc_p(pattern, p, syntax)) - BUF_PUSH(begline); - else - goto normal_char; - } - break; + Return an error code. + We use these short variable names so we can use the same macros as + `regex_compile' itself. */ - case '$': - { - if ( /* If at end of pattern, it's an operator. */ - p == pend - /* If context independent, it's an operator. */ - || syntax & RE_CONTEXT_INDEP_ANCHORS - /* Otherwise, depends on what's next. */ - || at_endline_loc_p(p, pend, syntax)) - BUF_PUSH(endline); - else - goto normal_char; - } - break; +static reg_errcode_t compile_range(p_ptr, pend, translate, syntax, b) +const char **p_ptr, *pend; +RE_TRANSLATE_TYPE translate; +reg_syntax_t syntax; +unsigned char *b; +{ + unsigned this_char; + const char *p = *p_ptr; + reg_errcode_t ret; + char range_start[2]; + char range_end[2]; + char ch[2]; - case '+': - case '?': - if ((syntax & RE_BK_PLUS_QM) - || (syntax & RE_LIMITED_OPS)) - goto normal_char; - handle_plus: - case '*': - /* If there is no previous pattern... */ - if (!laststart) { - if (syntax & RE_CONTEXT_INVALID_OPS) - FREE_STACK_RETURN(REG_BADRPT); - else if (!(syntax & RE_CONTEXT_INDEP_OPS)) - goto normal_char; - } + if (p == pend) + return REG_ERANGE; - { - /* Are we optimizing this jump? */ - boolean keep_string_p = false; + /* Fetch the endpoints without translating them; the + appropriate translation is done in the bit-setting loop below. */ + range_start[0] = p[-2]; + range_start[1] = '\0'; + range_end[0] = p[0]; + range_end[1] = '\0'; - /* 1 means zero (many) matches is allowed. */ - char zero_times_ok = 0, many_times_ok = 0; + /* Have to increment the pointer into the pattern string, so the + caller isn't still at the ending character. */ + (*p_ptr)++; - /* If there is a sequence of repetition chars, collapse it - down to just one (the right one). We can't combine - interval operators with these because of, e.g., `a{2}*', - which should only match an even number of `a's. */ + /* Report an error if the range is empty and the syntax prohibits this. */ + ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; - for (;;) { - zero_times_ok |= c != '+'; - many_times_ok |= c != '?'; + /* Here we see why `this_char' has to be larger than an `unsigned + char' -- we would otherwise go into an infinite loop, since all + characters <= 0xff. */ + ch[1] = '\0'; + for (this_char = 0; this_char <= (unsigned char) -1; ++this_char) { + ch[0] = this_char; + if (strcoll(range_start, ch) <= 0 && strcoll(ch, range_end) <= 0) { + SET_LIST_BIT(TRANSLATE(this_char)); + ret = REG_NOERROR; + } + } - if (p == pend) - break; + return ret; +} + +/* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in + BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible + characters can start a string that matches the pattern. This fastmap + is used by re_search to skip quickly over impossible starting points. - PATFETCH(c); + The caller must supply the address of a (1 << BYTEWIDTH)-byte data + area as BUFP->fastmap. - if (c == '*' - || (!(syntax & RE_BK_PLUS_QM) - && (c == '+' || c == '?'))); + We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in + the pattern buffer. - else if (syntax & RE_BK_PLUS_QM && c == '\\') { - if (p == pend) - FREE_STACK_RETURN(REG_EESCAPE); + Returns 0 if we succeed, -2 if an internal error. */ - PATFETCH(c1); - if (!(c1 == '+' || c1 == '?')) { - PATUNFETCH; - PATUNFETCH; - break; - } +int re_compile_fastmap(bufp) +struct re_pattern_buffer *bufp; +{ + int j, k; - c = c1; - } else { - PATUNFETCH; - break; - } +#ifdef MATCH_MAY_ALLOCATE + fail_stack_type fail_stack; +#endif +#ifndef REGEX_MALLOC + char *destination; +#endif - /* If we get here, we found another repeat character. */ - } + register char *fastmap = bufp->fastmap; + unsigned char *pattern = bufp->buffer; + unsigned char *p = pattern; + register unsigned char *pend = pattern + bufp->used; - /* Star, etc. applied to an empty pattern is equivalent - to an empty pattern. */ - if (!laststart) - break; +#ifdef REL_ALLOC + /* This holds the pointer to the failure stack, when + it is allocated relocatably. */ + fail_stack_elt_t *failure_stack_ptr; +#endif - /* Now we know whether or not zero matches is allowed - and also whether or not two or more matches is allowed. */ - if (many_times_ok) { /* More than one repetition is allowed, so put in at the - end a backward relative jump from `b' to before the next - jump we're going to put in below (which jumps from - laststart to after this jump). + /* Assume that each path through the pattern can be null until + proven otherwise. We set this false at the bottom of switch + statement, to which we get only if a particular path doesn't + match the empty string. */ + boolean path_can_be_null = true; - But if we are at the `*' in the exact sequence `.*\n', - insert an unconditional jump backwards to the ., - instead of the beginning of the loop. This way we only - push a failure point once, instead of every time - through the loop. */ - assert(p - 1 > pattern); + /* We aren't doing a `succeed_n' to begin with. */ + boolean succeed_n_p = false; - /* Allocate the space for the jump. */ - GET_BUFFER_SPACE(3); + assert(fastmap != NULL && p != NULL); - /* We know we are not at the first character of the pattern, - because laststart was nonzero. And we've already - incremented `p', by the way, to be the character after - the `*'. Do we have to do something analogous here - for null bytes, because of RE_DOT_NOT_NULL? */ - if (TRANSLATE(*(p - 2)) == TRANSLATE('.') - && zero_times_ok - && p < pend && TRANSLATE(*p) == TRANSLATE('\n') - && !(syntax & RE_DOT_NEWLINE)) { /* We have .*\n. */ - STORE_JUMP(jump, b, laststart); - keep_string_p = true; - } else - /* Anything else. */ - STORE_JUMP(maybe_pop_jump, b, laststart - 3); + INIT_FAIL_STACK(); + bzero(fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ + bufp->fastmap_accurate = 1; /* It will be when we're done. */ + bufp->can_be_null = 0; - /* We've added more stuff to the buffer. */ - b += 3; - } + while (1) { + if (p == pend || *p == succeed) { + /* We have reached the (effective) end of pattern. */ + if (!FAIL_STACK_EMPTY()) { + bufp->can_be_null |= path_can_be_null; - /* On failure, jump from laststart to b + 3, which will be the - end of the buffer after this jump is inserted. */ - GET_BUFFER_SPACE(3); - INSERT_JUMP(keep_string_p ? on_failure_keep_string_jump - : on_failure_jump, laststart, b + 3); - pending_exact = 0; - b += 3; + /* Reset for next path. */ + path_can_be_null = true; - if (!zero_times_ok) { - /* At least one repetition is required, so insert a - `dummy_failure_jump' before the initial - `on_failure_jump' instruction of the loop. This - effects a skip over that instruction the first time - we hit that loop. */ - GET_BUFFER_SPACE(3); - INSERT_JUMP(dummy_failure_jump, laststart, - laststart + 6); - b += 3; - } - } - break; + p = fail_stack.stack[--fail_stack.avail].pointer; + continue; + } else + break; + } - case '.': - laststart = b; - BUF_PUSH(anychar); - break; + /* We should never be about to go beyond the end of the pattern. */ + assert(p < pend); + switch (SWITCH_ENUM_CAST((re_opcode_t) * p++)) { - case '[': - { - boolean had_char_class = false; + /* I guess the idea here is to simply not bother with a fastmap + if a backreference is used, since it's too hard to figure out + the fastmap for the corresponding group. Setting + `can_be_null' stops `re_search_2' from using the fastmap, so + that is all we do. */ + case duplicate: + bufp->can_be_null = 1; + goto done; - if (p == pend) - FREE_STACK_RETURN(REG_EBRACK); - /* Ensure that we have enough space to push a charset: the - opcode, the length count, and the bitset; 34 bytes in all. */ - GET_BUFFER_SPACE(34); + /* Following are the cases which match a character. These end + with `break'. */ - laststart = b; + case exactn: + fastmap[p[1]] = 1; + break; - /* We test `*p == '^' twice, instead of using an if - statement, so we only need one BUF_PUSH. */ - BUF_PUSH(*p == '^' ? charset_not : charset); - if (*p == '^') - p++; - /* Remember the first position in the bracket expression. */ - p1 = p; + case charset: + for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) + if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) + fastmap[j] = 1; + break; - /* Push the number of bytes in the bitmap. */ - BUF_PUSH((1 << BYTEWIDTH) / BYTEWIDTH); - /* Clear the whole map. */ - bzero(b, (1 << BYTEWIDTH) / BYTEWIDTH); + case charset_not: + /* Chars beyond end of map must be allowed. */ + for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) + fastmap[j] = 1; - /* charset_not matches newline according to a syntax bit. */ - if ((re_opcode_t) b[-2] == charset_not - && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) SET_LIST_BIT('\n'); + for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) + if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) + fastmap[j] = 1; + break; - /* Read in characters and ranges, setting map bits. */ - for (;;) { - if (p == pend) - FREE_STACK_RETURN(REG_EBRACK); - PATFETCH(c); + case wordchar: + for (j = 0; j < (1 << BYTEWIDTH); j++) + if (SYNTAX(j) == Sword) + fastmap[j] = 1; + break; - /* \ might escape characters inside [...] and [^...]. */ - if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') { - if (p == pend) - FREE_STACK_RETURN(REG_EESCAPE); - PATFETCH(c1); - SET_LIST_BIT(c1); - continue; - } + case notwordchar: + for (j = 0; j < (1 << BYTEWIDTH); j++) + if (SYNTAX(j) != Sword) + fastmap[j] = 1; + break; - /* Could be the end of the bracket expression. If it's - not (i.e., when the bracket expression is `[]' so - far), the ']' character bit gets set way below. */ - if (c == ']' && p != p1 + 1) - break; - /* Look ahead to see if it's a range when the last thing - was a character class. */ - if (had_char_class && c == '-' && *p != ']') - FREE_STACK_RETURN(REG_ERANGE); + case anychar: + { + int fastmap_newline = fastmap['\n']; - /* Look ahead to see if it's a range when the last thing - was a character: if this is a hyphen not at the - beginning or the end of a list, then it's the range - operator. */ - if (c == '-' && !(p - 2 >= pattern && p[-2] == '[') - && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') - && *p != ']') { - reg_errcode_t ret - = compile_range(&p, pend, translate, syntax, b); + /* `.' matches anything ... */ + for (j = 0; j < (1 << BYTEWIDTH); j++) + fastmap[j] = 1; - if (ret != REG_NOERROR) - FREE_STACK_RETURN(ret); - } + /* ... except perhaps newline. */ + if (!(bufp->syntax & RE_DOT_NEWLINE)) + fastmap['\n'] = fastmap_newline; - else if (p[0] == '-' && p[1] != ']') { /* This handles ranges made up of characters only. */ - reg_errcode_t ret; + /* Return if we have already set `can_be_null'; if we have, + then the fastmap is irrelevant. Something's wrong here. */ + else if (bufp->can_be_null) + goto done; - /* Move past the `-'. */ - PATFETCH(c1); + /* Otherwise, have to check alternative paths. */ + break; + } - ret = compile_range(&p, pend, translate, syntax, b); - if (ret != REG_NOERROR) - FREE_STACK_RETURN(ret); - } +#ifdef emacs + case syntaxspec: + k = *p++; + for (j = 0; j < (1 << BYTEWIDTH); j++) + if (SYNTAX(j) == (enum syntaxcode) k) + fastmap[j] = 1; + break; - /* See if we're at the beginning of a possible character - class. */ - else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') { /* Leave room for the null. */ - char str[CHAR_CLASS_MAX_LENGTH + 1]; + case notsyntaxspec: + k = *p++; + for (j = 0; j < (1 << BYTEWIDTH); j++) + if (SYNTAX(j) != (enum syntaxcode) k) + fastmap[j] = 1; + break; - PATFETCH(c); - c1 = 0; - /* If pattern is `[[:'. */ - if (p == pend) - FREE_STACK_RETURN(REG_EBRACK); + /* All cases after this match the empty string. These end with + `continue'. */ - for (;;) { - PATFETCH(c); - if ((c == ':' && *p == ']') || p == pend) - break; - if (c1 < CHAR_CLASS_MAX_LENGTH) - str[c1++] = c; - else - /* This is in any case an invalid class name. */ - str[0] = '\0'; - } - str[c1] = '\0'; - /* If isn't a word bracketed by `[:' and `:]': - undo the ending character, the letters, and leave - the leading `:' and `[' (but set bits for them). */ - if (c == ':' && *p == ']') { -#if defined _LIBC || WIDE_CHAR_SUPPORT - boolean is_lower = STREQ(str, "lower"); - boolean is_upper = STREQ(str, "upper"); - wctype_t wt; - int ch; + case before_dot: + case at_dot: + case after_dot: + continue; +#endif /* emacs */ - wt = IS_CHAR_CLASS(str); - if (wt == 0) - FREE_STACK_RETURN(REG_ECTYPE); - /* Throw away the ] at the end of the character - class. */ - PATFETCH(c); + case no_op: + case begline: + case endline: + case begbuf: + case endbuf: + case wordbound: + case notwordbound: + case wordbeg: + case wordend: + case push_dummy_failure: + continue; - if (p == pend) - FREE_STACK_RETURN(REG_EBRACK); - for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) { -# ifdef _LIBC - if (__iswctype(__btowc(ch), wt)) - SET_LIST_BIT(ch); -# else - if (iswctype(btowc(ch), wt)) - SET_LIST_BIT(ch); -# endif + case jump_n: + case pop_failure_jump: + case maybe_pop_jump: + case jump: + case jump_past_alt: + case dummy_failure_jump: + EXTRACT_NUMBER_AND_INCR(j, p); + p += j; + if (j > 0) + continue; - if (translate && (is_upper || is_lower) - && (ISUPPER(ch) || ISLOWER(ch))) - SET_LIST_BIT(ch); - } + /* Jump backward implies we just went through the body of a + loop and matched nothing. Opcode jumped to should be + `on_failure_jump' or `succeed_n'. Just treat it like an + ordinary jump. For a * loop, it has pushed its failure + point already; if so, discard that as redundant. */ + if ((re_opcode_t) * p != on_failure_jump + && (re_opcode_t) * p != succeed_n) + continue; - had_char_class = true; -#else - int ch; - boolean is_alnum = STREQ(str, "alnum"); - boolean is_alpha = STREQ(str, "alpha"); - boolean is_blank = STREQ(str, "blank"); - boolean is_cntrl = STREQ(str, "cntrl"); - boolean is_digit = STREQ(str, "digit"); - boolean is_graph = STREQ(str, "graph"); - boolean is_lower = STREQ(str, "lower"); - boolean is_print = STREQ(str, "print"); - boolean is_punct = STREQ(str, "punct"); - boolean is_space = STREQ(str, "space"); - boolean is_upper = STREQ(str, "upper"); - boolean is_xdigit = STREQ(str, "xdigit"); + p++; + EXTRACT_NUMBER_AND_INCR(j, p); + p += j; - if (!IS_CHAR_CLASS(str)) - FREE_STACK_RETURN(REG_ECTYPE); + /* If what's on the stack is where we are now, pop it. */ + if (!FAIL_STACK_EMPTY() + && fail_stack.stack[fail_stack.avail - 1].pointer == p) + fail_stack.avail--; - /* Throw away the ] at the end of the character - class. */ - PATFETCH(c); + continue; - if (p == pend) - FREE_STACK_RETURN(REG_EBRACK); - for (ch = 0; ch < 1 << BYTEWIDTH; ch++) { - /* This was split into 3 if's to - avoid an arbitrary limit in some compiler. */ - if ((is_alnum && ISALNUM(ch)) - || (is_alpha && ISALPHA(ch)) - || (is_blank && ISBLANK(ch)) - || (is_cntrl && ISCNTRL(ch))) - SET_LIST_BIT(ch); - if ((is_digit && ISDIGIT(ch)) - || (is_graph && ISGRAPH(ch)) - || (is_lower && ISLOWER(ch)) - || (is_print && ISPRINT(ch))) - SET_LIST_BIT(ch); - if ((is_punct && ISPUNCT(ch)) - || (is_space && ISSPACE(ch)) - || (is_upper && ISUPPER(ch)) - || (is_xdigit && ISXDIGIT(ch))) - SET_LIST_BIT(ch); - if (translate && (is_upper || is_lower) - && (ISUPPER(ch) || ISLOWER(ch))) - SET_LIST_BIT(ch); - } - had_char_class = true; -#endif /* libc || wctype.h */ - } else { - c1++; - while (c1--) - PATUNFETCH; - SET_LIST_BIT('['); - SET_LIST_BIT(':'); - had_char_class = false; - } - } else { - had_char_class = false; - SET_LIST_BIT(c); + case on_failure_jump: + case on_failure_keep_string_jump: + handle_on_failure_jump: + EXTRACT_NUMBER_AND_INCR(j, p); + + /* For some patterns, e.g., `(a?)?', `p+j' here points to the + end of the pattern. We don't want to push such a point, + since when we restore it above, entering the switch will + increment `p' past the end of the pattern. We don't need + to push such a point since we obviously won't find any more + fastmap entries beyond `pend'. Such a pattern can match + the null string, though. */ + if (p + j < pend) { + if (!PUSH_PATTERN_OP(p + j, fail_stack)) { + RESET_FAIL_STACK(); + return -2; } + } else + bufp->can_be_null = 1; + + if (succeed_n_p) { + EXTRACT_NUMBER_AND_INCR(k, p); /* Skip the n. */ + succeed_n_p = false; } - /* Discard any (non)matching list bytes that are all 0 at the - end of the map. Decrease the map-length byte too. */ - while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) - b[-1]--; - b += b[-1]; - } - break; + continue; - case '(': - if (syntax & RE_NO_BK_PARENS) - goto handle_open; - else - goto normal_char; + case succeed_n: + /* Get to the number of times to succeed. */ + p += 2; + /* Increment p past the n for when k != 0. */ + EXTRACT_NUMBER_AND_INCR(k, p); + if (k == 0) { + p -= 4; + succeed_n_p = true; /* Spaghetti code alert. */ + goto handle_on_failure_jump; + } + continue; - case ')': - if (syntax & RE_NO_BK_PARENS) - goto handle_close; - else - goto normal_char; + case set_number_at: + p += 4; + continue; - case '\n': - if (syntax & RE_NEWLINE_ALT) - goto handle_alt; - else - goto normal_char; + case start_memory: + case stop_memory: + p += 2; + continue; - case '|': - if (syntax & RE_NO_BK_VBAR) - goto handle_alt; - else - goto normal_char; + default: + abort(); /* We have listed all the cases. */ + } /* switch *p++ */ - case '{': - if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) - goto handle_interval; - else - goto normal_char; + /* Getting here means we have found the possible starting + characters for one path of the pattern -- and that the empty + string does not match. We need not follow this path further. + Instead, look at the next alternative (remembered on the + stack), or quit if no more. The test at the top of the loop + does these things. */ + path_can_be_null = false; + p = pend; + } /* while p */ + /* Set `can_be_null' for the last path (also the first path, if the + pattern is empty). */ + bufp->can_be_null |= path_can_be_null; - case '\\': - if (p == pend) - FREE_STACK_RETURN(REG_EESCAPE); + done: + RESET_FAIL_STACK(); + return 0; +} /* re_compile_fastmap */ - /* Do not translate the character after the \, so that we can - distinguish, e.g., \B from \b, even if we normally would - translate, e.g., B to b. */ - PATFETCH_RAW(c); +#ifdef _LIBC +weak_alias(__re_compile_fastmap, re_compile_fastmap) +#endif + /* Set REGS to hold NUM_REGS registers, storing them in STARTS and + ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use + this memory for recording register information. STARTS and ENDS + must be allocated using the malloc library routine, and must each + be at least NUM_REGS * sizeof (regoff_t) bytes long. - switch (c) { - case '(': - if (syntax & RE_NO_BK_PARENS) - goto normal_backslash; + If NUM_REGS == 0, then subsequent matches should allocate their own + register data. - handle_open: - bufp->re_nsub++; - regnum++; + Unless this function is called, the first search or match using + PATTERN_BUFFER will allocate its own register data, without + freeing the old data. */ +void re_set_registers(bufp, regs, num_regs, starts, ends) +struct re_pattern_buffer *bufp; +struct re_registers *regs; +unsigned num_regs; +regoff_t *starts, *ends; +{ + if (num_regs) { + bufp->regs_allocated = REGS_REALLOCATE; + regs->num_regs = num_regs; + regs->start = starts; + regs->end = ends; + } else { + bufp->regs_allocated = REGS_UNALLOCATED; + regs->num_regs = 0; + regs->start = regs->end = (regoff_t *) 0; + } +} - if (COMPILE_STACK_FULL) { - RETALLOC(compile_stack.stack, compile_stack.size << 1, - compile_stack_elt_t); - if (compile_stack.stack == NULL) - return REG_ESPACE; +#ifdef _LIBC +weak_alias(__re_set_registers, re_set_registers) +#endif + /* Searching routines. */ +/* Like re_search_2, below, but only one string is specified, and + doesn't let you say where to stop matching. */ +int re_search(bufp, string, size, startpos, range, regs) +struct re_pattern_buffer *bufp; +const char *string; +int size, startpos, range; +struct re_registers *regs; +{ + return re_search_2(bufp, NULL, 0, string, size, startpos, range, + regs, size); +} - compile_stack.size <<= 1; - } +#ifdef _LIBC +weak_alias(__re_search, re_search) +#endif +/* Using the compiled pattern in BUFP->buffer, first tries to match the + virtual concatenation of STRING1 and STRING2, starting first at index + STARTPOS, then at STARTPOS + 1, and so on. - /* These are the values to restore when we hit end of this - group. They are all relative offsets, so that if the - whole pattern moves because of realloc, they will still - be valid. */ - COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; - COMPILE_STACK_TOP.fixup_alt_jump - = - fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; - COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; - COMPILE_STACK_TOP.regnum = regnum; + STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. - /* We will eventually replace the 0 with the number of - groups inner to this one. But do not push a - start_memory for groups beyond the last one we can - represent in the compiled pattern. */ - if (regnum <= MAX_REGNUM) { - COMPILE_STACK_TOP.inner_group_offset = - b - bufp->buffer + 2; - BUF_PUSH_3(start_memory, regnum, 0); - } + RANGE is how far to scan while trying to match. RANGE = 0 means try + only at STARTPOS; in general, the last start tried is STARTPOS + + RANGE. - compile_stack.avail++; + In REGS, return the indices of the virtual concatenation of STRING1 + and STRING2 that matched the entire BUFP->buffer and its contained + subexpressions. - fixup_alt_jump = 0; - laststart = 0; - begalt = b; - /* If we've reached MAX_REGNUM groups, then this open - won't actually generate any code, so we'll have to - clear pending_exact explicitly. */ - pending_exact = 0; - break; + Do not consider matching one past the index STOP in the virtual + concatenation of STRING1 and STRING2. + We return either the position in the strings at which the match was + found, -1 if no match, or -2 if error (such as failure + stack overflow). */ +int +re_search_2(bufp, string1, size1, string2, size2, startpos, range, regs, + stop) +struct re_pattern_buffer *bufp; +const char *string1, *string2; +int size1, size2; +int startpos; +int range; +struct re_registers *regs; +int stop; +{ + int val; + register char *fastmap = bufp->fastmap; + register RE_TRANSLATE_TYPE translate = bufp->translate; + int total_size = size1 + size2; + int endpos = startpos + range; - case ')': - if (syntax & RE_NO_BK_PARENS) - goto normal_backslash; + /* Check for out-of-range STARTPOS. */ + if (startpos < 0 || startpos > total_size) + return -1; - if (COMPILE_STACK_EMPTY) { - if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) - goto normal_backslash; - else - FREE_STACK_RETURN(REG_ERPAREN); - } + /* Fix up RANGE if it might eventually take us outside + the virtual concatenation of STRING1 and STRING2. + Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ + if (endpos < 0) + range = 0 - startpos; + else if (endpos > total_size) + range = total_size - startpos; - handle_close: - if (fixup_alt_jump) { /* Push a dummy failure point at the end of the - alternative for a possible future - `pop_failure_jump' to pop. See comments at - `push_dummy_failure' in `re_match_2'. */ - BUF_PUSH(push_dummy_failure); + /* If the search isn't to be a backwards one, don't waste time in a + search for a pattern that must be anchored. */ + if (bufp->used > 0 && range > 0 + && ((re_opcode_t) bufp->buffer[0] == begbuf + /* `begline' is like `begbuf' if it cannot match at newlines. */ + || ((re_opcode_t) bufp->buffer[0] == begline + && !bufp->newline_anchor))) { + if (startpos > 0) + return -1; + else + range = 1; + } +#ifdef emacs + /* In a forward search for something that starts with \=. + don't keep searching past point. */ + if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot + && range > 0) { + range = PT - startpos; + if (range <= 0) + return -1; + } +#endif /* emacs */ - /* We allocated space for this jump when we assigned - to `fixup_alt_jump', in the `handle_alt' case below. */ - STORE_JUMP(jump_past_alt, fixup_alt_jump, b - 1); - } + /* Update the fastmap now if not correct already. */ + if (fastmap && !bufp->fastmap_accurate) + if (re_compile_fastmap(bufp) == -2) + return -2; - /* See similar code for backslashed left paren above. */ - if (COMPILE_STACK_EMPTY) { - if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) - goto normal_char; - else - FREE_STACK_RETURN(REG_ERPAREN); - } + /* Loop through the string, looking for a place to start matching. */ + for (;;) { + /* If a fastmap is supplied, skip quickly over characters that + cannot be the start of a match. If the pattern can match the + null string, however, we don't need to skip characters; we want + the first null string. */ + if (fastmap && startpos < total_size && !bufp->can_be_null) { + if (range > 0) { /* Searching forwards. */ + register const char *d; + register int lim = 0; + int irange = range; - /* Since we just checked for an empty stack above, this - ``can't happen''. */ - assert(compile_stack.avail != 0); - { - /* We don't just want to restore into `regnum', because - later groups should continue to be numbered higher, - as in `(ab)c(de)' -- the second group is #2. */ - regnum_t this_group_regnum; + if (startpos < size1 && startpos + range >= size1) + lim = range - (size1 - startpos); - compile_stack.avail--; - begalt = - bufp->buffer + COMPILE_STACK_TOP.begalt_offset; - fixup_alt_jump = - COMPILE_STACK_TOP.fixup_alt_jump ? bufp->buffer + - COMPILE_STACK_TOP.fixup_alt_jump - 1 : 0; - laststart = - bufp->buffer + COMPILE_STACK_TOP.laststart_offset; - this_group_regnum = COMPILE_STACK_TOP.regnum; - /* If we've reached MAX_REGNUM groups, then this open - won't actually generate any code, so we'll have to - clear pending_exact explicitly. */ - pending_exact = 0; - - /* We're at the end of the group, so now we know how many - groups were inside this one. */ - if (this_group_regnum <= MAX_REGNUM) { - unsigned char *inner_group_loc - - = - bufp->buffer + - COMPILE_STACK_TOP.inner_group_offset; - - *inner_group_loc = regnum - this_group_regnum; - BUF_PUSH_3(stop_memory, this_group_regnum, - regnum - this_group_regnum); - } - } - break; - - - case '|': /* `\|'. */ - if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) - goto normal_backslash; - handle_alt: - if (syntax & RE_LIMITED_OPS) - goto normal_char; - - /* Insert before the previous alternative a jump which - jumps to this alternative if the former fails. */ - GET_BUFFER_SPACE(3); - INSERT_JUMP(on_failure_jump, begalt, b + 6); - pending_exact = 0; - b += 3; - - /* The alternative before this one has a jump after it - which gets executed if it gets matched. Adjust that - jump so it will jump to this alternative's analogous - jump (put in below, which in turn will jump to the next - (if any) alternative's such jump, etc.). The last such - jump jumps to the correct final destination. A picture: - _____ _____ - | | | | - | v | v - a | b | c - - If we are at `b', then fixup_alt_jump right now points to a - three-byte space after `a'. We'll put in the jump, set - fixup_alt_jump to right after `b', and leave behind three - bytes which we'll fill in when we get to after `c'. */ - - if (fixup_alt_jump) - STORE_JUMP(jump_past_alt, fixup_alt_jump, b); - - /* Mark and leave space for a jump after this alternative, - to be filled in later either by next alternative or - when know we're at the end of a series of alternatives. */ - fixup_alt_jump = b; - GET_BUFFER_SPACE(3); - b += 3; - - laststart = 0; - begalt = b; - break; - - - case '{': - /* If \{ is a literal. */ - if (!(syntax & RE_INTERVALS) - /* If we're at `\{' and it's not the open-interval - operator. */ - || ((syntax & RE_INTERVALS) - && (syntax & RE_NO_BK_BRACES)) || (p - 2 == pattern - && p == pend)) - goto normal_backslash; - - handle_interval: - { - /* If got here, then the syntax allows intervals. */ + d = + (startpos >= + size1 ? string2 - size1 : string1) + startpos; - /* At least (most) this many matches must be made. */ - int lower_bound = -1, upper_bound = -1; + /* Written out as an if-else to avoid testing `translate' + inside the loop. */ + if (translate) + while (range > lim && !fastmap[(unsigned char) + translate[ + (unsigned + char) *d++]]) + range--; + else + while (range > lim && !fastmap[(unsigned char) *d++]) + range--; - beg_interval = p - 1; + startpos += irange - range; + } else { /* Searching backwards. */ - if (p == pend) { - if (!(syntax & RE_INTERVALS) - && (syntax & RE_NO_BK_BRACES)) goto - unfetch_interval; - else - FREE_STACK_RETURN(REG_EBRACE); - } + register char c = (size1 == 0 || startpos >= size1 + ? string2[startpos - size1] + : string1[startpos]); - GET_UNSIGNED_NUMBER(lower_bound); + if (!fastmap[(unsigned char) TRANSLATE(c)]) + goto advance; + } + } - if (c == ',') { - GET_UNSIGNED_NUMBER(upper_bound); - if ((!(syntax & RE_NO_BK_BRACES) && c != '\\') - || ((syntax & RE_NO_BK_BRACES) && c != '}')) - FREE_STACK_RETURN(REG_BADBR); + /* If can't match the null string, and that's all we have left, fail. */ + if (range >= 0 && startpos == total_size && fastmap + && !bufp->can_be_null) return -1; - if (upper_bound < 0) - upper_bound = RE_DUP_MAX; - } else - /* Interval such as `{1}' => match exactly once. */ - upper_bound = lower_bound; + val = re_match_2_internal(bufp, string1, size1, string2, size2, + startpos, regs, stop); +#ifndef REGEX_MALLOC +# ifdef C_ALLOCA + alloca(0); +# endif +#endif - if (lower_bound < 0 || upper_bound > RE_DUP_MAX - || lower_bound > upper_bound) { - if (!(syntax & RE_INTERVALS) - && (syntax & RE_NO_BK_BRACES)) goto - unfetch_interval; - else - FREE_STACK_RETURN(REG_BADBR); - } + if (val >= 0) + return startpos; - if (!(syntax & RE_NO_BK_BRACES)) { - if (c != '\\') - FREE_STACK_RETURN(REG_EBRACE); + if (val == -2) + return -2; - PATFETCH(c); - } + advance: + if (!range) + break; + else if (range > 0) { + range--; + startpos++; + } else { + range++; + startpos--; + } + } + return -1; +} /* re_search_2 */ - if (c != '}') { - if (!(syntax & RE_INTERVALS) - && (syntax & RE_NO_BK_BRACES)) goto - unfetch_interval; - else - FREE_STACK_RETURN(REG_BADBR); - } +#ifdef _LIBC +weak_alias(__re_search_2, re_search_2) +#endif + /* This converts PTR, a pointer into one of the search strings `string1' + and `string2' into an offset from the beginning of that string. */ +#define POINTER_TO_OFFSET(ptr) \ + (FIRST_STRING_P (ptr) \ + ? ((regoff_t) ((ptr) - string1)) \ + : ((regoff_t) ((ptr) - string2 + size1))) +/* Macros for dealing with the split strings in re_match_2. */ +#define MATCHING_IN_FIRST_STRING (dend == end_match_1) +/* Call before fetching a character with *d. This switches over to + string2 if necessary. */ +#define PREFETCH() \ + while (d == dend) \ + { \ + /* End of string2 => fail. */ \ + if (dend == end_match_2) \ + goto fail; \ + /* End of string1 => advance to string2. */ \ + d = string2; \ + dend = end_match_2; \ + } +/* Test if at very beginning or at very end of the virtual concatenation + of `string1' and `string2'. If only one string, it's `string2'. */ +#define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) +#define AT_STRINGS_END(d) ((d) == end2) +/* Test if D points to a character which is word-constituent. We have + two special cases to check for: if past the end of string1, look at + the first character in string2; and if before the beginning of + string2, look at the last character in string1. */ +#define WORDCHAR_P(d) \ + (SYNTAX ((d) == end1 ? *string2 \ + : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ + == Sword) +/* Disabled due to a compiler bug -- see comment at case wordbound */ +#if 0 +/* Test if the character before D and the one at D differ with respect + to being word-constituent. */ +#define AT_WORD_BOUNDARY(d) \ + (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ + || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) +#endif +/* Free everything we malloc. */ +#ifdef MATCH_MAY_ALLOCATE +# define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL +# define FREE_VARIABLES() \ + do { \ + REGEX_FREE_STACK (fail_stack.stack); \ + FREE_VAR (regstart); \ + FREE_VAR (regend); \ + FREE_VAR (old_regstart); \ + FREE_VAR (old_regend); \ + FREE_VAR (best_regstart); \ + FREE_VAR (best_regend); \ + FREE_VAR (reg_info); \ + FREE_VAR (reg_dummy); \ + FREE_VAR (reg_info_dummy); \ + } while (0) +#else +# define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ +#endif /* not MATCH_MAY_ALLOCATE */ +/* These values must meet several constraints. They must not be valid + register values; since we have a limit of 255 registers (because + we use only one byte in the pattern for the register number), we can + use numbers larger than 255. They must differ by 1, because of + NUM_FAILURE_ITEMS above. And the value for the lowest register must + be larger than the value for the highest register, so we do not try + to actually save any registers when none are active. */ +#define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) +#define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) + /* Matching routines. */ +#ifndef emacs /* Emacs never uses this. */ +/* re_match is like re_match_2 except it takes only a single string. */ +int re_match(bufp, string, size, pos, regs) +struct re_pattern_buffer *bufp; +const char *string; +int size, pos; +struct re_registers *regs; +{ + int result = re_match_2_internal(bufp, NULL, 0, string, size, + pos, regs, size); - /* We just parsed a valid interval. */ +# ifndef REGEX_MALLOC +# ifdef C_ALLOCA + alloca(0); +# endif +# endif + return result; +} - /* If it's invalid to have no preceding re. */ - if (!laststart) { - if (syntax & RE_CONTEXT_INVALID_OPS) - FREE_STACK_RETURN(REG_BADRPT); - else if (syntax & RE_CONTEXT_INDEP_OPS) - laststart = b; - else - goto unfetch_interval; - } +# ifdef _LIBC +weak_alias(__re_match, re_match) +# endif +#endif /* not emacs */ +static boolean group_match_null_string_p _RE_ARGS((unsigned char **p, + unsigned char *end, + register_info_type * - /* If the upper bound is zero, don't want to succeed at - all; jump from `laststart' to `b + 3', which will be - the end of the buffer after we insert the jump. */ - if (upper_bound == 0) { - GET_BUFFER_SPACE(3); - INSERT_JUMP(jump, laststart, b + 3); - b += 3; - } + reg_info)); +static boolean alt_match_null_string_p +_RE_ARGS( - /* Otherwise, we have a nontrivial interval. When - we're all done, the pattern will look like: - set_number_at - set_number_at - succeed_n - - jump_n - (The upper bound and `jump_n' are omitted if - `upper_bound' is 1, though.) */ - else { /* If the upper bound is > 1, we need to insert - more at the end of the loop. */ - unsigned nbytes = 10 + (upper_bound > 1) * 10; + (unsigned char *p, unsigned char *end, + register_info_type * reg_info)); +static boolean common_op_match_null_string_p +_RE_ARGS( - GET_BUFFER_SPACE(nbytes); + (unsigned char **p, unsigned char *end, + register_info_type * reg_info)); +static int bcmp_translate +_RE_ARGS((const char *s1, const char *s2, int len, char *translate)); - /* Initialize lower bound of the `succeed_n', even - though it will be set during matching by its - attendant `set_number_at' (inserted next), - because `re_compile_fastmap' needs to know. - Jump to the `jump_n' we might insert below. */ - INSERT_JUMP2(succeed_n, laststart, - b + 5 + (upper_bound > 1) * 5, - lower_bound); - b += 5; +/* re_match_2 matches the compiled pattern in BUFP against the + the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 + and SIZE2, respectively). We start matching at POS, and stop + matching at STOP. - /* Code to initialize the lower bound. Insert - before the `succeed_n'. The `5' is the last two - bytes of this `set_number_at', plus 3 bytes of - the following `succeed_n'. */ - insert_op2(set_number_at, laststart, 5, - lower_bound, b); - b += 5; + If REGS is non-null and the `no_sub' field of BUFP is nonzero, we + store offsets for the substring each group matched in REGS. See the + documentation for exactly how many groups we fill. - if (upper_bound > 1) { /* More than one repetition is allowed, so - append a backward jump to the `succeed_n' - that starts this interval. + We return -1 if no match, -2 if an internal error (such as the + failure stack overflowing). Otherwise, we return the length of the + matched substring. */ - When we've reached this during matching, - we'll have matched the interval once, so - jump back only `upper_bound - 1' times. */ - STORE_JUMP2(jump_n, b, laststart + 5, - upper_bound - 1); - b += 5; +int re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) +struct re_pattern_buffer *bufp; +const char *string1, *string2; +int size1, size2; +int pos; +struct re_registers *regs; +int stop; +{ + int result = re_match_2_internal(bufp, string1, size1, string2, size2, + pos, regs, stop); - /* The location we want to set is the second - parameter of the `jump_n'; that is `b-2' as - an absolute address. `laststart' will be - the `set_number_at' we're about to insert; - `laststart+3' the number to set, the source - for the relative address. But we are - inserting into the middle of the pattern -- - so everything is getting moved up by 5. - Conclusion: (b - 2) - (laststart + 3) + 5, - i.e., b - laststart. +#ifndef REGEX_MALLOC +# ifdef C_ALLOCA + alloca(0); +# endif +#endif + return result; +} - We insert this at the beginning of the loop - so that if we fail during matching, we'll - reinitialize the bounds. */ - insert_op2(set_number_at, laststart, - b - laststart, upper_bound - 1, b); - b += 5; - } - } - pending_exact = 0; - beg_interval = NULL; - } - break; +#ifdef _LIBC +weak_alias(__re_match_2, re_match_2) +#endif +/* This is a separate function so that we can force an alloca cleanup + afterwards. */ +static int +re_match_2_internal(bufp, string1, size1, string2, size2, pos, regs, stop) +struct re_pattern_buffer *bufp; +const char *string1, *string2; +int size1, size2; +int pos; +struct re_registers *regs; +int stop; +{ + /* General temporaries. */ + int mcnt; + unsigned char *p1; - unfetch_interval: - /* If an invalid interval, match the characters as literals. */ - assert(beg_interval); - p = beg_interval; - beg_interval = NULL; + /* Just past the end of the corresponding string. */ + const char *end1, *end2; - /* normal_char and normal_backslash need `c'. */ - PATFETCH(c); + /* Pointers into string1 and string2, just past the last characters in + each to consider matching. */ + const char *end_match_1, *end_match_2; - if (!(syntax & RE_NO_BK_BRACES)) { - if (p > pattern && p[-1] == '\\') - goto normal_backslash; - } - goto normal_char; + /* Where we are in the data, and the end of the current string. */ + const char *d, *dend; -#ifdef emacs - /* There is no way to specify the before_dot and after_dot - operators. rms says this is ok. --karl */ - case '=': - BUF_PUSH(at_dot); - break; + /* Where we are in the pattern, and the end of the pattern. */ + unsigned char *p = bufp->buffer; + register unsigned char *pend = p + bufp->used; - case 's': - laststart = b; - PATFETCH(c); - BUF_PUSH_2(syntaxspec, syntax_spec_code[c]); - break; + /* Mark the opcode just after a start_memory, so we can test for an + empty subpattern when we get to the stop_memory. */ + unsigned char *just_past_start_mem = 0; - case 'S': - laststart = b; - PATFETCH(c); - BUF_PUSH_2(notsyntaxspec, syntax_spec_code[c]); - break; -#endif /* emacs */ + /* We use this to map every character in the string. */ + RE_TRANSLATE_TYPE translate = bufp->translate; + /* Failure point stack. Each place that can handle a failure further + down the line pushes a failure point on this stack. It consists of + restart, regend, and reg_info for all registers corresponding to + the subexpressions we're currently inside, plus the number of such + registers, and, finally, two char *'s. The first char * is where + to resume scanning the pattern; the second one is where to resume + scanning the strings. If the latter is zero, the failure point is + a ``dummy''; if a failure happens and the failure point is a dummy, + it gets discarded and the next next one is tried. */ +#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ + fail_stack_type fail_stack; +#endif +#ifdef DEBUG + static unsigned failure_id; + unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; +#endif - case 'w': - if (syntax & RE_NO_GNU_OPS) - goto normal_char; - laststart = b; - BUF_PUSH(wordchar); - break; +#ifdef REL_ALLOC + /* This holds the pointer to the failure stack, when + it is allocated relocatably. */ + fail_stack_elt_t *failure_stack_ptr; +#endif + /* We fill all the registers internally, independent of what we + return, for use in backreferences. The number here includes + an element for register zero. */ + size_t num_regs = bufp->re_nsub + 1; - case 'W': - if (syntax & RE_NO_GNU_OPS) - goto normal_char; - laststart = b; - BUF_PUSH(notwordchar); - break; + /* The currently active registers. */ + active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG; + active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG; + /* Information on the contents of registers. These are pointers into + the input strings; they record just what was matched (on this + attempt) by a subexpression part of the pattern, that is, the + regnum-th regstart pointer points to where in the pattern we began + matching and the regnum-th regend points to right after where we + stopped matching the regnum-th subexpression. (The zeroth register + keeps track of what the whole pattern matches.) */ +#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ + const char **regstart, **regend; +#endif - case '<': - if (syntax & RE_NO_GNU_OPS) - goto normal_char; - BUF_PUSH(wordbeg); - break; + /* If a group that's operated upon by a repetition operator fails to + match anything, then the register for its start will need to be + restored because it will have been set to wherever in the string we + are when we last see its open-group operator. Similarly for a + register's end. */ +#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ + const char **old_regstart, **old_regend; +#endif - case '>': - if (syntax & RE_NO_GNU_OPS) - goto normal_char; - BUF_PUSH(wordend); - break; + /* The is_active field of reg_info helps us keep track of which (possibly + nested) subexpressions we are currently in. The matched_something + field of reg_info[reg_num] helps us tell whether or not we have + matched any of the pattern so far this time through the reg_num-th + subexpression. These two fields get reset each time through any + loop their register is in. */ +#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ + register_info_type *reg_info; +#endif - case 'b': - if (syntax & RE_NO_GNU_OPS) - goto normal_char; - BUF_PUSH(wordbound); - break; + /* The following record the register info as found in the above + variables when we find a match better than any we've seen before. + This happens as we backtrack through the failure points, which in + turn happens only if we have not yet matched the entire string. */ + unsigned best_regs_set = false; - case 'B': - if (syntax & RE_NO_GNU_OPS) - goto normal_char; - BUF_PUSH(notwordbound); - break; +#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ + const char **best_regstart, **best_regend; +#endif - case '`': - if (syntax & RE_NO_GNU_OPS) - goto normal_char; - BUF_PUSH(begbuf); - break; + /* Logically, this is `best_regend[0]'. But we don't want to have to + allocate space for that if we're not allocating space for anything + else (see below). Also, we never need info about register 0 for + any of the other register vectors, and it seems rather a kludge to + treat `best_regend' differently than the rest. So we keep track of + the end of the best match so far in a separate variable. We + initialize this to NULL so that when we backtrack the first time + and need to test it, it's not garbage. */ + const char *match_end = NULL; - case '\'': - if (syntax & RE_NO_GNU_OPS) - goto normal_char; - BUF_PUSH(endbuf); - break; + /* This helps SET_REGS_MATCHED avoid doing redundant work. */ + int set_regs_matched_done = 0; - case '1': - case '2': - case '3': - case '4': - case '5': - case '6': - case '7': - case '8': - case '9': - if (syntax & RE_NO_BK_REFS) - goto normal_char; + /* Used when we pop values we don't care about. */ +#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ + const char **reg_dummy; + register_info_type *reg_info_dummy; +#endif - c1 = c - '0'; +#ifdef DEBUG + /* Counts the total number of registers pushed. */ + unsigned num_regs_pushed = 0; +#endif - if (c1 > regnum) - FREE_STACK_RETURN(REG_ESUBREG); + DEBUG_PRINT1("\n\nEntering re_match_2.\n"); - /* Can't back reference to a subexpression if inside of it. */ - if (group_in_compile_stack(compile_stack, (regnum_t) c1)) - goto normal_char; + INIT_FAIL_STACK(); - laststart = b; - BUF_PUSH_2(duplicate, c1); - break; +#ifdef MATCH_MAY_ALLOCATE + /* Do not bother to initialize all the register variables if there are + no groups in the pattern, as it takes a fair amount of time. If + there are groups, we include space for register 0 (the whole + pattern), even though we never use it, since it simplifies the + array indexing. We should fix this. */ + if (bufp->re_nsub) { + regstart = REGEX_TALLOC(num_regs, const char *); + regend = REGEX_TALLOC(num_regs, const char *); + old_regstart = REGEX_TALLOC(num_regs, const char *); + old_regend = REGEX_TALLOC(num_regs, const char *); + best_regstart = REGEX_TALLOC(num_regs, const char *); + best_regend = REGEX_TALLOC(num_regs, const char *); + reg_info = REGEX_TALLOC(num_regs, register_info_type); + reg_dummy = REGEX_TALLOC(num_regs, const char *); - case '+': - case '?': - if (syntax & RE_BK_PLUS_QM) - goto handle_plus; - else - goto normal_backslash; + reg_info_dummy = REGEX_TALLOC(num_regs, register_info_type); - default: - normal_backslash: - /* You might think it would be useful for \ to mean - not to translate; but if we don't translate it - it will never match anything. */ - c = TRANSLATE(c); - goto normal_char; - } - break; + if (!(regstart && regend && old_regstart && old_regend && reg_info + && best_regstart && best_regend && reg_dummy + && reg_info_dummy)) { + FREE_VARIABLES(); + return -2; + } + } else { + /* We must initialize all our variables to NULL, so that + `FREE_VARIABLES' doesn't try to free them. */ + regstart = regend = old_regstart = old_regend = best_regstart + = best_regend = reg_dummy = NULL; + reg_info = reg_info_dummy = (register_info_type *) NULL; + } +#endif /* MATCH_MAY_ALLOCATE */ + /* The starting position is bogus. */ + if (pos < 0 || pos > size1 + size2) { + FREE_VARIABLES(); + return -1; + } - default: - /* Expects the character in `c'. */ - normal_char: - /* If no exactn currently being built. */ - if (!pending_exact - /* If last exactn not at current position. */ - || pending_exact + *pending_exact + 1 != b - /* We have only one byte following the exactn for the count. */ - || *pending_exact == (1 << BYTEWIDTH) - 1 - /* If followed by a repetition operator. */ - || *p == '*' || *p == '^' || ((syntax & RE_BK_PLUS_QM) - ? *p == '\\' && (p[1] == '+' - || p[1] == - '?') : (*p - == - '+' - || - *p - == - '?')) - || ((syntax & RE_INTERVALS) - && ((syntax & RE_NO_BK_BRACES) - ? *p == '{' : (p[0] == '\\' && p[1] == '{')))) { - /* Start building a new exactn. */ + /* Initialize subexpression text positions to -1 to mark ones that no + start_memory/stop_memory has been seen for. Also initialize the + register information struct. */ + for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) { + regstart[mcnt] = regend[mcnt] + = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; - laststart = b; + REG_MATCH_NULL_STRING_P(reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; + IS_ACTIVE(reg_info[mcnt]) = 0; + MATCHED_SOMETHING(reg_info[mcnt]) = 0; + EVER_MATCHED_SOMETHING(reg_info[mcnt]) = 0; + } - BUF_PUSH_2(exactn, 0); - pending_exact = b - 1; - } + /* We move `string1' into `string2' if the latter's empty -- but not if + `string1' is null. */ + if (size2 == 0 && string1 != NULL) { + string2 = string1; + size2 = size1; + string1 = 0; + size1 = 0; + } + end1 = string1 + size1; + end2 = string2 + size2; - BUF_PUSH(c); - (*pending_exact)++; - break; - } /* switch (c) */ - } /* while p != pend */ + /* Compute where to stop matching, within the two strings. */ + if (stop <= size1) { + end_match_1 = string1 + stop; + end_match_2 = string2; + } else { + end_match_1 = end1; + end_match_2 = string2 + stop - size1; + } + /* `p' scans through the pattern as `d' scans through the data. + `dend' is the end of the input string that `d' points within. `d' + is advanced into the following input string whenever necessary, but + this happens before fetching; therefore, at the beginning of the + loop, `d' can be pointing at the end of a string, but it cannot + equal `string2'. */ + if (size1 > 0 && pos <= size1) { + d = string1 + pos; + dend = end_match_1; + } else { + d = string2 + pos - size1; + dend = end_match_2; + } - /* Through the pattern now. */ + DEBUG_PRINT1("The compiled pattern is:\n"); + DEBUG_PRINT_COMPILED_PATTERN(bufp, p, pend); + DEBUG_PRINT1("The string to match is: `"); + DEBUG_PRINT_DOUBLE_STRING(d, string1, size1, string2, size2); + DEBUG_PRINT1("'\n"); - if (fixup_alt_jump) - STORE_JUMP(jump_past_alt, fixup_alt_jump, b); + /* This loops over pattern commands. It exits by returning from the + function if the match is complete, or it drops through if the match + fails at this starting point in the input data. */ + for (;;) { +#ifdef _LIBC + DEBUG_PRINT2("\n%p: ", p); +#else + DEBUG_PRINT2("\n0x%x: ", p); +#endif - if (!COMPILE_STACK_EMPTY) - FREE_STACK_RETURN(REG_EPAREN); + if (p == pend) { /* End of pattern means we might have succeeded. */ + DEBUG_PRINT1("end of pattern ... "); - /* If we don't want backtracking, force success - the first time we reach the end of the compiled pattern. */ - if (syntax & RE_NO_POSIX_BACKTRACKING) - BUF_PUSH(succeed); + /* If we haven't matched the entire string, and we want the + longest match, try backtracking. */ + if (d != end_match_2) { + /* 1 if this match ends in the same string (string1 or string2) + as the best previous match. */ + boolean same_str_p = (FIRST_STRING_P(match_end) + == MATCHING_IN_FIRST_STRING); - free(compile_stack.stack); + /* 1 if this match is the best seen so far. */ + boolean best_match_p; - /* We have succeeded; set the length of the buffer. */ - bufp->used = b - bufp->buffer; + /* AIX compiler got confused when this was combined + with the previous declaration. */ + if (same_str_p) + best_match_p = d > match_end; + else + best_match_p = !MATCHING_IN_FIRST_STRING; -#ifdef DEBUG - if (debug) { - DEBUG_PRINT1("\nCompiled pattern: \n"); - print_compiled_pattern(bufp); - } -#endif /* DEBUG */ + DEBUG_PRINT1("backtracking.\n"); -#ifndef MATCH_MAY_ALLOCATE - /* Initialize the failure stack to the largest possible stack. This - isn't necessary unless we're trying to avoid calling alloca in - the search and match routines. */ - { - int num_regs = bufp->re_nsub + 1; + if (!FAIL_STACK_EMPTY()) { /* More failure points to try. */ - /* Since DOUBLE_FAIL_STACK refuses to double only if the current size - is strictly greater than re_max_failures, the largest possible stack - is 2 * re_max_failures failure points. */ - if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) { - fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); + /* If exceeds best match so far, save it. */ + if (!best_regs_set || best_match_p) { + best_regs_set = true; + match_end = d; -# ifdef emacs - if (!fail_stack.stack) - fail_stack.stack - = (fail_stack_elt_t *) xmalloc(fail_stack.size - * - sizeof - (fail_stack_elt_t)); - else - fail_stack.stack = - (fail_stack_elt_t *) xrealloc(fail_stack.stack, - (fail_stack.size * - sizeof - (fail_stack_elt_t))); -# else /* not emacs */ - if (!fail_stack.stack) - fail_stack.stack - = (fail_stack_elt_t *) malloc(fail_stack.size - * - sizeof - (fail_stack_elt_t)); - else - fail_stack.stack = - (fail_stack_elt_t *) realloc(fail_stack.stack, - (fail_stack.size * - sizeof - (fail_stack_elt_t))); -# endif /* not emacs */ - } + DEBUG_PRINT1("\nSAVING match as best so far.\n"); - regex_grow_registers(num_regs); - } -#endif /* not MATCH_MAY_ALLOCATE */ + for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) { + best_regstart[mcnt] = regstart[mcnt]; + best_regend[mcnt] = regend[mcnt]; + } + } + goto fail; + } - return REG_NOERROR; -} /* regex_compile */ - -/* Subroutines for `regex_compile'. */ + /* If no failure points, don't restore garbage. And if + last match is real best match, don't restore second + best one. */ + else if (best_regs_set && !best_match_p) { + restore_best_regs: + /* Restore best match. It may happen that `dend == + end_match_1' while the restored d is in string2. + For example, the pattern `x.*y.*z' against the + strings `x-' and `y-z-', if the two strings are + not consecutive in memory. */ + DEBUG_PRINT1("Restoring best registers.\n"); -/* Store OP at LOC followed by two-byte integer parameter ARG. */ + d = match_end; + dend = ((d >= string1 && d <= end1) + ? end_match_1 : end_match_2); -static void store_op1(op, loc, arg) -re_opcode_t op; -unsigned char *loc; -int arg; -{ - *loc = (unsigned char) op; - STORE_NUMBER(loc + 1, arg); -} + for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) { + regstart[mcnt] = best_regstart[mcnt]; + regend[mcnt] = best_regend[mcnt]; + } + } + } + /* d != end_match_2 */ + succeed_label: + DEBUG_PRINT1("Accepting match.\n"); + /* If caller wants register contents data back, do it. */ + if (regs && !bufp->no_sub) { + /* Have the register data arrays been allocated? */ + if (bufp->regs_allocated == REGS_UNALLOCATED) { /* No. So allocate them with malloc. We need one + extra element beyond `num_regs' for the `-1' marker + GNU code uses. */ + regs->num_regs = MAX(RE_NREGS, num_regs + 1); + regs->start = TALLOC(regs->num_regs, regoff_t); + regs->end = TALLOC(regs->num_regs, regoff_t); + if (regs->start == NULL || regs->end == NULL) { + FREE_VARIABLES(); + return -2; + } + bufp->regs_allocated = REGS_REALLOCATE; + } else if (bufp->regs_allocated == REGS_REALLOCATE) { /* Yes. If we need more elements than were already + allocated, reallocate them. If we need fewer, just + leave it alone. */ + if (regs->num_regs < num_regs + 1) { + regs->num_regs = num_regs + 1; + RETALLOC(regs->start, regs->num_regs, regoff_t); + RETALLOC(regs->end, regs->num_regs, regoff_t); + if (regs->start == NULL || regs->end == NULL) { + FREE_VARIABLES(); + return -2; + } + } + } else { + /* These braces fend off a "empty body in an else-statement" + warning under GCC when assert expands to nothing. */ + assert(bufp->regs_allocated == REGS_FIXED); + } -/* Like `store_op1', but for two two-byte parameters ARG1 and ARG2. */ + /* Convert the pointer data in `regstart' and `regend' to + indices. Register zero has to be set differently, + since we haven't kept track of any info for it. */ + if (regs->num_regs > 0) { + regs->start[0] = pos; + regs->end[0] = (MATCHING_IN_FIRST_STRING + ? ((regoff_t) (d - string1)) + : ((regoff_t) (d - string2 + size1))); + } -static void store_op2(op, loc, arg1, arg2) -re_opcode_t op; -unsigned char *loc; -int arg1, arg2; -{ - *loc = (unsigned char) op; - STORE_NUMBER(loc + 1, arg1); - STORE_NUMBER(loc + 3, arg2); -} + /* Go through the first `min (num_regs, regs->num_regs)' + registers, since that is all we initialized. */ + for (mcnt = 1; + (unsigned) mcnt < MIN(num_regs, regs->num_regs); + mcnt++) { + if (REG_UNSET(regstart[mcnt]) + || REG_UNSET(regend[mcnt])) regs->start[mcnt] = + regs->end[mcnt] = -1; + else { + regs->start[mcnt] + = (regoff_t) POINTER_TO_OFFSET(regstart[mcnt]); + regs->end[mcnt] + = (regoff_t) POINTER_TO_OFFSET(regend[mcnt]); + } + } + /* If the regs structure we return has more elements than + were in the pattern, set the extra elements to -1. If + we (re)allocated the registers, this is the case, + because we always allocate enough to have at least one + -1 at the end. */ + for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; + mcnt++) + regs->start[mcnt] = regs->end[mcnt] = -1; + } + /* regs && !bufp->no_sub */ + DEBUG_PRINT4 + ("%u failure points pushed, %u popped (%u remain).\n", + nfailure_points_pushed, nfailure_points_popped, + nfailure_points_pushed - nfailure_points_popped); + DEBUG_PRINT2("%u registers pushed.\n", num_regs_pushed); -/* Copy the bytes from LOC to END to open up three bytes of space at LOC - for OP followed by two-byte integer parameter ARG. */ + mcnt = d - pos - (MATCHING_IN_FIRST_STRING + ? string1 : string2 - size1); -static void insert_op1(op, loc, arg, end) -re_opcode_t op; -unsigned char *loc; -int arg; -unsigned char *end; -{ - register unsigned char *pfrom = end; - register unsigned char *pto = end + 3; + DEBUG_PRINT2("Returning %d from re_match_2.\n", mcnt); - while (pfrom != loc) - *--pto = *--pfrom; + FREE_VARIABLES(); + return mcnt; + } - store_op1(op, loc, arg); -} + /* Otherwise match next pattern command. */ + switch (SWITCH_ENUM_CAST((re_opcode_t) * p++)) { + /* Ignore these. Used to ignore the n of succeed_n's which + currently have n == 0. */ + case no_op: + DEBUG_PRINT1("EXECUTING no_op.\n"); + break; + case succeed: + DEBUG_PRINT1("EXECUTING succeed.\n"); + goto succeed_label; -/* Like `insert_op1', but for two two-byte parameters ARG1 and ARG2. */ + /* Match the next n pattern characters exactly. The following + byte in the pattern defines n, and the n bytes after that + are the characters to match. */ + case exactn: + mcnt = *p++; + DEBUG_PRINT2("EXECUTING exactn %d.\n", mcnt); -static void insert_op2(op, loc, arg1, arg2, end) -re_opcode_t op; -unsigned char *loc; -int arg1, arg2; -unsigned char *end; -{ - register unsigned char *pfrom = end; - register unsigned char *pto = end + 5; + /* This is written out as an if-else so we don't waste time + testing `translate' inside the loop. */ + if (translate) { + do { + PREFETCH(); + if ((unsigned char) translate[(unsigned char) *d++] + != (unsigned char) *p++) + goto fail; + } + while (--mcnt); + } else { + do { + PREFETCH(); + if (*d++ != (char) *p++) + goto fail; + } + while (--mcnt); + } + SET_REGS_MATCHED(); + break; - while (pfrom != loc) - *--pto = *--pfrom; - store_op2(op, loc, arg1, arg2); -} + /* Match any character except possibly a newline or a null. */ + case anychar: + DEBUG_PRINT1("EXECUTING anychar.\n"); + PREFETCH(); -/* P points to just after a ^ in PATTERN. Return true if that ^ comes - after an alternative or a begin-subexpression. We assume there is at - least one character before the ^. */ + if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE(*d) == '\n') + || (bufp->syntax & RE_DOT_NOT_NULL + && TRANSLATE(*d) == '\000')) goto fail; -static boolean at_begline_loc_p(pattern, p, syntax) -const char *pattern, *p; -reg_syntax_t syntax; -{ - const char *prev = p - 2; - boolean prev_prev_backslash = prev > pattern && prev[-1] == '\\'; + SET_REGS_MATCHED(); + DEBUG_PRINT2(" Matched `%d'.\n", *d); + d++; + break; - return - /* After a subexpression? */ - (*prev == '(' && (syntax & RE_NO_BK_PARENS || prev_prev_backslash)) - /* After an alternative? */ - || (*prev == '|' - && (syntax & RE_NO_BK_VBAR || prev_prev_backslash)); -} + case charset: + case charset_not: + { + register unsigned char c; + boolean not = (re_opcode_t) * (p - 1) == charset_not; + + DEBUG_PRINT2("EXECUTING charset%s.\n", not ? "_not" : ""); -/* The dual of at_begline_loc_p. This one is for $. We assume there is - at least one character after the $, i.e., `P < PEND'. */ + PREFETCH(); + c = TRANSLATE(*d); /* The character to match. */ -static boolean at_endline_loc_p(p, pend, syntax) -const char *p, *pend; -reg_syntax_t syntax; -{ - const char *next = p; - boolean next_backslash = *next == '\\'; - const char *next_next = p + 1 < pend ? p + 1 : 0; + /* Cast to `unsigned' instead of `unsigned char' in case the + bit list is a full 32 bytes long. */ + if (c < (unsigned) (*p * BYTEWIDTH) + && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) + not = !not; - return - /* Before a subexpression? */ - (syntax & RE_NO_BK_PARENS ? *next == ')' - : next_backslash && next_next && *next_next == ')') - /* Before an alternative? */ - || (syntax & RE_NO_BK_VBAR ? *next == '|' - : next_backslash && next_next && *next_next == '|'); -} + p += 1 + *p; + if (!not) + goto fail; -/* Returns true if REGNUM is in one of COMPILE_STACK's elements and - false if it's not. */ + SET_REGS_MATCHED(); + d++; + break; + } -static boolean group_in_compile_stack(compile_stack, regnum) -compile_stack_type compile_stack; -regnum_t regnum; -{ - int this_element; - for (this_element = compile_stack.avail - 1; - this_element >= 0; this_element--) - if (compile_stack.stack[this_element].regnum == regnum) - return true; + /* The beginning of a group is represented by start_memory. + The arguments are the register number in the next byte, and the + number of groups inner to this one in the next. The text + matched within the group is recorded (in the internal + registers data structure) under the register number. */ + case start_memory: + DEBUG_PRINT3("EXECUTING start_memory %d (%d):\n", *p, p[1]); - return false; -} + /* Find out if this group can match the empty string. */ + p1 = p; /* To send to group_match_null_string_p. */ + if (REG_MATCH_NULL_STRING_P(reg_info[*p]) == + MATCH_NULL_UNSET_VALUE) + REG_MATCH_NULL_STRING_P(reg_info[*p]) = + group_match_null_string_p(&p1, pend, reg_info); -/* Read the ending character of a range (in a bracket expression) from the - uncompiled pattern *P_PTR (which ends at PEND). We assume the - starting character is in `P[-2]'. (`P[-1]' is the character `-'.) - Then we set the translation of all bits between the starting and - ending characters (inclusive) in the compiled pattern B. + /* Save the position in the string where we were the last time + we were at this open-group operator in case the group is + operated upon by a repetition operator, e.g., with `(a*)*b' + against `ab'; then we want to ignore where we are now in + the string in case this attempt to match fails. */ + old_regstart[*p] = REG_MATCH_NULL_STRING_P(reg_info[*p]) + ? REG_UNSET(regstart[*p]) ? d : regstart[*p] + : regstart[*p]; + DEBUG_PRINT2(" old_regstart: %d\n", + POINTER_TO_OFFSET(old_regstart[*p])); - Return an error code. + regstart[*p] = d; + DEBUG_PRINT2(" regstart: %d\n", + POINTER_TO_OFFSET(regstart[*p])); - We use these short variable names so we can use the same macros as - `regex_compile' itself. */ + IS_ACTIVE(reg_info[*p]) = 1; + MATCHED_SOMETHING(reg_info[*p]) = 0; -static reg_errcode_t compile_range(p_ptr, pend, translate, syntax, b) -const char **p_ptr, *pend; -RE_TRANSLATE_TYPE translate; -reg_syntax_t syntax; -unsigned char *b; -{ - unsigned this_char; + /* Clear this whenever we change the register activity status. */ + set_regs_matched_done = 0; - const char *p = *p_ptr; - reg_errcode_t ret; - char range_start[2]; - char range_end[2]; - char ch[2]; + /* This is the new highest active register. */ + highest_active_reg = *p; - if (p == pend) - return REG_ERANGE; + /* If nothing was active before, this is the new lowest active + register. */ + if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) + lowest_active_reg = *p; - /* Fetch the endpoints without translating them; the - appropriate translation is done in the bit-setting loop below. */ - range_start[0] = p[-2]; - range_start[1] = '\0'; - range_end[0] = p[0]; - range_end[1] = '\0'; + /* Move past the register number and inner group count. */ + p += 2; + just_past_start_mem = p; - /* Have to increment the pointer into the pattern string, so the - caller isn't still at the ending character. */ - (*p_ptr)++; + break; - /* Report an error if the range is empty and the syntax prohibits this. */ - ret = syntax & RE_NO_EMPTY_RANGES ? REG_ERANGE : REG_NOERROR; - /* Here we see why `this_char' has to be larger than an `unsigned - char' -- we would otherwise go into an infinite loop, since all - characters <= 0xff. */ - ch[1] = '\0'; - for (this_char = 0; this_char <= (unsigned char) -1; ++this_char) { - ch[0] = this_char; - if (strcoll(range_start, ch) <= 0 && strcoll(ch, range_end) <= 0) { - SET_LIST_BIT(TRANSLATE(this_char)); - ret = REG_NOERROR; - } - } + /* The stop_memory opcode represents the end of a group. Its + arguments are the same as start_memory's: the register + number, and the number of inner groups. */ + case stop_memory: + DEBUG_PRINT3("EXECUTING stop_memory %d (%d):\n", *p, p[1]); - return ret; -} - -/* re_compile_fastmap computes a ``fastmap'' for the compiled pattern in - BUFP. A fastmap records which of the (1 << BYTEWIDTH) possible - characters can start a string that matches the pattern. This fastmap - is used by re_search to skip quickly over impossible starting points. + /* We need to save the string position the last time we were at + this close-group operator in case the group is operated + upon by a repetition operator, e.g., with `((a*)*(b*)*)*' + against `aba'; then we want to ignore where we are now in + the string in case this attempt to match fails. */ + old_regend[*p] = REG_MATCH_NULL_STRING_P(reg_info[*p]) + ? REG_UNSET(regend[*p]) ? d : regend[*p] + : regend[*p]; + DEBUG_PRINT2(" old_regend: %d\n", + POINTER_TO_OFFSET(old_regend[*p])); - The caller must supply the address of a (1 << BYTEWIDTH)-byte data - area as BUFP->fastmap. + regend[*p] = d; + DEBUG_PRINT2(" regend: %d\n", + POINTER_TO_OFFSET(regend[*p])); - We set the `fastmap', `fastmap_accurate', and `can_be_null' fields in - the pattern buffer. + /* This register isn't active anymore. */ + IS_ACTIVE(reg_info[*p]) = 0; - Returns 0 if we succeed, -2 if an internal error. */ + /* Clear this whenever we change the register activity status. */ + set_regs_matched_done = 0; -int re_compile_fastmap(bufp) -struct re_pattern_buffer *bufp; -{ - int j, k; + /* If this was the only register active, nothing is active + anymore. */ + if (lowest_active_reg == highest_active_reg) { + lowest_active_reg = NO_LOWEST_ACTIVE_REG; + highest_active_reg = NO_HIGHEST_ACTIVE_REG; + } else { /* We must scan for the new highest active register, since + it isn't necessarily one less than now: consider + (a(b)c(d(e)f)g). When group 3 ends, after the f), the + new highest active register is 1. */ + unsigned char r = *p - 1; -#ifdef MATCH_MAY_ALLOCATE - fail_stack_type fail_stack; -#endif -#ifndef REGEX_MALLOC - char *destination; -#endif + while (r > 0 && !IS_ACTIVE(reg_info[r])) + r--; - register char *fastmap = bufp->fastmap; - unsigned char *pattern = bufp->buffer; - unsigned char *p = pattern; - register unsigned char *pend = pattern + bufp->used; + /* If we end up at register zero, that means that we saved + the registers as the result of an `on_failure_jump', not + a `start_memory', and we jumped to past the innermost + `stop_memory'. For example, in ((.)*) we save + registers 1 and 2 as a result of the *, but when we pop + back to the second ), we are at the stop_memory 1. + Thus, nothing is active. */ + if (r == 0) { + lowest_active_reg = NO_LOWEST_ACTIVE_REG; + highest_active_reg = NO_HIGHEST_ACTIVE_REG; + } else + highest_active_reg = r; + } -#ifdef REL_ALLOC - /* This holds the pointer to the failure stack, when - it is allocated relocatably. */ - fail_stack_elt_t *failure_stack_ptr; -#endif + /* If just failed to match something this time around with a + group that's operated on by a repetition operator, try to + force exit from the ``loop'', and restore the register + information for this group that we had before trying this + last match. */ + if ((!MATCHED_SOMETHING(reg_info[*p]) + || just_past_start_mem == p - 1) + && (p + 2) < pend) { + boolean is_a_jump_n = false; - /* Assume that each path through the pattern can be null until - proven otherwise. We set this false at the bottom of switch - statement, to which we get only if a particular path doesn't - match the empty string. */ - boolean path_can_be_null = true; + p1 = p + 2; + mcnt = 0; + switch ((re_opcode_t) * p1++) { + case jump_n: + is_a_jump_n = true; + case pop_failure_jump: + case maybe_pop_jump: + case jump: + case dummy_failure_jump: + EXTRACT_NUMBER_AND_INCR(mcnt, p1); + if (is_a_jump_n) + p1 += 2; + break; + + default: + /* do nothing */ ; + } + p1 += mcnt; + + /* If the next operation is a jump backwards in the pattern + to an on_failure_jump right before the start_memory + corresponding to this stop_memory, exit from the loop + by forcing a failure after pushing on the stack the + on_failure_jump's jump in the pattern, and d. */ + if (mcnt < 0 && (re_opcode_t) * p1 == on_failure_jump + && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) { + /* If this group ever matched anything, then restore + what its registers were before trying this last + failed match, e.g., with `(a*)*b' against `ab' for + regstart[1], and, e.g., with `((a*)*(b*)*)*' + against `aba' for regend[3]. + + Also restore the registers for inner groups for, + e.g., `((a*)(b*))*' against `aba' (register 3 would + otherwise get trashed). */ - /* We aren't doing a `succeed_n' to begin with. */ - boolean succeed_n_p = false; + if (EVER_MATCHED_SOMETHING(reg_info[*p])) { + unsigned r; - assert(fastmap != NULL && p != NULL); + EVER_MATCHED_SOMETHING(reg_info[*p]) = 0; - INIT_FAIL_STACK(); - bzero(fastmap, 1 << BYTEWIDTH); /* Assume nothing's valid. */ - bufp->fastmap_accurate = 1; /* It will be when we're done. */ - bufp->can_be_null = 0; + /* Restore this and inner groups' (if any) registers. */ + for (r = *p; + r < (unsigned) *p + (unsigned) *(p + 1); r++) { + regstart[r] = old_regstart[r]; - while (1) { - if (p == pend || *p == succeed) { - /* We have reached the (effective) end of pattern. */ - if (!FAIL_STACK_EMPTY()) { - bufp->can_be_null |= path_can_be_null; + /* xx why this test? */ + if (old_regend[r] >= regstart[r]) + regend[r] = old_regend[r]; + } + } + p1++; + EXTRACT_NUMBER_AND_INCR(mcnt, p1); + PUSH_FAILURE_POINT(p1 + mcnt, d, -2); - /* Reset for next path. */ - path_can_be_null = true; + goto fail; + } + } - p = fail_stack.stack[--fail_stack.avail].pointer; + /* Move past the register number and the inner group count. */ + p += 2; + break; - continue; - } else - break; - } - /* We should never be about to go beyond the end of the pattern. */ - assert(p < pend); + /* \ has been turned into a `duplicate' command which is + followed by the numeric value of as the register number. */ + case duplicate: + { + register const char *d2, *dend2; + int regno = *p++; /* Get which register to match against. */ - switch (SWITCH_ENUM_CAST((re_opcode_t) * p++)) { + DEBUG_PRINT2("EXECUTING duplicate %d.\n", regno); - /* I guess the idea here is to simply not bother with a fastmap - if a backreference is used, since it's too hard to figure out - the fastmap for the corresponding group. Setting - `can_be_null' stops `re_search_2' from using the fastmap, so - that is all we do. */ - case duplicate: - bufp->can_be_null = 1; - goto done; + /* Can't back reference a group which we've never matched. */ + if (REG_UNSET(regstart[regno]) || REG_UNSET(regend[regno])) + goto fail; + /* Where in input to try to start matching. */ + d2 = regstart[regno]; - /* Following are the cases which match a character. These end - with `break'. */ + /* Where to stop matching; if both the place to start and + the place to stop matching are in the same string, then + set to the place to stop, otherwise, for now have to use + the end of the first string. */ - case exactn: - fastmap[p[1]] = 1; - break; + dend2 = ((FIRST_STRING_P(regstart[regno]) + == FIRST_STRING_P(regend[regno])) + ? regend[regno] : end_match_1); + for (;;) { + /* If necessary, advance to next segment in register + contents. */ + while (d2 == dend2) { + if (dend2 == end_match_2) + break; + if (dend2 == regend[regno]) + break; + /* End of string1 => advance to string2. */ + d2 = string2; + dend2 = regend[regno]; + } + /* At end of register contents => success */ + if (d2 == dend2) + break; - case charset: - for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) - if (p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH))) - fastmap[j] = 1; - break; + /* If necessary, advance to next segment in data. */ + PREFETCH(); + /* How many characters left in this segment to match. */ + mcnt = dend - d; - case charset_not: - /* Chars beyond end of map must be allowed. */ - for (j = *p * BYTEWIDTH; j < (1 << BYTEWIDTH); j++) - fastmap[j] = 1; + /* Want how many consecutive characters we can match in + one shot, so, if necessary, adjust the count. */ + if (mcnt > dend2 - d2) + mcnt = dend2 - d2; - for (j = *p++ * BYTEWIDTH - 1; j >= 0; j--) - if (!(p[j / BYTEWIDTH] & (1 << (j % BYTEWIDTH)))) - fastmap[j] = 1; + /* Compare that many; failure if mismatch, else move + past them. */ + if (translate ? bcmp_translate(d, d2, mcnt, translate) + : memcmp(d, d2, mcnt)) + goto fail; + d += mcnt, d2 += mcnt; + + /* Do this because we've match some characters. */ + SET_REGS_MATCHED(); + } + } break; - case wordchar: - for (j = 0; j < (1 << BYTEWIDTH); j++) - if (SYNTAX(j) == Sword) - fastmap[j] = 1; - break; + /* begline matches the empty string at the beginning of the string + (unless `not_bol' is set in `bufp'), and, if + `newline_anchor' is set, after newlines. */ + case begline: + DEBUG_PRINT1("EXECUTING begline.\n"); + if (AT_STRINGS_BEG(d)) { + if (!bufp->not_bol) + break; + } else if (d[-1] == '\n' && bufp->newline_anchor) { + break; + } + /* In all other cases, we fail. */ + goto fail; - case notwordchar: - for (j = 0; j < (1 << BYTEWIDTH); j++) - if (SYNTAX(j) != Sword) - fastmap[j] = 1; - break; + /* endline is the dual of begline. */ + case endline: + DEBUG_PRINT1("EXECUTING endline.\n"); - case anychar: - { - int fastmap_newline = fastmap['\n']; + if (AT_STRINGS_END(d)) { + if (!bufp->not_eol) + break; + } - /* `.' matches anything ... */ - for (j = 0; j < (1 << BYTEWIDTH); j++) - fastmap[j] = 1; + /* We have to ``prefetch'' the next character. */ + else if ((d == end1 ? *string2 : *d) == '\n' + && bufp->newline_anchor) { + break; + } + goto fail; - /* ... except perhaps newline. */ - if (!(bufp->syntax & RE_DOT_NEWLINE)) - fastmap['\n'] = fastmap_newline; - /* Return if we have already set `can_be_null'; if we have, - then the fastmap is irrelevant. Something's wrong here. */ - else if (bufp->can_be_null) - goto done; + /* Match at the very beginning of the data. */ + case begbuf: + DEBUG_PRINT1("EXECUTING begbuf.\n"); + if (AT_STRINGS_BEG(d)) + break; + goto fail; - /* Otherwise, have to check alternative paths. */ - break; - } -#ifdef emacs - case syntaxspec: - k = *p++; - for (j = 0; j < (1 << BYTEWIDTH); j++) - if (SYNTAX(j) == (enum syntaxcode) k) - fastmap[j] = 1; - break; + /* Match at the very end of the data. */ + case endbuf: + DEBUG_PRINT1("EXECUTING endbuf.\n"); + if (AT_STRINGS_END(d)) + break; + goto fail; - case notsyntaxspec: - k = *p++; - for (j = 0; j < (1 << BYTEWIDTH); j++) - if (SYNTAX(j) != (enum syntaxcode) k) - fastmap[j] = 1; - break; + /* on_failure_keep_string_jump is used to optimize `.*\n'. It + pushes NULL as the value for the string on the stack. Then + `pop_failure_point' will keep the current value for the + string, instead of restoring it. To see why, consider + matching `foo\nbar' against `.*\n'. The .* matches the foo; + then the . fails against the \n. But the next thing we want + to do is match the \n against the \n; if we restored the + string value, we would be back at the foo. + + Because this is used only in specific cases, we don't need to + check all the things that `on_failure_jump' does, to make + sure the right things get saved on the stack. Hence we don't + share its code. The only reason to push anything on the + stack at all is that otherwise we would have to change + `anychar's code to do something besides goto fail in this + case; that seems worse than this. */ + case on_failure_keep_string_jump: + DEBUG_PRINT1("EXECUTING on_failure_keep_string_jump"); + EXTRACT_NUMBER_AND_INCR(mcnt, p); +#ifdef _LIBC + DEBUG_PRINT3(" %d (to %p):\n", mcnt, p + mcnt); +#else + DEBUG_PRINT3(" %d (to 0x%x):\n", mcnt, p + mcnt); +#endif - /* All cases after this match the empty string. These end with - `continue'. */ + PUSH_FAILURE_POINT(p + mcnt, NULL, -2); + break; - case before_dot: - case at_dot: - case after_dot: - continue; -#endif /* emacs */ + /* Uses of on_failure_jump: + Each alternative starts with an on_failure_jump that points + to the beginning of the next alternative. Each alternative + except the last ends with a jump that in effect jumps past + the rest of the alternatives. (They really jump to the + ending jump of the following alternative, because tensioning + these jumps is a hassle.) - case no_op: - case begline: - case endline: - case begbuf: - case endbuf: - case wordbound: - case notwordbound: - case wordbeg: - case wordend: - case push_dummy_failure: - continue; + Repeats start with an on_failure_jump that points past both + the repetition text and either the following jump or + pop_failure_jump back to this on_failure_jump. */ + case on_failure_jump: + on_failure: + DEBUG_PRINT1("EXECUTING on_failure_jump"); + EXTRACT_NUMBER_AND_INCR(mcnt, p); +#ifdef _LIBC + DEBUG_PRINT3(" %d (to %p)", mcnt, p + mcnt); +#else + DEBUG_PRINT3(" %d (to 0x%x)", mcnt, p + mcnt); +#endif - case jump_n: - case pop_failure_jump: - case maybe_pop_jump: - case jump: - case jump_past_alt: - case dummy_failure_jump: - EXTRACT_NUMBER_AND_INCR(j, p); - p += j; - if (j > 0) - continue; + /* If this on_failure_jump comes right before a group (i.e., + the original * applied to a group), save the information + for that group and all inner ones, so that if we fail back + to this point, the group's information will be correct. + For example, in \(a*\)*\1, we need the preceding group, + and in \(zz\(a*\)b*\)\2, we need the inner group. */ - /* Jump backward implies we just went through the body of a - loop and matched nothing. Opcode jumped to should be - `on_failure_jump' or `succeed_n'. Just treat it like an - ordinary jump. For a * loop, it has pushed its failure - point already; if so, discard that as redundant. */ - if ((re_opcode_t) * p != on_failure_jump - && (re_opcode_t) * p != succeed_n) - continue; + /* We can't use `p' to check ahead because we push + a failure point to `p + mcnt' after we do this. */ + p1 = p; - p++; - EXTRACT_NUMBER_AND_INCR(j, p); - p += j; + /* We need to skip no_op's before we look for the + start_memory in case this on_failure_jump is happening as + the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 + against aba. */ + while (p1 < pend && (re_opcode_t) * p1 == no_op) + p1++; - /* If what's on the stack is where we are now, pop it. */ - if (!FAIL_STACK_EMPTY() - && fail_stack.stack[fail_stack.avail - 1].pointer == p) - fail_stack.avail--; + if (p1 < pend && (re_opcode_t) * p1 == start_memory) { + /* We have a new highest active register now. This will + get reset at the start_memory we are about to get to, + but we will have saved all the registers relevant to + this repetition op, as described above. */ + highest_active_reg = *(p1 + 1) + *(p1 + 2); + if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) + lowest_active_reg = *(p1 + 1); + } - continue; + DEBUG_PRINT1(":\n"); + PUSH_FAILURE_POINT(p + mcnt, d, -2); + break; - case on_failure_jump: - case on_failure_keep_string_jump: - handle_on_failure_jump: - EXTRACT_NUMBER_AND_INCR(j, p); + /* A smart repeat ends with `maybe_pop_jump'. + We change it to either `pop_failure_jump' or `jump'. */ + case maybe_pop_jump: + EXTRACT_NUMBER_AND_INCR(mcnt, p); + DEBUG_PRINT2("EXECUTING maybe_pop_jump %d.\n", mcnt); + { + register unsigned char *p2 = p; - /* For some patterns, e.g., `(a?)?', `p+j' here points to the - end of the pattern. We don't want to push such a point, - since when we restore it above, entering the switch will - increment `p' past the end of the pattern. We don't need - to push such a point since we obviously won't find any more - fastmap entries beyond `pend'. Such a pattern can match - the null string, though. */ - if (p + j < pend) { - if (!PUSH_PATTERN_OP(p + j, fail_stack)) { - RESET_FAIL_STACK(); - return -2; - } - } else - bufp->can_be_null = 1; + /* Compare the beginning of the repeat with what in the + pattern follows its end. If we can establish that there + is nothing that they would both match, i.e., that we + would have to backtrack because of (as in, e.g., `a*a') + then we can change to pop_failure_jump, because we'll + never have to backtrack. - if (succeed_n_p) { - EXTRACT_NUMBER_AND_INCR(k, p); /* Skip the n. */ - succeed_n_p = false; - } + This is not true in the case of alternatives: in + `(a|ab)*' we do need to backtrack to the `ab' alternative + (e.g., if the string was `ab'). But instead of trying to + detect that here, the alternative has put on a dummy + failure point which is what we will end up popping. */ - continue; + /* Skip over open/close-group commands. + If what follows this loop is a ...+ construct, + look at what begins its body, since we will have to + match at least one of that. */ + while (1) { + if (p2 + 2 < pend + && ((re_opcode_t) * p2 == stop_memory + || (re_opcode_t) * p2 == start_memory)) + p2 += 3; + else if (p2 + 6 < pend + && (re_opcode_t) * p2 == dummy_failure_jump) + p2 += 6; + else + break; + } + p1 = p + mcnt; + /* p1[0] ... p1[2] are the `on_failure_jump' corresponding + to the `maybe_finalize_jump' of this case. Examine what + follows. */ - case succeed_n: - /* Get to the number of times to succeed. */ - p += 2; + /* If we're at the end of the pattern, we can change. */ + if (p2 == pend) { + /* Consider what happens when matching ":\(.*\)" + against ":/". I don't really understand this code + yet. */ + p[-3] = (unsigned char) pop_failure_jump; + DEBUG_PRINT1 + (" End of pattern: change to `pop_failure_jump'.\n"); + } - /* Increment p past the n for when k != 0. */ - EXTRACT_NUMBER_AND_INCR(k, p); - if (k == 0) { - p -= 4; - succeed_n_p = true; /* Spaghetti code alert. */ - goto handle_on_failure_jump; - } - continue; + else if ((re_opcode_t) * p2 == exactn + || (bufp->newline_anchor + && (re_opcode_t) * p2 == endline)) { + register unsigned char c = + *p2 == (unsigned char) endline ? '\n' : p2[2]; + if ((re_opcode_t) p1[3] == exactn && p1[5] != c) { + p[-3] = (unsigned char) pop_failure_jump; + DEBUG_PRINT3(" %c != %c => pop_failure_jump.\n", + c, p1[5]); + } - case set_number_at: - p += 4; - continue; + else if ((re_opcode_t) p1[3] == charset + || (re_opcode_t) p1[3] == charset_not) { + int not = (re_opcode_t) p1[3] == charset_not; + if (c < (unsigned char) (p1[4] * BYTEWIDTH) + && p1[5 + + c / BYTEWIDTH] & (1 << (c % + BYTEWIDTH))) not + = !not; - case start_memory: - case stop_memory: - p += 2; - continue; + /* `not' is equal to 1 if c would match, which means + that we can't change to pop_failure_jump. */ + if (!not) { + p[-3] = (unsigned char) pop_failure_jump; + DEBUG_PRINT1 + (" No match => pop_failure_jump.\n"); + } + } + } else if ((re_opcode_t) * p2 == charset) { + /* We win if the first character of the loop is not part + of the charset. */ + if ((re_opcode_t) p1[3] == exactn + && !((int) p2[1] * BYTEWIDTH > (int) p1[5] + && (p2[2 + p1[5] / BYTEWIDTH] + & (1 << (p1[5] % BYTEWIDTH))))) { + p[-3] = (unsigned char) pop_failure_jump; + DEBUG_PRINT1(" No match => pop_failure_jump.\n"); + } + else if ((re_opcode_t) p1[3] == charset_not) { + int idx; - default: - abort(); /* We have listed all the cases. */ - } /* switch *p++ */ + /* We win if the charset_not inside the loop + lists every character listed in the charset after. */ + for (idx = 0; idx < (int) p2[1]; idx++) + if (!(p2[2 + idx] == 0 || (idx < (int) p1[4] + && + ((p2 + [2 + + idx] & ~p1[5 + + idx]) + == 0)))) + break; - /* Getting here means we have found the possible starting - characters for one path of the pattern -- and that the empty - string does not match. We need not follow this path further. - Instead, look at the next alternative (remembered on the - stack), or quit if no more. The test at the top of the loop - does these things. */ - path_can_be_null = false; - p = pend; - } /* while p */ + if (idx == p2[1]) { + p[-3] = (unsigned char) pop_failure_jump; + DEBUG_PRINT1 + (" No match => pop_failure_jump.\n"); + } + } else if ((re_opcode_t) p1[3] == charset) { + int idx; - /* Set `can_be_null' for the last path (also the first path, if the - pattern is empty). */ - bufp->can_be_null |= path_can_be_null; + /* We win if the charset inside the loop + has no overlap with the one after the loop. */ + for (idx = 0; + idx < (int) p2[1] && idx < (int) p1[4]; idx++) + if ((p2[2 + idx] & p1[5 + idx]) != 0) + break; - done: - RESET_FAIL_STACK(); - return 0; -} /* re_compile_fastmap */ + if (idx == p2[1] || idx == p1[4]) { + p[-3] = (unsigned char) pop_failure_jump; + DEBUG_PRINT1 + (" No match => pop_failure_jump.\n"); + } + } + } + } + p -= 2; /* Point at relative address again. */ + if ((re_opcode_t) p[-1] != pop_failure_jump) { + p[-1] = (unsigned char) jump; + DEBUG_PRINT1(" Match => jump.\n"); + goto unconditional_jump; + } + /* Note fall through. */ -#ifdef _LIBC -weak_alias(__re_compile_fastmap, re_compile_fastmap) -#endif - /* Set REGS to hold NUM_REGS registers, storing them in STARTS and - ENDS. Subsequent matches using PATTERN_BUFFER and REGS will use - this memory for recording register information. STARTS and ENDS - must be allocated using the malloc library routine, and must each - be at least NUM_REGS * sizeof (regoff_t) bytes long. - If NUM_REGS == 0, then subsequent matches should allocate their own - register data. + /* The end of a simple repeat has a pop_failure_jump back to + its matching on_failure_jump, where the latter will push a + failure point. The pop_failure_jump takes off failure + points put on by this pop_failure_jump's matching + on_failure_jump; we got through the pattern to here from the + matching on_failure_jump, so didn't fail. */ + case pop_failure_jump: + { + /* We need to pass separate storage for the lowest and + highest registers, even though we don't care about the + actual values. Otherwise, we will restore only one + register from the stack, since lowest will == highest in + `pop_failure_point'. */ + active_reg_t dummy_low_reg, dummy_high_reg; + unsigned char *pdummy; + const char *sdummy; - Unless this function is called, the first search or match using - PATTERN_BUFFER will allocate its own register data, without - freeing the old data. */ -void re_set_registers(bufp, regs, num_regs, starts, ends) -struct re_pattern_buffer *bufp; -struct re_registers *regs; -unsigned num_regs; -regoff_t *starts, *ends; -{ - if (num_regs) { - bufp->regs_allocated = REGS_REALLOCATE; - regs->num_regs = num_regs; - regs->start = starts; - regs->end = ends; - } else { - bufp->regs_allocated = REGS_UNALLOCATED; - regs->num_regs = 0; - regs->start = regs->end = (regoff_t *) 0; - } -} + DEBUG_PRINT1("EXECUTING pop_failure_jump.\n"); + POP_FAILURE_POINT(sdummy, pdummy, + dummy_low_reg, dummy_high_reg, + reg_dummy, reg_dummy, reg_info_dummy); + } + /* Note fall through. */ + unconditional_jump: #ifdef _LIBC -weak_alias(__re_set_registers, re_set_registers) + DEBUG_PRINT2("\n%p: ", p); +#else + DEBUG_PRINT2("\n0x%x: ", p); #endif - /* Searching routines. */ -/* Like re_search_2, below, but only one string is specified, and - doesn't let you say where to stop matching. */ -int re_search(bufp, string, size, startpos, range, regs) -struct re_pattern_buffer *bufp; -const char *string; -int size, startpos, range; -struct re_registers *regs; -{ - return re_search_2(bufp, NULL, 0, string, size, startpos, range, - regs, size); -} + /* Note fall through. */ + /* Unconditionally jump (without popping any failure points). */ + case jump: + EXTRACT_NUMBER_AND_INCR(mcnt, p); /* Get the amount to jump. */ + DEBUG_PRINT2("EXECUTING jump %d ", mcnt); + p += mcnt; /* Do the jump. */ #ifdef _LIBC -weak_alias(__re_search, re_search) + DEBUG_PRINT2("(to %p).\n", p); +#else + DEBUG_PRINT2("(to 0x%x).\n", p); #endif -/* Using the compiled pattern in BUFP->buffer, first tries to match the - virtual concatenation of STRING1 and STRING2, starting first at index - STARTPOS, then at STARTPOS + 1, and so on. + break; - STRING1 and STRING2 have length SIZE1 and SIZE2, respectively. - RANGE is how far to scan while trying to match. RANGE = 0 means try - only at STARTPOS; in general, the last start tried is STARTPOS + - RANGE. + /* We need this opcode so we can detect where alternatives end + in `group_match_null_string_p' et al. */ + case jump_past_alt: + DEBUG_PRINT1("EXECUTING jump_past_alt.\n"); + goto unconditional_jump; - In REGS, return the indices of the virtual concatenation of STRING1 - and STRING2 that matched the entire BUFP->buffer and its contained - subexpressions. - Do not consider matching one past the index STOP in the virtual - concatenation of STRING1 and STRING2. + /* Normally, the on_failure_jump pushes a failure point, which + then gets popped at pop_failure_jump. We will end up at + pop_failure_jump, also, and with a pattern of, say, `a+', we + are skipping over the on_failure_jump, so we have to push + something meaningless for pop_failure_jump to pop. */ + case dummy_failure_jump: + DEBUG_PRINT1("EXECUTING dummy_failure_jump.\n"); + /* It doesn't matter what we push for the string here. What + the code at `fail' tests is the value for the pattern. */ + PUSH_FAILURE_POINT(NULL, NULL, -2); + goto unconditional_jump; - We return either the position in the strings at which the match was - found, -1 if no match, or -2 if error (such as failure - stack overflow). */ -int -re_search_2(bufp, string1, size1, string2, size2, startpos, range, regs, - stop) -struct re_pattern_buffer *bufp; -const char *string1, *string2; -int size1, size2; -int startpos; -int range; -struct re_registers *regs; -int stop; -{ - int val; - register char *fastmap = bufp->fastmap; - register RE_TRANSLATE_TYPE translate = bufp->translate; - int total_size = size1 + size2; - int endpos = startpos + range; - /* Check for out-of-range STARTPOS. */ - if (startpos < 0 || startpos > total_size) - return -1; + /* At the end of an alternative, we need to push a dummy failure + point in case we are followed by a `pop_failure_jump', because + we don't want the failure point for the alternative to be + popped. For example, matching `(a|ab)*' against `aab' + requires that we match the `ab' alternative. */ + case push_dummy_failure: + DEBUG_PRINT1("EXECUTING push_dummy_failure.\n"); + /* See comments just above at `dummy_failure_jump' about the + two zeroes. */ + PUSH_FAILURE_POINT(NULL, NULL, -2); + break; - /* Fix up RANGE if it might eventually take us outside - the virtual concatenation of STRING1 and STRING2. - Make sure we won't move STARTPOS below 0 or above TOTAL_SIZE. */ - if (endpos < 0) - range = 0 - startpos; - else if (endpos > total_size) - range = total_size - startpos; + /* Have to succeed matching what follows at least n times. + After that, handle like `on_failure_jump'. */ + case succeed_n: + EXTRACT_NUMBER(mcnt, p + 2); + DEBUG_PRINT2("EXECUTING succeed_n %d.\n", mcnt); - /* If the search isn't to be a backwards one, don't waste time in a - search for a pattern that must be anchored. */ - if (bufp->used > 0 && range > 0 - && ((re_opcode_t) bufp->buffer[0] == begbuf - /* `begline' is like `begbuf' if it cannot match at newlines. */ - || ((re_opcode_t) bufp->buffer[0] == begline - && !bufp->newline_anchor))) { - if (startpos > 0) - return -1; - else - range = 1; - } -#ifdef emacs - /* In a forward search for something that starts with \=. - don't keep searching past point. */ - if (bufp->used > 0 && (re_opcode_t) bufp->buffer[0] == at_dot - && range > 0) { - range = PT - startpos; - if (range <= 0) - return -1; - } -#endif /* emacs */ + assert(mcnt >= 0); + /* Originally, this is how many times we HAVE to succeed. */ + if (mcnt > 0) { + mcnt--; + p += 2; + STORE_NUMBER_AND_INCR(p, mcnt); +#ifdef _LIBC + DEBUG_PRINT3(" Setting %p to %d.\n", p - 2, mcnt); +#else + DEBUG_PRINT3(" Setting 0x%x to %d.\n", p - 2, mcnt); +#endif + } else if (mcnt == 0) { +#ifdef _LIBC + DEBUG_PRINT2(" Setting two bytes from %p to no_op.\n", + p + 2); +#else + DEBUG_PRINT2(" Setting two bytes from 0x%x to no_op.\n", + p + 2); +#endif + p[2] = (unsigned char) no_op; + p[3] = (unsigned char) no_op; + goto on_failure; + } + break; - /* Update the fastmap now if not correct already. */ - if (fastmap && !bufp->fastmap_accurate) - if (re_compile_fastmap(bufp) == -2) - return -2; + case jump_n: + EXTRACT_NUMBER(mcnt, p + 2); + DEBUG_PRINT2("EXECUTING jump_n %d.\n", mcnt); - /* Loop through the string, looking for a place to start matching. */ - for (;;) { - /* If a fastmap is supplied, skip quickly over characters that - cannot be the start of a match. If the pattern can match the - null string, however, we don't need to skip characters; we want - the first null string. */ - if (fastmap && startpos < total_size && !bufp->can_be_null) { - if (range > 0) { /* Searching forwards. */ - register const char *d; - register int lim = 0; - int irange = range; + /* Originally, this is how many times we CAN jump. */ + if (mcnt) { + mcnt--; + STORE_NUMBER(p + 2, mcnt); +#ifdef _LIBC + DEBUG_PRINT3(" Setting %p to %d.\n", p + 2, mcnt); +#else + DEBUG_PRINT3(" Setting 0x%x to %d.\n", p + 2, mcnt); +#endif + goto unconditional_jump; + } + /* If don't have to jump any more, skip over the rest of command. */ + else + p += 4; + break; - if (startpos < size1 && startpos + range >= size1) - lim = range - (size1 - startpos); + case set_number_at: + { + DEBUG_PRINT1("EXECUTING set_number_at.\n"); - d = - (startpos >= - size1 ? string2 - size1 : string1) + startpos; + EXTRACT_NUMBER_AND_INCR(mcnt, p); + p1 = p + mcnt; + EXTRACT_NUMBER_AND_INCR(mcnt, p); +#ifdef _LIBC + DEBUG_PRINT3(" Setting %p to %d.\n", p1, mcnt); +#else + DEBUG_PRINT3(" Setting 0x%x to %d.\n", p1, mcnt); +#endif + STORE_NUMBER(p1, mcnt); + break; + } - /* Written out as an if-else to avoid testing `translate' - inside the loop. */ - if (translate) - while (range > lim && !fastmap[(unsigned char) - translate[ - (unsigned - char) *d++]]) - range--; - else - while (range > lim && !fastmap[(unsigned char) *d++]) - range--; +#if 0 + /* The DEC Alpha C compiler 3.x generates incorrect code for the + test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of + AT_WORD_BOUNDARY, so this code is disabled. Expanding the + macro and introducing temporary variables works around the bug. */ + + case wordbound: + DEBUG_PRINT1("EXECUTING wordbound.\n"); + if (AT_WORD_BOUNDARY(d)) + break; + goto fail; - startpos += irange - range; - } else { /* Searching backwards. */ + case notwordbound: + DEBUG_PRINT1("EXECUTING notwordbound.\n"); + if (AT_WORD_BOUNDARY(d)) + goto fail; + break; +#else + case wordbound: + { + boolean prevchar, thischar; - register char c = (size1 == 0 || startpos >= size1 - ? string2[startpos - size1] - : string1[startpos]); + DEBUG_PRINT1("EXECUTING wordbound.\n"); + if (AT_STRINGS_BEG(d) || AT_STRINGS_END(d)) + break; - if (!fastmap[(unsigned char) TRANSLATE(c)]) - goto advance; - } + prevchar = WORDCHAR_P(d - 1); + thischar = WORDCHAR_P(d); + if (prevchar != thischar) + break; + goto fail; } - /* If can't match the null string, and that's all we have left, fail. */ - if (range >= 0 && startpos == total_size && fastmap - && !bufp->can_be_null) return -1; - - val = re_match_2_internal(bufp, string1, size1, string2, size2, - startpos, regs, stop); -#ifndef REGEX_MALLOC -# ifdef C_ALLOCA - alloca(0); -# endif -#endif - - if (val >= 0) - return startpos; + case notwordbound: + { + boolean prevchar, thischar; - if (val == -2) - return -2; + DEBUG_PRINT1("EXECUTING notwordbound.\n"); + if (AT_STRINGS_BEG(d) || AT_STRINGS_END(d)) + goto fail; - advance: - if (!range) + prevchar = WORDCHAR_P(d - 1); + thischar = WORDCHAR_P(d); + if (prevchar != thischar) + goto fail; break; - else if (range > 0) { - range--; - startpos++; - } else { - range++; - startpos--; } - } - return -1; -} /* re_search_2 */ - -#ifdef _LIBC -weak_alias(__re_search_2, re_search_2) -#endif - /* This converts PTR, a pointer into one of the search strings `string1' - and `string2' into an offset from the beginning of that string. */ -#define POINTER_TO_OFFSET(ptr) \ - (FIRST_STRING_P (ptr) \ - ? ((regoff_t) ((ptr) - string1)) \ - : ((regoff_t) ((ptr) - string2 + size1))) -/* Macros for dealing with the split strings in re_match_2. */ -#define MATCHING_IN_FIRST_STRING (dend == end_match_1) -/* Call before fetching a character with *d. This switches over to - string2 if necessary. */ -#define PREFETCH() \ - while (d == dend) \ - { \ - /* End of string2 => fail. */ \ - if (dend == end_match_2) \ - goto fail; \ - /* End of string1 => advance to string2. */ \ - d = string2; \ - dend = end_match_2; \ - } -/* Test if at very beginning or at very end of the virtual concatenation - of `string1' and `string2'. If only one string, it's `string2'. */ -#define AT_STRINGS_BEG(d) ((d) == (size1 ? string1 : string2) || !size2) -#define AT_STRINGS_END(d) ((d) == end2) -/* Test if D points to a character which is word-constituent. We have - two special cases to check for: if past the end of string1, look at - the first character in string2; and if before the beginning of - string2, look at the last character in string1. */ -#define WORDCHAR_P(d) \ - (SYNTAX ((d) == end1 ? *string2 \ - : (d) == string2 - 1 ? *(end1 - 1) : *(d)) \ - == Sword) -/* Disabled due to a compiler bug -- see comment at case wordbound */ -#if 0 -/* Test if the character before D and the one at D differ with respect - to being word-constituent. */ -#define AT_WORD_BOUNDARY(d) \ - (AT_STRINGS_BEG (d) || AT_STRINGS_END (d) \ - || WORDCHAR_P (d - 1) != WORDCHAR_P (d)) #endif -/* Free everything we malloc. */ -#ifdef MATCH_MAY_ALLOCATE -# define FREE_VAR(var) if (var) REGEX_FREE (var); var = NULL -# define FREE_VARIABLES() \ - do { \ - REGEX_FREE_STACK (fail_stack.stack); \ - FREE_VAR (regstart); \ - FREE_VAR (regend); \ - FREE_VAR (old_regstart); \ - FREE_VAR (old_regend); \ - FREE_VAR (best_regstart); \ - FREE_VAR (best_regend); \ - FREE_VAR (reg_info); \ - FREE_VAR (reg_dummy); \ - FREE_VAR (reg_info_dummy); \ - } while (0) -#else -# define FREE_VARIABLES() ((void)0) /* Do nothing! But inhibit gcc warning. */ -#endif /* not MATCH_MAY_ALLOCATE */ -/* These values must meet several constraints. They must not be valid - register values; since we have a limit of 255 registers (because - we use only one byte in the pattern for the register number), we can - use numbers larger than 255. They must differ by 1, because of - NUM_FAILURE_ITEMS above. And the value for the lowest register must - be larger than the value for the highest register, so we do not try - to actually save any registers when none are active. */ -#define NO_HIGHEST_ACTIVE_REG (1 << BYTEWIDTH) -#define NO_LOWEST_ACTIVE_REG (NO_HIGHEST_ACTIVE_REG + 1) - /* Matching routines. */ -#ifndef emacs /* Emacs never uses this. */ -/* re_match is like re_match_2 except it takes only a single string. */ -int re_match(bufp, string, size, pos, regs) -struct re_pattern_buffer *bufp; -const char *string; -int size, pos; -struct re_registers *regs; -{ - int result = re_match_2_internal(bufp, NULL, 0, string, size, - pos, regs, size); -# ifndef REGEX_MALLOC -# ifdef C_ALLOCA - alloca(0); -# endif -# endif - return result; -} + case wordbeg: + DEBUG_PRINT1("EXECUTING wordbeg.\n"); + if (WORDCHAR_P(d) && (AT_STRINGS_BEG(d) || !WORDCHAR_P(d - 1))) + break; + goto fail; -# ifdef _LIBC -weak_alias(__re_match, re_match) -# endif -#endif /* not emacs */ -static boolean group_match_null_string_p _RE_ARGS((unsigned char **p, - unsigned char *end, - register_info_type * + case wordend: + DEBUG_PRINT1("EXECUTING wordend.\n"); + if (!AT_STRINGS_BEG(d) && WORDCHAR_P(d - 1) + && (!WORDCHAR_P(d) || AT_STRINGS_END(d))) + break; + goto fail; - reg_info)); -static boolean alt_match_null_string_p -_RE_ARGS( +#ifdef emacs + case before_dot: + DEBUG_PRINT1("EXECUTING before_dot.\n"); + if (PTR_CHAR_POS((unsigned char *) d) >= point) + goto fail; + break; - (unsigned char *p, unsigned char *end, - register_info_type * reg_info)); -static boolean common_op_match_null_string_p -_RE_ARGS( + case at_dot: + DEBUG_PRINT1("EXECUTING at_dot.\n"); + if (PTR_CHAR_POS((unsigned char *) d) != point) + goto fail; + break; - (unsigned char **p, unsigned char *end, - register_info_type * reg_info)); -static int bcmp_translate -_RE_ARGS((const char *s1, const char *s2, int len, char *translate)); + case after_dot: + DEBUG_PRINT1("EXECUTING after_dot.\n"); + if (PTR_CHAR_POS((unsigned char *) d) <= point) + goto fail; + break; -/* re_match_2 matches the compiled pattern in BUFP against the - the (virtual) concatenation of STRING1 and STRING2 (of length SIZE1 - and SIZE2, respectively). We start matching at POS, and stop - matching at STOP. + case syntaxspec: + DEBUG_PRINT2("EXECUTING syntaxspec %d.\n", mcnt); + mcnt = *p++; + goto matchsyntax; - If REGS is non-null and the `no_sub' field of BUFP is nonzero, we - store offsets for the substring each group matched in REGS. See the - documentation for exactly how many groups we fill. + case wordchar: + DEBUG_PRINT1("EXECUTING Emacs wordchar.\n"); + mcnt = (int) Sword; + matchsyntax: + PREFETCH(); + /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ + d++; + if (SYNTAX(d[-1]) != (enum syntaxcode) mcnt) + goto fail; + SET_REGS_MATCHED(); + break; - We return -1 if no match, -2 if an internal error (such as the - failure stack overflowing). Otherwise, we return the length of the - matched substring. */ + case notsyntaxspec: + DEBUG_PRINT2("EXECUTING notsyntaxspec %d.\n", mcnt); + mcnt = *p++; + goto matchnotsyntax; -int re_match_2(bufp, string1, size1, string2, size2, pos, regs, stop) -struct re_pattern_buffer *bufp; -const char *string1, *string2; -int size1, size2; -int pos; -struct re_registers *regs; -int stop; -{ - int result = re_match_2_internal(bufp, string1, size1, string2, size2, - pos, regs, stop); + case notwordchar: + DEBUG_PRINT1("EXECUTING Emacs notwordchar.\n"); + mcnt = (int) Sword; + matchnotsyntax: + PREFETCH(); + /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ + d++; + if (SYNTAX(d[-1]) == (enum syntaxcode) mcnt) + goto fail; + SET_REGS_MATCHED(); + break; -#ifndef REGEX_MALLOC -# ifdef C_ALLOCA - alloca(0); -# endif -#endif - return result; -} +#else /* not emacs */ + case wordchar: + DEBUG_PRINT1("EXECUTING non-Emacs wordchar.\n"); + PREFETCH(); + if (!WORDCHAR_P(d)) + goto fail; + SET_REGS_MATCHED(); + d++; + break; -#ifdef _LIBC -weak_alias(__re_match_2, re_match_2) -#endif -/* This is a separate function so that we can force an alloca cleanup - afterwards. */ -static int -re_match_2_internal(bufp, string1, size1, string2, size2, pos, regs, stop) -struct re_pattern_buffer *bufp; -const char *string1, *string2; -int size1, size2; -int pos; -struct re_registers *regs; -int stop; -{ - /* General temporaries. */ - int mcnt; - unsigned char *p1; + case notwordchar: + DEBUG_PRINT1("EXECUTING non-Emacs notwordchar.\n"); + PREFETCH(); + if (WORDCHAR_P(d)) + goto fail; + SET_REGS_MATCHED(); + d++; + break; +#endif /* not emacs */ - /* Just past the end of the corresponding string. */ - const char *end1, *end2; + default: + abort(); + } + continue; /* Successfully executed one pattern command; keep going. */ - /* Pointers into string1 and string2, just past the last characters in - each to consider matching. */ - const char *end_match_1, *end_match_2; - /* Where we are in the data, and the end of the current string. */ - const char *d, *dend; + /* We goto here if a matching operation fails. */ + fail: + if (!FAIL_STACK_EMPTY()) { /* A restart point is known. Restore to that state. */ + DEBUG_PRINT1("\nFAIL:\n"); + POP_FAILURE_POINT(d, p, + lowest_active_reg, highest_active_reg, + regstart, regend, reg_info); - /* Where we are in the pattern, and the end of the pattern. */ - unsigned char *p = bufp->buffer; - register unsigned char *pend = p + bufp->used; + /* If this failure point is a dummy, try the next one. */ + if (!p) + goto fail; - /* Mark the opcode just after a start_memory, so we can test for an - empty subpattern when we get to the stop_memory. */ - unsigned char *just_past_start_mem = 0; + /* If we failed to the end of the pattern, don't examine *p. */ + assert(p <= pend); + if (p < pend) { + boolean is_a_jump_n = false; - /* We use this to map every character in the string. */ - RE_TRANSLATE_TYPE translate = bufp->translate; + /* If failed to a backwards jump that's part of a repetition + loop, need to pop this failure point and use the next one. */ + switch ((re_opcode_t) * p) { + case jump_n: + is_a_jump_n = true; + case maybe_pop_jump: + case pop_failure_jump: + case jump: + p1 = p + 1; + EXTRACT_NUMBER_AND_INCR(mcnt, p1); + p1 += mcnt; - /* Failure point stack. Each place that can handle a failure further - down the line pushes a failure point on this stack. It consists of - restart, regend, and reg_info for all registers corresponding to - the subexpressions we're currently inside, plus the number of such - registers, and, finally, two char *'s. The first char * is where - to resume scanning the pattern; the second one is where to resume - scanning the strings. If the latter is zero, the failure point is - a ``dummy''; if a failure happens and the failure point is a dummy, - it gets discarded and the next next one is tried. */ -#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ - fail_stack_type fail_stack; -#endif -#ifdef DEBUG - static unsigned failure_id; - unsigned nfailure_points_pushed = 0, nfailure_points_popped = 0; -#endif + if ((is_a_jump_n && (re_opcode_t) * p1 == succeed_n) + || (!is_a_jump_n + && (re_opcode_t) * p1 == on_failure_jump)) + goto fail; + break; + default: + /* do nothing */ ; + } + } -#ifdef REL_ALLOC - /* This holds the pointer to the failure stack, when - it is allocated relocatably. */ - fail_stack_elt_t *failure_stack_ptr; -#endif + if (d >= string1 && d <= end1) + dend = end_match_1; + } else + break; /* Matching at this starting point really fails. */ + } /* for (;;) */ - /* We fill all the registers internally, independent of what we - return, for use in backreferences. The number here includes - an element for register zero. */ - size_t num_regs = bufp->re_nsub + 1; + if (best_regs_set) + goto restore_best_regs; - /* The currently active registers. */ - active_reg_t lowest_active_reg = NO_LOWEST_ACTIVE_REG; - active_reg_t highest_active_reg = NO_HIGHEST_ACTIVE_REG; + FREE_VARIABLES(); - /* Information on the contents of registers. These are pointers into - the input strings; they record just what was matched (on this - attempt) by a subexpression part of the pattern, that is, the - regnum-th regstart pointer points to where in the pattern we began - matching and the regnum-th regend points to right after where we - stopped matching the regnum-th subexpression. (The zeroth register - keeps track of what the whole pattern matches.) */ -#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ - const char **regstart, **regend; -#endif + return -1; /* Failure to match. */ +} /* re_match_2 */ + +/* Subroutine definitions for re_match_2. */ - /* If a group that's operated upon by a repetition operator fails to - match anything, then the register for its start will need to be - restored because it will have been set to wherever in the string we - are when we last see its open-group operator. Similarly for a - register's end. */ -#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ - const char **old_regstart, **old_regend; -#endif - /* The is_active field of reg_info helps us keep track of which (possibly - nested) subexpressions we are currently in. The matched_something - field of reg_info[reg_num] helps us tell whether or not we have - matched any of the pattern so far this time through the reg_num-th - subexpression. These two fields get reset each time through any - loop their register is in. */ -#ifdef MATCH_MAY_ALLOCATE /* otherwise, this is global. */ - register_info_type *reg_info; -#endif +/* We are passed P pointing to a register number after a start_memory. - /* The following record the register info as found in the above - variables when we find a match better than any we've seen before. - This happens as we backtrack through the failure points, which in - turn happens only if we have not yet matched the entire string. */ - unsigned best_regs_set = false; + Return true if the pattern up to the corresponding stop_memory can + match the empty string, and false otherwise. -#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ - const char **best_regstart, **best_regend; -#endif + If we find the matching stop_memory, sets P to point to one past its number. + Otherwise, sets P to an undefined byte less than or equal to END. - /* Logically, this is `best_regend[0]'. But we don't want to have to - allocate space for that if we're not allocating space for anything - else (see below). Also, we never need info about register 0 for - any of the other register vectors, and it seems rather a kludge to - treat `best_regend' differently than the rest. So we keep track of - the end of the best match so far in a separate variable. We - initialize this to NULL so that when we backtrack the first time - and need to test it, it's not garbage. */ - const char *match_end = NULL; + We don't handle duplicates properly (yet). */ - /* This helps SET_REGS_MATCHED avoid doing redundant work. */ - int set_regs_matched_done = 0; +static boolean group_match_null_string_p(p, end, reg_info) +unsigned char **p, *end; +register_info_type *reg_info; +{ + int mcnt; - /* Used when we pop values we don't care about. */ -#ifdef MATCH_MAY_ALLOCATE /* otherwise, these are global. */ - const char **reg_dummy; - register_info_type *reg_info_dummy; -#endif + /* Point to after the args to the start_memory. */ + unsigned char *p1 = *p + 2; -#ifdef DEBUG - /* Counts the total number of registers pushed. */ - unsigned num_regs_pushed = 0; -#endif + while (p1 < end) { + /* Skip over opcodes that can match nothing, and return true or + false, as appropriate, when we get to one that can't, or to the + matching stop_memory. */ - DEBUG_PRINT1("\n\nEntering re_match_2.\n"); + switch ((re_opcode_t) * p1) { + /* Could be either a loop or a series of alternatives. */ + case on_failure_jump: + p1++; + EXTRACT_NUMBER_AND_INCR(mcnt, p1); - INIT_FAIL_STACK(); + /* If the next operation is not a jump backwards in the + pattern. */ -#ifdef MATCH_MAY_ALLOCATE - /* Do not bother to initialize all the register variables if there are - no groups in the pattern, as it takes a fair amount of time. If - there are groups, we include space for register 0 (the whole - pattern), even though we never use it, since it simplifies the - array indexing. We should fix this. */ - if (bufp->re_nsub) { - regstart = REGEX_TALLOC(num_regs, const char *); - regend = REGEX_TALLOC(num_regs, const char *); - old_regstart = REGEX_TALLOC(num_regs, const char *); - old_regend = REGEX_TALLOC(num_regs, const char *); - best_regstart = REGEX_TALLOC(num_regs, const char *); - best_regend = REGEX_TALLOC(num_regs, const char *); + if (mcnt >= 0) { + /* Go through the on_failure_jumps of the alternatives, + seeing if any of the alternatives cannot match nothing. + The last alternative starts with only a jump, + whereas the rest start with on_failure_jump and end + with a jump, e.g., here is the pattern for `a|b|c': - reg_info = REGEX_TALLOC(num_regs, register_info_type); - reg_dummy = REGEX_TALLOC(num_regs, const char *); + /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 + /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 + /exactn/1/c - reg_info_dummy = REGEX_TALLOC(num_regs, register_info_type); + So, we have to first go through the first (n-1) + alternatives and then deal with the last one separately. */ - if (!(regstart && regend && old_regstart && old_regend && reg_info - && best_regstart && best_regend && reg_dummy - && reg_info_dummy)) { - FREE_VARIABLES(); - return -2; - } - } else { - /* We must initialize all our variables to NULL, so that - `FREE_VARIABLES' doesn't try to free them. */ - regstart = regend = old_regstart = old_regend = best_regstart - = best_regend = reg_dummy = NULL; - reg_info = reg_info_dummy = (register_info_type *) NULL; - } -#endif /* MATCH_MAY_ALLOCATE */ - /* The starting position is bogus. */ - if (pos < 0 || pos > size1 + size2) { - FREE_VARIABLES(); - return -1; - } + /* Deal with the first (n-1) alternatives, which start + with an on_failure_jump (see above) that jumps to right + past a jump_past_alt. */ - /* Initialize subexpression text positions to -1 to mark ones that no - start_memory/stop_memory has been seen for. Also initialize the - register information struct. */ - for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) { - regstart[mcnt] = regend[mcnt] - = old_regstart[mcnt] = old_regend[mcnt] = REG_UNSET_VALUE; + while ((re_opcode_t) p1[mcnt - 3] == jump_past_alt) { + /* `mcnt' holds how many bytes long the alternative + is, including the ending `jump_past_alt' and + its number. */ - REG_MATCH_NULL_STRING_P(reg_info[mcnt]) = MATCH_NULL_UNSET_VALUE; - IS_ACTIVE(reg_info[mcnt]) = 0; - MATCHED_SOMETHING(reg_info[mcnt]) = 0; - EVER_MATCHED_SOMETHING(reg_info[mcnt]) = 0; - } + if (!alt_match_null_string_p(p1, p1 + mcnt - 3, + reg_info)) return false; - /* We move `string1' into `string2' if the latter's empty -- but not if - `string1' is null. */ - if (size2 == 0 && string1 != NULL) { - string2 = string1; - size2 = size1; - string1 = 0; - size1 = 0; - } - end1 = string1 + size1; - end2 = string2 + size2; + /* Move to right after this alternative, including the + jump_past_alt. */ + p1 += mcnt; - /* Compute where to stop matching, within the two strings. */ - if (stop <= size1) { - end_match_1 = string1 + stop; - end_match_2 = string2; - } else { - end_match_1 = end1; - end_match_2 = string2 + stop - size1; - } + /* Break if it's the beginning of an n-th alternative + that doesn't begin with an on_failure_jump. */ + if ((re_opcode_t) * p1 != on_failure_jump) + break; - /* `p' scans through the pattern as `d' scans through the data. - `dend' is the end of the input string that `d' points within. `d' - is advanced into the following input string whenever necessary, but - this happens before fetching; therefore, at the beginning of the - loop, `d' can be pointing at the end of a string, but it cannot - equal `string2'. */ - if (size1 > 0 && pos <= size1) { - d = string1 + pos; - dend = end_match_1; - } else { - d = string2 + pos - size1; - dend = end_match_2; - } + /* Still have to check that it's not an n-th + alternative that starts with an on_failure_jump. */ + p1++; + EXTRACT_NUMBER_AND_INCR(mcnt, p1); + if ((re_opcode_t) p1[mcnt - 3] != jump_past_alt) { + /* Get to the beginning of the n-th alternative. */ + p1 -= 3; + break; + } + } - DEBUG_PRINT1("The compiled pattern is:\n"); - DEBUG_PRINT_COMPILED_PATTERN(bufp, p, pend); - DEBUG_PRINT1("The string to match is: `"); - DEBUG_PRINT_DOUBLE_STRING(d, string1, size1, string2, size2); - DEBUG_PRINT1("'\n"); + /* Deal with the last alternative: go back and get number + of the `jump_past_alt' just before it. `mcnt' contains + the length of the alternative. */ + EXTRACT_NUMBER(mcnt, p1 - 2); - /* This loops over pattern commands. It exits by returning from the - function if the match is complete, or it drops through if the match - fails at this starting point in the input data. */ - for (;;) { -#ifdef _LIBC - DEBUG_PRINT2("\n%p: ", p); -#else - DEBUG_PRINT2("\n0x%x: ", p); -#endif + if (!alt_match_null_string_p(p1, p1 + mcnt, reg_info)) + return false; - if (p == pend) { /* End of pattern means we might have succeeded. */ - DEBUG_PRINT1("end of pattern ... "); + p1 += mcnt; /* Get past the n-th alternative. */ + } /* if mcnt > 0 */ + break; - /* If we haven't matched the entire string, and we want the - longest match, try backtracking. */ - if (d != end_match_2) { - /* 1 if this match ends in the same string (string1 or string2) - as the best previous match. */ - boolean same_str_p = (FIRST_STRING_P(match_end) - == MATCHING_IN_FIRST_STRING); - /* 1 if this match is the best seen so far. */ - boolean best_match_p; + case stop_memory: + assert(p1[1] == **p); + *p = p1 + 2; + return true; - /* AIX compiler got confused when this was combined - with the previous declaration. */ - if (same_str_p) - best_match_p = d > match_end; - else - best_match_p = !MATCHING_IN_FIRST_STRING; - DEBUG_PRINT1("backtracking.\n"); + default: + if (!common_op_match_null_string_p(&p1, end, reg_info)) + return false; + } + } /* while p1 < end */ - if (!FAIL_STACK_EMPTY()) { /* More failure points to try. */ + return false; +} /* group_match_null_string_p */ - /* If exceeds best match so far, save it. */ - if (!best_regs_set || best_match_p) { - best_regs_set = true; - match_end = d; - DEBUG_PRINT1("\nSAVING match as best so far.\n"); +/* Similar to group_match_null_string_p, but doesn't deal with alternatives: + It expects P to be the first byte of a single alternative and END one + byte past the last. The alternative can contain groups. */ - for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) { - best_regstart[mcnt] = regstart[mcnt]; - best_regend[mcnt] = regend[mcnt]; - } - } - goto fail; - } +static boolean alt_match_null_string_p(p, end, reg_info) +unsigned char *p, *end; +register_info_type *reg_info; +{ + int mcnt; + unsigned char *p1 = p; - /* If no failure points, don't restore garbage. And if - last match is real best match, don't restore second - best one. */ - else if (best_regs_set && !best_match_p) { - restore_best_regs: - /* Restore best match. It may happen that `dend == - end_match_1' while the restored d is in string2. - For example, the pattern `x.*y.*z' against the - strings `x-' and `y-z-', if the two strings are - not consecutive in memory. */ - DEBUG_PRINT1("Restoring best registers.\n"); + while (p1 < end) { + /* Skip over opcodes that can match nothing, and break when we get + to one that can't. */ - d = match_end; - dend = ((d >= string1 && d <= end1) - ? end_match_1 : end_match_2); + switch ((re_opcode_t) * p1) { + /* It's a loop. */ + case on_failure_jump: + p1++; + EXTRACT_NUMBER_AND_INCR(mcnt, p1); + p1 += mcnt; + break; - for (mcnt = 1; (unsigned) mcnt < num_regs; mcnt++) { - regstart[mcnt] = best_regstart[mcnt]; - regend[mcnt] = best_regend[mcnt]; - } - } - } - /* d != end_match_2 */ - succeed_label: - DEBUG_PRINT1("Accepting match.\n"); + default: + if (!common_op_match_null_string_p(&p1, end, reg_info)) + return false; + } + } /* while p1 < end */ - /* If caller wants register contents data back, do it. */ - if (regs && !bufp->no_sub) { - /* Have the register data arrays been allocated? */ - if (bufp->regs_allocated == REGS_UNALLOCATED) { /* No. So allocate them with malloc. We need one - extra element beyond `num_regs' for the `-1' marker - GNU code uses. */ - regs->num_regs = MAX(RE_NREGS, num_regs + 1); - regs->start = TALLOC(regs->num_regs, regoff_t); - regs->end = TALLOC(regs->num_regs, regoff_t); - if (regs->start == NULL || regs->end == NULL) { - FREE_VARIABLES(); - return -2; - } - bufp->regs_allocated = REGS_REALLOCATE; - } else if (bufp->regs_allocated == REGS_REALLOCATE) { /* Yes. If we need more elements than were already - allocated, reallocate them. If we need fewer, just - leave it alone. */ - if (regs->num_regs < num_regs + 1) { - regs->num_regs = num_regs + 1; - RETALLOC(regs->start, regs->num_regs, regoff_t); - RETALLOC(regs->end, regs->num_regs, regoff_t); - if (regs->start == NULL || regs->end == NULL) { - FREE_VARIABLES(); - return -2; - } - } - } else { - /* These braces fend off a "empty body in an else-statement" - warning under GCC when assert expands to nothing. */ - assert(bufp->regs_allocated == REGS_FIXED); - } + return true; +} /* alt_match_null_string_p */ - /* Convert the pointer data in `regstart' and `regend' to - indices. Register zero has to be set differently, - since we haven't kept track of any info for it. */ - if (regs->num_regs > 0) { - regs->start[0] = pos; - regs->end[0] = (MATCHING_IN_FIRST_STRING - ? ((regoff_t) (d - string1)) - : ((regoff_t) (d - string2 + size1))); - } - /* Go through the first `min (num_regs, regs->num_regs)' - registers, since that is all we initialized. */ - for (mcnt = 1; - (unsigned) mcnt < MIN(num_regs, regs->num_regs); - mcnt++) { - if (REG_UNSET(regstart[mcnt]) - || REG_UNSET(regend[mcnt])) regs->start[mcnt] = - regs->end[mcnt] = -1; - else { - regs->start[mcnt] - = (regoff_t) POINTER_TO_OFFSET(regstart[mcnt]); - regs->end[mcnt] - = (regoff_t) POINTER_TO_OFFSET(regend[mcnt]); - } - } +/* Deals with the ops common to group_match_null_string_p and + alt_match_null_string_p. - /* If the regs structure we return has more elements than - were in the pattern, set the extra elements to -1. If - we (re)allocated the registers, this is the case, - because we always allocate enough to have at least one - -1 at the end. */ - for (mcnt = num_regs; (unsigned) mcnt < regs->num_regs; - mcnt++) - regs->start[mcnt] = regs->end[mcnt] = -1; - } - /* regs && !bufp->no_sub */ - DEBUG_PRINT4 - ("%u failure points pushed, %u popped (%u remain).\n", - nfailure_points_pushed, nfailure_points_popped, - nfailure_points_pushed - nfailure_points_popped); - DEBUG_PRINT2("%u registers pushed.\n", num_regs_pushed); + Sets P to one after the op and its arguments, if any. */ + +static boolean common_op_match_null_string_p(p, end, reg_info) +unsigned char **p, *end; +register_info_type *reg_info; +{ + int mcnt; + boolean ret; + int reg_no; + unsigned char *p1 = *p; - mcnt = d - pos - (MATCHING_IN_FIRST_STRING - ? string1 : string2 - size1); + switch ((re_opcode_t) * p1++) { + case no_op: + case begline: + case endline: + case begbuf: + case endbuf: + case wordbeg: + case wordend: + case wordbound: + case notwordbound: +#ifdef emacs + case before_dot: + case at_dot: + case after_dot: +#endif + break; - DEBUG_PRINT2("Returning %d from re_match_2.\n", mcnt); + case start_memory: + reg_no = *p1; + assert(reg_no > 0 && reg_no <= MAX_REGNUM); + ret = group_match_null_string_p(&p1, end, reg_info); - FREE_VARIABLES(); - return mcnt; - } + /* Have to set this here in case we're checking a group which + contains a group and a back reference to it. */ - /* Otherwise match next pattern command. */ - switch (SWITCH_ENUM_CAST((re_opcode_t) * p++)) { - /* Ignore these. Used to ignore the n of succeed_n's which - currently have n == 0. */ - case no_op: - DEBUG_PRINT1("EXECUTING no_op.\n"); - break; + if (REG_MATCH_NULL_STRING_P(reg_info[reg_no]) == + MATCH_NULL_UNSET_VALUE) + REG_MATCH_NULL_STRING_P(reg_info[reg_no]) = ret; - case succeed: - DEBUG_PRINT1("EXECUTING succeed.\n"); - goto succeed_label; + if (!ret) + return false; + break; - /* Match the next n pattern characters exactly. The following - byte in the pattern defines n, and the n bytes after that - are the characters to match. */ - case exactn: - mcnt = *p++; - DEBUG_PRINT2("EXECUTING exactn %d.\n", mcnt); + /* If this is an optimized succeed_n for zero times, make the jump. */ + case jump: + EXTRACT_NUMBER_AND_INCR(mcnt, p1); + if (mcnt >= 0) + p1 += mcnt; + else + return false; + break; - /* This is written out as an if-else so we don't waste time - testing `translate' inside the loop. */ - if (translate) { - do { - PREFETCH(); - if ((unsigned char) translate[(unsigned char) *d++] - != (unsigned char) *p++) - goto fail; - } - while (--mcnt); - } else { - do { - PREFETCH(); - if (*d++ != (char) *p++) - goto fail; - } - while (--mcnt); - } - SET_REGS_MATCHED(); - break; + case succeed_n: + /* Get to the number of times to succeed. */ + p1 += 2; + EXTRACT_NUMBER_AND_INCR(mcnt, p1); + if (mcnt == 0) { + p1 -= 4; + EXTRACT_NUMBER_AND_INCR(mcnt, p1); + p1 += mcnt; + } else + return false; + break; - /* Match any character except possibly a newline or a null. */ - case anychar: - DEBUG_PRINT1("EXECUTING anychar.\n"); + case duplicate: + if (!REG_MATCH_NULL_STRING_P(reg_info[*p1])) + return false; + break; - PREFETCH(); + case set_number_at: + p1 += 4; - if ((!(bufp->syntax & RE_DOT_NEWLINE) && TRANSLATE(*d) == '\n') - || (bufp->syntax & RE_DOT_NOT_NULL - && TRANSLATE(*d) == '\000')) goto fail; + default: + /* All other opcodes mean we cannot match the empty string. */ + return false; + } - SET_REGS_MATCHED(); - DEBUG_PRINT2(" Matched `%d'.\n", *d); - d++; - break; + *p = p1; + return true; +} /* common_op_match_null_string_p */ - case charset: - case charset_not: - { - register unsigned char c; - boolean not = (re_opcode_t) * (p - 1) == charset_not; +/* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN + bytes; nonzero otherwise. */ - DEBUG_PRINT2("EXECUTING charset%s.\n", not ? "_not" : ""); +static int bcmp_translate(s1, s2, len, translate) +const char *s1, *s2; +register int len; +RE_TRANSLATE_TYPE translate; +{ + register const unsigned char *p1 = (const unsigned char *) s1; + register const unsigned char *p2 = (const unsigned char *) s2; - PREFETCH(); - c = TRANSLATE(*d); /* The character to match. */ + while (len) { + if (translate[*p1++] != translate[*p2++]) + return 1; + len--; + } + return 0; +} + +/* Entry points for GNU code. */ - /* Cast to `unsigned' instead of `unsigned char' in case the - bit list is a full 32 bytes long. */ - if (c < (unsigned) (*p * BYTEWIDTH) - && p[1 + c / BYTEWIDTH] & (1 << (c % BYTEWIDTH))) - not = !not; +/* re_compile_pattern is the GNU regular expression compiler: it + compiles PATTERN (of length SIZE) and puts the result in BUFP. + Returns 0 if the pattern was valid, otherwise an error string. - p += 1 + *p; + Assumes the `allocated' (and perhaps `buffer') and `translate' fields + are set in BUFP on entry. - if (!not) - goto fail; + We call regex_compile to do the actual compilation. */ - SET_REGS_MATCHED(); - d++; - break; - } +const char *re_compile_pattern(pattern, length, bufp) +const char *pattern; +size_t length; +struct re_pattern_buffer *bufp; +{ + reg_errcode_t ret; + /* GNU code is written to assume at least RE_NREGS registers will be set + (and at least one extra will be -1). */ + bufp->regs_allocated = REGS_UNALLOCATED; - /* The beginning of a group is represented by start_memory. - The arguments are the register number in the next byte, and the - number of groups inner to this one in the next. The text - matched within the group is recorded (in the internal - registers data structure) under the register number. */ - case start_memory: - DEBUG_PRINT3("EXECUTING start_memory %d (%d):\n", *p, p[1]); + /* And GNU code determines whether or not to get register information + by passing null for the REGS argument to re_match, etc., not by + setting no_sub. */ + bufp->no_sub = 0; - /* Find out if this group can match the empty string. */ - p1 = p; /* To send to group_match_null_string_p. */ + /* Match anchors at newline. */ + bufp->newline_anchor = 1; - if (REG_MATCH_NULL_STRING_P(reg_info[*p]) == - MATCH_NULL_UNSET_VALUE) - REG_MATCH_NULL_STRING_P(reg_info[*p]) = - group_match_null_string_p(&p1, pend, reg_info); + ret = regex_compile(pattern, length, re_syntax_options, bufp); - /* Save the position in the string where we were the last time - we were at this open-group operator in case the group is - operated upon by a repetition operator, e.g., with `(a*)*b' - against `ab'; then we want to ignore where we are now in - the string in case this attempt to match fails. */ - old_regstart[*p] = REG_MATCH_NULL_STRING_P(reg_info[*p]) - ? REG_UNSET(regstart[*p]) ? d : regstart[*p] - : regstart[*p]; - DEBUG_PRINT2(" old_regstart: %d\n", - POINTER_TO_OFFSET(old_regstart[*p])); + if (!ret) + return NULL; + return gettext(re_error_msgid + re_error_msgid_idx[(int) ret]); +} - regstart[*p] = d; - DEBUG_PRINT2(" regstart: %d\n", - POINTER_TO_OFFSET(regstart[*p])); +#ifdef _LIBC +weak_alias(__re_compile_pattern, re_compile_pattern) +#endif + /* Entry points compatible with 4.2 BSD regex library. We don't define + them unless specifically requested. */ +#if defined _REGEX_RE_COMP || defined _LIBC +/* BSD has one and only one pattern buffer. */ +static struct re_pattern_buffer re_comp_buf; - IS_ACTIVE(reg_info[*p]) = 1; - MATCHED_SOMETHING(reg_info[*p]) = 0; +char * +#ifdef _LIBC +/* Make these definitions weak in libc, so POSIX programs can redefine + these names if they don't use our functions, and still use + regcomp/regexec below without link errors. */ weak_function +#endif +re_comp(s) +const char *s; +{ + reg_errcode_t ret; - /* Clear this whenever we change the register activity status. */ - set_regs_matched_done = 0; + if (!s) { + if (!re_comp_buf.buffer) + return gettext("No previous regular expression"); + return 0; + } - /* This is the new highest active register. */ - highest_active_reg = *p; + if (!re_comp_buf.buffer) { + re_comp_buf.buffer = (unsigned char *) malloc(200); + if (re_comp_buf.buffer == NULL) + return (char *) gettext(re_error_msgid + + + re_error_msgid_idx[(int) REG_ESPACE]); + re_comp_buf.allocated = 200; - /* If nothing was active before, this is the new lowest active - register. */ - if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) - lowest_active_reg = *p; + re_comp_buf.fastmap = (char *) malloc(1 << BYTEWIDTH); + if (re_comp_buf.fastmap == NULL) + return (char *) gettext(re_error_msgid + + + re_error_msgid_idx[(int) REG_ESPACE]); + } + + /* Since `re_exec' always passes NULL for the `regs' argument, we + don't need to initialize the pattern buffer fields which affect it. */ - /* Move past the register number and inner group count. */ - p += 2; - just_past_start_mem = p; + /* Match anchors at newlines. */ + re_comp_buf.newline_anchor = 1; - break; + ret = regex_compile(s, strlen(s), re_syntax_options, &re_comp_buf); + if (!ret) + return NULL; - /* The stop_memory opcode represents the end of a group. Its - arguments are the same as start_memory's: the register - number, and the number of inner groups. */ - case stop_memory: - DEBUG_PRINT3("EXECUTING stop_memory %d (%d):\n", *p, p[1]); + /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ + return (char *) gettext(re_error_msgid + + re_error_msgid_idx[(int) ret]); +} - /* We need to save the string position the last time we were at - this close-group operator in case the group is operated - upon by a repetition operator, e.g., with `((a*)*(b*)*)*' - against `aba'; then we want to ignore where we are now in - the string in case this attempt to match fails. */ - old_regend[*p] = REG_MATCH_NULL_STRING_P(reg_info[*p]) - ? REG_UNSET(regend[*p]) ? d : regend[*p] - : regend[*p]; - DEBUG_PRINT2(" old_regend: %d\n", - POINTER_TO_OFFSET(old_regend[*p])); - regend[*p] = d; - DEBUG_PRINT2(" regend: %d\n", - POINTER_TO_OFFSET(regend[*p])); +int +#ifdef _LIBC + weak_function +#endif +re_exec(s) +const char *s; +{ + const int len = strlen(s); - /* This register isn't active anymore. */ - IS_ACTIVE(reg_info[*p]) = 0; + return + 0 <= re_search(&re_comp_buf, s, len, 0, len, + (struct re_registers *) 0); +} - /* Clear this whenever we change the register activity status. */ - set_regs_matched_done = 0; +#endif /* _REGEX_RE_COMP */ + +/* POSIX.2 functions. Don't define these for Emacs. */ - /* If this was the only register active, nothing is active - anymore. */ - if (lowest_active_reg == highest_active_reg) { - lowest_active_reg = NO_LOWEST_ACTIVE_REG; - highest_active_reg = NO_HIGHEST_ACTIVE_REG; - } else { /* We must scan for the new highest active register, since - it isn't necessarily one less than now: consider - (a(b)c(d(e)f)g). When group 3 ends, after the f), the - new highest active register is 1. */ - unsigned char r = *p - 1; +#ifndef emacs - while (r > 0 && !IS_ACTIVE(reg_info[r])) - r--; +/* regcomp takes a regular expression as a string and compiles it. - /* If we end up at register zero, that means that we saved - the registers as the result of an `on_failure_jump', not - a `start_memory', and we jumped to past the innermost - `stop_memory'. For example, in ((.)*) we save - registers 1 and 2 as a result of the *, but when we pop - back to the second ), we are at the stop_memory 1. - Thus, nothing is active. */ - if (r == 0) { - lowest_active_reg = NO_LOWEST_ACTIVE_REG; - highest_active_reg = NO_HIGHEST_ACTIVE_REG; - } else - highest_active_reg = r; - } + PREG is a regex_t *. We do not expect any fields to be initialized, + since POSIX says we shouldn't. Thus, we set - /* If just failed to match something this time around with a - group that's operated on by a repetition operator, try to - force exit from the ``loop'', and restore the register - information for this group that we had before trying this - last match. */ - if ((!MATCHED_SOMETHING(reg_info[*p]) - || just_past_start_mem == p - 1) - && (p + 2) < pend) { - boolean is_a_jump_n = false; + `buffer' to the compiled pattern; + `used' to the length of the compiled pattern; + `syntax' to RE_SYNTAX_POSIX_EXTENDED if the + REG_EXTENDED bit in CFLAGS is set; otherwise, to + RE_SYNTAX_POSIX_BASIC; + `newline_anchor' to REG_NEWLINE being set in CFLAGS; + `fastmap' to an allocated space for the fastmap; + `fastmap_accurate' to zero; + `re_nsub' to the number of subexpressions in PATTERN. - p1 = p + 2; - mcnt = 0; - switch ((re_opcode_t) * p1++) { - case jump_n: - is_a_jump_n = true; - case pop_failure_jump: - case maybe_pop_jump: - case jump: - case dummy_failure_jump: - EXTRACT_NUMBER_AND_INCR(mcnt, p1); - if (is_a_jump_n) - p1 += 2; - break; + PATTERN is the address of the pattern string. - default: - /* do nothing */ ; - } - p1 += mcnt; + CFLAGS is a series of bits which affect compilation. - /* If the next operation is a jump backwards in the pattern - to an on_failure_jump right before the start_memory - corresponding to this stop_memory, exit from the loop - by forcing a failure after pushing on the stack the - on_failure_jump's jump in the pattern, and d. */ - if (mcnt < 0 && (re_opcode_t) * p1 == on_failure_jump - && (re_opcode_t) p1[3] == start_memory && p1[4] == *p) { - /* If this group ever matched anything, then restore - what its registers were before trying this last - failed match, e.g., with `(a*)*b' against `ab' for - regstart[1], and, e.g., with `((a*)*(b*)*)*' - against `aba' for regend[3]. + If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we + use POSIX basic syntax. - Also restore the registers for inner groups for, - e.g., `((a*)(b*))*' against `aba' (register 3 would - otherwise get trashed). */ + If REG_NEWLINE is set, then . and [^...] don't match newline. + Also, regexec will try a match beginning after every newline. - if (EVER_MATCHED_SOMETHING(reg_info[*p])) { - unsigned r; + If REG_ICASE is set, then we considers upper- and lowercase + versions of letters to be equivalent when matching. - EVER_MATCHED_SOMETHING(reg_info[*p]) = 0; + If REG_NOSUB is set, then when PREG is passed to regexec, that + routine will report only success or failure, and nothing about the + registers. - /* Restore this and inner groups' (if any) registers. */ - for (r = *p; - r < (unsigned) *p + (unsigned) *(p + 1); r++) { - regstart[r] = old_regstart[r]; + It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for + the return codes and their meanings.) */ - /* xx why this test? */ - if (old_regend[r] >= regstart[r]) - regend[r] = old_regend[r]; - } - } - p1++; - EXTRACT_NUMBER_AND_INCR(mcnt, p1); - PUSH_FAILURE_POINT(p1 + mcnt, d, -2); +int regcomp(preg, pattern, cflags) +regex_t *preg; +const char *pattern; +int cflags; +{ + reg_errcode_t ret; + reg_syntax_t syntax + = (cflags & REG_EXTENDED) ? - goto fail; - } - } + RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; - /* Move past the register number and the inner group count. */ - p += 2; - break; + /* regex_compile will allocate the space for the compiled pattern. */ + preg->buffer = 0; + preg->allocated = 0; + preg->used = 0; + /* Try to allocate space for the fastmap. */ + preg->fastmap = (char *) malloc(1 << BYTEWIDTH); - /* \ has been turned into a `duplicate' command which is - followed by the numeric value of as the register number. */ - case duplicate: - { - register const char *d2, *dend2; - int regno = *p++; /* Get which register to match against. */ + if (cflags & REG_ICASE) { + unsigned i; - DEBUG_PRINT2("EXECUTING duplicate %d.\n", regno); + preg->translate + = (RE_TRANSLATE_TYPE) malloc(CHAR_SET_SIZE + * sizeof(*(RE_TRANSLATE_TYPE) 0)); + if (preg->translate == NULL) + return (int) REG_ESPACE; - /* Can't back reference a group which we've never matched. */ - if (REG_UNSET(regstart[regno]) || REG_UNSET(regend[regno])) - goto fail; + /* Map uppercase characters to corresponding lowercase ones. */ + for (i = 0; i < CHAR_SET_SIZE; i++) + preg->translate[i] = ISUPPER(i) ? TOLOWER(i) : i; + } else + preg->translate = NULL; - /* Where in input to try to start matching. */ - d2 = regstart[regno]; + /* If REG_NEWLINE is set, newlines are treated differently. */ + if (cflags & REG_NEWLINE) { /* REG_NEWLINE implies neither . nor [^...] match newline. */ + syntax &= ~RE_DOT_NEWLINE; + syntax |= RE_HAT_LISTS_NOT_NEWLINE; + /* It also changes the matching behavior. */ + preg->newline_anchor = 1; + } else + preg->newline_anchor = 0; - /* Where to stop matching; if both the place to start and - the place to stop matching are in the same string, then - set to the place to stop, otherwise, for now have to use - the end of the first string. */ + preg->no_sub = !!(cflags & REG_NOSUB); - dend2 = ((FIRST_STRING_P(regstart[regno]) - == FIRST_STRING_P(regend[regno])) - ? regend[regno] : end_match_1); - for (;;) { - /* If necessary, advance to next segment in register - contents. */ - while (d2 == dend2) { - if (dend2 == end_match_2) - break; - if (dend2 == regend[regno]) - break; + /* POSIX says a null character in the pattern terminates it, so we + can use strlen here in compiling the pattern. */ + ret = regex_compile(pattern, strlen(pattern), syntax, preg); - /* End of string1 => advance to string2. */ - d2 = string2; - dend2 = regend[regno]; - } - /* At end of register contents => success */ - if (d2 == dend2) - break; + /* POSIX doesn't distinguish between an unmatched open-group and an + unmatched close-group: both are REG_EPAREN. */ + if (ret == REG_ERPAREN) + ret = REG_EPAREN; - /* If necessary, advance to next segment in data. */ - PREFETCH(); + if (ret == REG_NOERROR && preg->fastmap) { + /* Compute the fastmap now, since regexec cannot modify the pattern + buffer. */ + if (re_compile_fastmap(preg) == -2) { + /* Some error occurred while computing the fastmap, just forget + about it. */ + free(preg->fastmap); + preg->fastmap = NULL; + } + } + + return (int) ret; +} - /* How many characters left in this segment to match. */ - mcnt = dend - d; +#ifdef _LIBC +weak_alias(__regcomp, regcomp) +#endif +/* regexec searches for a given pattern, specified by PREG, in the + string STRING. - /* Want how many consecutive characters we can match in - one shot, so, if necessary, adjust the count. */ - if (mcnt > dend2 - d2) - mcnt = dend2 - d2; + If NMATCH is zero or REG_NOSUB was set in the cflags argument to + `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at + least NMATCH elements, and we set them to the offsets of the + corresponding matched substrings. - /* Compare that many; failure if mismatch, else move - past them. */ - if (translate ? bcmp_translate(d, d2, mcnt, translate) - : memcmp(d, d2, mcnt)) - goto fail; - d += mcnt, d2 += mcnt; + EFLAGS specifies `execution flags' which affect matching: if + REG_NOTBOL is set, then ^ does not match at the beginning of the + string; if REG_NOTEOL is set, then $ does not match at the end. - /* Do this because we've match some characters. */ - SET_REGS_MATCHED(); - } - } - break; + We return 0 if we find a match and REG_NOMATCH if not. */ +int regexec(preg, string, nmatch, pmatch, eflags) +const regex_t *preg; +const char *string; +size_t nmatch; +regmatch_t pmatch[]; +int eflags; +{ + int ret; + struct re_registers regs; + regex_t private_preg; + int len = strlen(string); + boolean want_reg_info = !preg->no_sub && nmatch > 0; + private_preg = *preg; - /* begline matches the empty string at the beginning of the string - (unless `not_bol' is set in `bufp'), and, if - `newline_anchor' is set, after newlines. */ - case begline: - DEBUG_PRINT1("EXECUTING begline.\n"); + private_preg.not_bol = !!(eflags & REG_NOTBOL); + private_preg.not_eol = !!(eflags & REG_NOTEOL); - if (AT_STRINGS_BEG(d)) { - if (!bufp->not_bol) - break; - } else if (d[-1] == '\n' && bufp->newline_anchor) { - break; - } - /* In all other cases, we fail. */ - goto fail; + /* The user has told us exactly how many registers to return + information about, via `nmatch'. We have to pass that on to the + matching routines. */ + private_preg.regs_allocated = REGS_FIXED; + if (want_reg_info) { + regs.num_regs = nmatch; + regs.start = TALLOC(nmatch * 2, regoff_t); + if (regs.start == NULL) + return (int) REG_NOMATCH; + regs.end = regs.start + nmatch; + } - /* endline is the dual of begline. */ - case endline: - DEBUG_PRINT1("EXECUTING endline.\n"); + /* Perform the searching operation. */ + ret = re_search(&private_preg, string, len, + /* start: */ 0, /* range: */ len, + want_reg_info ? ®s : (struct re_registers *) 0); - if (AT_STRINGS_END(d)) { - if (!bufp->not_eol) - break; - } + /* Copy the register information to the POSIX structure. */ + if (want_reg_info) { + if (ret >= 0) { + unsigned r; - /* We have to ``prefetch'' the next character. */ - else if ((d == end1 ? *string2 : *d) == '\n' - && bufp->newline_anchor) { - break; + for (r = 0; r < nmatch; r++) { + pmatch[r].rm_so = regs.start[r]; + pmatch[r].rm_eo = regs.end[r]; } - goto fail; - + } - /* Match at the very beginning of the data. */ - case begbuf: - DEBUG_PRINT1("EXECUTING begbuf.\n"); - if (AT_STRINGS_BEG(d)) - break; - goto fail; + /* If we needed the temporary register info, free the space now. */ + free(regs.start); + } + /* We want zero return to mean success, unlike `re_search'. */ + return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; +} - /* Match at the very end of the data. */ - case endbuf: - DEBUG_PRINT1("EXECUTING endbuf.\n"); - if (AT_STRINGS_END(d)) - break; - goto fail; +#ifdef _LIBC +weak_alias(__regexec, regexec) +#endif +/* Returns a message corresponding to an error code, ERRCODE, returned + from either regcomp or regexec. We don't use PREG here. */ + size_t regerror(errcode, preg, errbuf, errbuf_size) +int errcode; +const regex_t *preg; +char *errbuf; +size_t errbuf_size; +{ + const char *msg; + size_t msg_size; + if (errcode < 0 || errcode >= (int) (sizeof(re_error_msgid_idx) + / sizeof(re_error_msgid_idx[0]))) + /* Only error codes returned by the rest of the code should be passed + to this routine. If we are given anything else, or if other regex + code generates an invalid error code, then the program has a bug. + Dump core so we can fix it. */ + abort(); - /* on_failure_keep_string_jump is used to optimize `.*\n'. It - pushes NULL as the value for the string on the stack. Then - `pop_failure_point' will keep the current value for the - string, instead of restoring it. To see why, consider - matching `foo\nbar' against `.*\n'. The .* matches the foo; - then the . fails against the \n. But the next thing we want - to do is match the \n against the \n; if we restored the - string value, we would be back at the foo. + msg = gettext(re_error_msgid + re_error_msgid_idx[errcode]); - Because this is used only in specific cases, we don't need to - check all the things that `on_failure_jump' does, to make - sure the right things get saved on the stack. Hence we don't - share its code. The only reason to push anything on the - stack at all is that otherwise we would have to change - `anychar's code to do something besides goto fail in this - case; that seems worse than this. */ - case on_failure_keep_string_jump: - DEBUG_PRINT1("EXECUTING on_failure_keep_string_jump"); + msg_size = strlen(msg) + 1; /* Includes the null. */ - EXTRACT_NUMBER_AND_INCR(mcnt, p); -#ifdef _LIBC - DEBUG_PRINT3(" %d (to %p):\n", mcnt, p + mcnt); + if (errbuf_size != 0) { + if (msg_size > errbuf_size) { +#if defined HAVE_MEMPCPY || defined _LIBC + *((char *) __mempcpy(errbuf, msg, errbuf_size - 1)) = '\0'; #else - DEBUG_PRINT3(" %d (to 0x%x):\n", mcnt, p + mcnt); + memcpy(errbuf, msg, errbuf_size - 1); + errbuf[errbuf_size - 1] = 0; #endif + } else + memcpy(errbuf, msg, msg_size); + } - PUSH_FAILURE_POINT(p + mcnt, NULL, -2); - break; + return msg_size; +} +#ifdef _LIBC +weak_alias(__regerror, regerror) +#endif +/* Free dynamically allocated space used by PREG. */ +void regfree(preg) +regex_t *preg; +{ + if (preg->buffer != NULL) + free(preg->buffer); + preg->buffer = NULL; - /* Uses of on_failure_jump: + preg->allocated = 0; + preg->used = 0; - Each alternative starts with an on_failure_jump that points - to the beginning of the next alternative. Each alternative - except the last ends with a jump that in effect jumps past - the rest of the alternatives. (They really jump to the - ending jump of the following alternative, because tensioning - these jumps is a hassle.) + if (preg->fastmap != NULL) + free(preg->fastmap); + preg->fastmap = NULL; + preg->fastmap_accurate = 0; - Repeats start with an on_failure_jump that points past both - the repetition text and either the following jump or - pop_failure_jump back to this on_failure_jump. */ - case on_failure_jump: - on_failure: - DEBUG_PRINT1("EXECUTING on_failure_jump"); + if (preg->translate != NULL) + free(preg->translate); + preg->translate = NULL; +} - EXTRACT_NUMBER_AND_INCR(mcnt, p); #ifdef _LIBC - DEBUG_PRINT3(" %d (to %p)", mcnt, p + mcnt); -#else - DEBUG_PRINT3(" %d (to 0x%x)", mcnt, p + mcnt); +weak_alias(__regfree, regfree) #endif +#endif /* not emacs */ + +/* `regex_compile' compiles PATTERN (of length SIZE) according to SYNTAX. + Returns one of error codes defined in `regex.h', or zero for success. - /* If this on_failure_jump comes right before a group (i.e., - the original * applied to a group), save the information - for that group and all inner ones, so that if we fail back - to this point, the group's information will be correct. - For example, in \(a*\)*\1, we need the preceding group, - and in \(zz\(a*\)b*\)\2, we need the inner group. */ + Assumes the `allocated' (and perhaps `buffer') and `translate' + fields are set in BUFP on entry. - /* We can't use `p' to check ahead because we push - a failure point to `p + mcnt' after we do this. */ - p1 = p; + If it succeeds, results are put in BUFP (if it returns an error, the + contents of BUFP are undefined): + `buffer' is the compiled pattern; + `syntax' is set to SYNTAX; + `used' is set to the length of the compiled pattern; + `fastmap_accurate' is zero; + `re_nsub' is the number of subexpressions in PATTERN; + `not_bol' and `not_eol' are zero; - /* We need to skip no_op's before we look for the - start_memory in case this on_failure_jump is happening as - the result of a completed succeed_n, as in \(a\)\{1,3\}b\1 - against aba. */ - while (p1 < pend && (re_opcode_t) * p1 == no_op) - p1++; + The `fastmap' and `newline_anchor' fields are neither + examined nor set. */ - if (p1 < pend && (re_opcode_t) * p1 == start_memory) { - /* We have a new highest active register now. This will - get reset at the start_memory we are about to get to, - but we will have saved all the registers relevant to - this repetition op, as described above. */ - highest_active_reg = *(p1 + 1) + *(p1 + 2); - if (lowest_active_reg == NO_LOWEST_ACTIVE_REG) - lowest_active_reg = *(p1 + 1); - } +/* Return, freeing storage we allocated. */ +#define FREE_STACK_RETURN(value) \ + return (free (compile_stack.stack), value) - DEBUG_PRINT1(":\n"); - PUSH_FAILURE_POINT(p + mcnt, d, -2); - break; +static reg_errcode_t regex_compile(pattern, size, syntax, bufp) +const char *pattern; +size_t size; +reg_syntax_t syntax; +struct re_pattern_buffer *bufp; +{ + /* We fetch characters from PATTERN here. Even though PATTERN is + `char *' (i.e., signed), we declare these variables as unsigned, so + they can be reliably used as array indices. */ + register unsigned char c, c1; + /* A random temporary spot in PATTERN. */ + const char *p1; - /* A smart repeat ends with `maybe_pop_jump'. - We change it to either `pop_failure_jump' or `jump'. */ - case maybe_pop_jump: - EXTRACT_NUMBER_AND_INCR(mcnt, p); - DEBUG_PRINT2("EXECUTING maybe_pop_jump %d.\n", mcnt); - { - register unsigned char *p2 = p; + /* Points to the end of the buffer, where we should append. */ + register unsigned char *b; - /* Compare the beginning of the repeat with what in the - pattern follows its end. If we can establish that there - is nothing that they would both match, i.e., that we - would have to backtrack because of (as in, e.g., `a*a') - then we can change to pop_failure_jump, because we'll - never have to backtrack. + /* Keeps track of unclosed groups. */ + compile_stack_type compile_stack; - This is not true in the case of alternatives: in - `(a|ab)*' we do need to backtrack to the `ab' alternative - (e.g., if the string was `ab'). But instead of trying to - detect that here, the alternative has put on a dummy - failure point which is what we will end up popping. */ + /* Points to the current (ending) position in the pattern. */ + const char *p = pattern; + const char *pend = pattern + size; - /* Skip over open/close-group commands. - If what follows this loop is a ...+ construct, - look at what begins its body, since we will have to - match at least one of that. */ - while (1) { - if (p2 + 2 < pend - && ((re_opcode_t) * p2 == stop_memory - || (re_opcode_t) * p2 == start_memory)) - p2 += 3; - else if (p2 + 6 < pend - && (re_opcode_t) * p2 == dummy_failure_jump) - p2 += 6; - else - break; - } + /* How to translate the characters in the pattern. */ + RE_TRANSLATE_TYPE translate = bufp->translate; - p1 = p + mcnt; - /* p1[0] ... p1[2] are the `on_failure_jump' corresponding - to the `maybe_finalize_jump' of this case. Examine what - follows. */ + /* Address of the count-byte of the most recently inserted `exactn' + command. This makes it possible to tell if a new exact-match + character can be added to that command or if the character requires + a new `exactn' command. */ + unsigned char *pending_exact = 0; - /* If we're at the end of the pattern, we can change. */ - if (p2 == pend) { - /* Consider what happens when matching ":\(.*\)" - against ":/". I don't really understand this code - yet. */ - p[-3] = (unsigned char) pop_failure_jump; - DEBUG_PRINT1 - (" End of pattern: change to `pop_failure_jump'.\n"); - } + /* Address of start of the most recently finished expression. + This tells, e.g., postfix * where to find the start of its + operand. Reset at the beginning of groups and alternatives. */ + unsigned char *laststart = 0; - else if ((re_opcode_t) * p2 == exactn - || (bufp->newline_anchor - && (re_opcode_t) * p2 == endline)) { - register unsigned char c = - *p2 == (unsigned char) endline ? '\n' : p2[2]; + /* Address of beginning of regexp, or inside of last group. */ + unsigned char *begalt; - if ((re_opcode_t) p1[3] == exactn && p1[5] != c) { - p[-3] = (unsigned char) pop_failure_jump; - DEBUG_PRINT3(" %c != %c => pop_failure_jump.\n", - c, p1[5]); - } + /* Place in the uncompiled pattern (i.e., the {) to + which to go back if the interval is invalid. */ + const char *beg_interval; - else if ((re_opcode_t) p1[3] == charset - || (re_opcode_t) p1[3] == charset_not) { - int not = (re_opcode_t) p1[3] == charset_not; + /* Address of the place where a forward jump should go to the end of + the containing expression. Each alternative of an `or' -- except the + last -- ends with a forward jump of this sort. */ + unsigned char *fixup_alt_jump = 0; - if (c < (unsigned char) (p1[4] * BYTEWIDTH) - && p1[5 + - c / BYTEWIDTH] & (1 << (c % - BYTEWIDTH))) not - = !not; + /* Counts open-groups as they are encountered. Remembered for the + matching close-group on the compile stack, so the same register + number is put in the stop_memory as the start_memory. */ + regnum_t regnum = 0; - /* `not' is equal to 1 if c would match, which means - that we can't change to pop_failure_jump. */ - if (!not) { - p[-3] = (unsigned char) pop_failure_jump; - DEBUG_PRINT1 - (" No match => pop_failure_jump.\n"); - } - } - } else if ((re_opcode_t) * p2 == charset) { - /* We win if the first character of the loop is not part - of the charset. */ - if ((re_opcode_t) p1[3] == exactn - && !((int) p2[1] * BYTEWIDTH > (int) p1[5] - && (p2[2 + p1[5] / BYTEWIDTH] - & (1 << (p1[5] % BYTEWIDTH))))) { - p[-3] = (unsigned char) pop_failure_jump; - DEBUG_PRINT1(" No match => pop_failure_jump.\n"); - } +#ifdef DEBUG + DEBUG_PRINT1("\nCompiling pattern: "); + if (debug) { + unsigned debug_count; - else if ((re_opcode_t) p1[3] == charset_not) { - int idx; + for (debug_count = 0; debug_count < size; debug_count++) + putchar(pattern[debug_count]); + putchar('\n'); + } +#endif /* DEBUG */ - /* We win if the charset_not inside the loop - lists every character listed in the charset after. */ - for (idx = 0; idx < (int) p2[1]; idx++) - if (!(p2[2 + idx] == 0 || (idx < (int) p1[4] - && - ((p2 - [2 + - idx] & ~p1[5 + - idx]) - == 0)))) - break; + /* Initialize the compile stack. */ + compile_stack.stack = + TALLOC(INIT_COMPILE_STACK_SIZE, compile_stack_elt_t); + if (compile_stack.stack == NULL) + return REG_ESPACE; - if (idx == p2[1]) { - p[-3] = (unsigned char) pop_failure_jump; - DEBUG_PRINT1 - (" No match => pop_failure_jump.\n"); - } - } else if ((re_opcode_t) p1[3] == charset) { - int idx; + compile_stack.size = INIT_COMPILE_STACK_SIZE; + compile_stack.avail = 0; - /* We win if the charset inside the loop - has no overlap with the one after the loop. */ - for (idx = 0; - idx < (int) p2[1] && idx < (int) p1[4]; idx++) - if ((p2[2 + idx] & p1[5 + idx]) != 0) - break; + /* Initialize the pattern buffer. */ + bufp->syntax = syntax; + bufp->fastmap_accurate = 0; + bufp->not_bol = bufp->not_eol = 0; - if (idx == p2[1] || idx == p1[4]) { - p[-3] = (unsigned char) pop_failure_jump; - DEBUG_PRINT1 - (" No match => pop_failure_jump.\n"); - } - } - } - } - p -= 2; /* Point at relative address again. */ - if ((re_opcode_t) p[-1] != pop_failure_jump) { - p[-1] = (unsigned char) jump; - DEBUG_PRINT1(" Match => jump.\n"); - goto unconditional_jump; - } - /* Note fall through. */ + /* Set `used' to zero, so that if we return an error, the pattern + printer (for debugging) will think there's no pattern. We reset it + at the end. */ + bufp->used = 0; + /* Always count groups, whether or not bufp->no_sub is set. */ + bufp->re_nsub = 0; - /* The end of a simple repeat has a pop_failure_jump back to - its matching on_failure_jump, where the latter will push a - failure point. The pop_failure_jump takes off failure - points put on by this pop_failure_jump's matching - on_failure_jump; we got through the pattern to here from the - matching on_failure_jump, so didn't fail. */ - case pop_failure_jump: - { - /* We need to pass separate storage for the lowest and - highest registers, even though we don't care about the - actual values. Otherwise, we will restore only one - register from the stack, since lowest will == highest in - `pop_failure_point'. */ - active_reg_t dummy_low_reg, dummy_high_reg; - unsigned char *pdummy; - const char *sdummy; +#if !defined emacs && !defined SYNTAX_TABLE + /* Initialize the syntax table. */ + init_syntax_once(); +#endif - DEBUG_PRINT1("EXECUTING pop_failure_jump.\n"); - POP_FAILURE_POINT(sdummy, pdummy, - dummy_low_reg, dummy_high_reg, - reg_dummy, reg_dummy, reg_info_dummy); + if (bufp->allocated == 0) { + if (bufp->buffer) { /* If zero allocated, but buffer is non-null, try to realloc + enough space. This loses if buffer's address is bogus, but + that is the user's responsibility. */ + RETALLOC(bufp->buffer, INIT_BUF_SIZE, unsigned char); + } else { /* Caller did not allocate a buffer. Do it for them. */ + bufp->buffer = TALLOC(INIT_BUF_SIZE, unsigned char); } - /* Note fall through. */ + if (!bufp->buffer) + FREE_STACK_RETURN(REG_ESPACE); - unconditional_jump: -#ifdef _LIBC - DEBUG_PRINT2("\n%p: ", p); -#else - DEBUG_PRINT2("\n0x%x: ", p); -#endif - /* Note fall through. */ + bufp->allocated = INIT_BUF_SIZE; + } - /* Unconditionally jump (without popping any failure points). */ - case jump: - EXTRACT_NUMBER_AND_INCR(mcnt, p); /* Get the amount to jump. */ - DEBUG_PRINT2("EXECUTING jump %d ", mcnt); - p += mcnt; /* Do the jump. */ -#ifdef _LIBC - DEBUG_PRINT2("(to %p).\n", p); -#else - DEBUG_PRINT2("(to 0x%x).\n", p); -#endif + begalt = b = bufp->buffer; + + /* Loop through the uncompiled pattern until we're at the end. */ + while (p != pend) { + PATFETCH(c); + + switch (c) { + case '^': + { + if ( /* If at start of pattern, it's an operator. */ + p == pattern + 1 + /* If context independent, it's an operator. */ + || syntax & RE_CONTEXT_INDEP_ANCHORS + /* Otherwise, depends on what's come before. */ + || at_begline_loc_p(pattern, p, syntax)) + BUF_PUSH(begline); + else + goto normal_char; + } break; - /* We need this opcode so we can detect where alternatives end - in `group_match_null_string_p' et al. */ - case jump_past_alt: - DEBUG_PRINT1("EXECUTING jump_past_alt.\n"); - goto unconditional_jump; + case '$': + { + if ( /* If at end of pattern, it's an operator. */ + p == pend + /* If context independent, it's an operator. */ + || syntax & RE_CONTEXT_INDEP_ANCHORS + /* Otherwise, depends on what's next. */ + || at_endline_loc_p(p, pend, syntax)) + BUF_PUSH(endline); + else + goto normal_char; + } + break; - /* Normally, the on_failure_jump pushes a failure point, which - then gets popped at pop_failure_jump. We will end up at - pop_failure_jump, also, and with a pattern of, say, `a+', we - are skipping over the on_failure_jump, so we have to push - something meaningless for pop_failure_jump to pop. */ - case dummy_failure_jump: - DEBUG_PRINT1("EXECUTING dummy_failure_jump.\n"); - /* It doesn't matter what we push for the string here. What - the code at `fail' tests is the value for the pattern. */ - PUSH_FAILURE_POINT(NULL, NULL, -2); - goto unconditional_jump; + case '+': + case '?': + if ((syntax & RE_BK_PLUS_QM) + || (syntax & RE_LIMITED_OPS)) + goto normal_char; + handle_plus: + case '*': + /* If there is no previous pattern... */ + if (!laststart) { + if (syntax & RE_CONTEXT_INVALID_OPS) + FREE_STACK_RETURN(REG_BADRPT); + else if (!(syntax & RE_CONTEXT_INDEP_OPS)) + goto normal_char; + } + { + /* Are we optimizing this jump? */ + boolean keep_string_p = false; - /* At the end of an alternative, we need to push a dummy failure - point in case we are followed by a `pop_failure_jump', because - we don't want the failure point for the alternative to be - popped. For example, matching `(a|ab)*' against `aab' - requires that we match the `ab' alternative. */ - case push_dummy_failure: - DEBUG_PRINT1("EXECUTING push_dummy_failure.\n"); - /* See comments just above at `dummy_failure_jump' about the - two zeroes. */ - PUSH_FAILURE_POINT(NULL, NULL, -2); - break; + /* 1 means zero (many) matches is allowed. */ + char zero_times_ok = 0, many_times_ok = 0; - /* Have to succeed matching what follows at least n times. - After that, handle like `on_failure_jump'. */ - case succeed_n: - EXTRACT_NUMBER(mcnt, p + 2); - DEBUG_PRINT2("EXECUTING succeed_n %d.\n", mcnt); + /* If there is a sequence of repetition chars, collapse it + down to just one (the right one). We can't combine + interval operators with these because of, e.g., `a{2}*', + which should only match an even number of `a's. */ - assert(mcnt >= 0); - /* Originally, this is how many times we HAVE to succeed. */ - if (mcnt > 0) { - mcnt--; - p += 2; - STORE_NUMBER_AND_INCR(p, mcnt); -#ifdef _LIBC - DEBUG_PRINT3(" Setting %p to %d.\n", p - 2, mcnt); -#else - DEBUG_PRINT3(" Setting 0x%x to %d.\n", p - 2, mcnt); -#endif - } else if (mcnt == 0) { -#ifdef _LIBC - DEBUG_PRINT2(" Setting two bytes from %p to no_op.\n", - p + 2); -#else - DEBUG_PRINT2(" Setting two bytes from 0x%x to no_op.\n", - p + 2); -#endif - p[2] = (unsigned char) no_op; - p[3] = (unsigned char) no_op; - goto on_failure; - } - break; + for (;;) { + zero_times_ok |= c != '+'; + many_times_ok |= c != '?'; - case jump_n: - EXTRACT_NUMBER(mcnt, p + 2); - DEBUG_PRINT2("EXECUTING jump_n %d.\n", mcnt); + if (p == pend) + break; - /* Originally, this is how many times we CAN jump. */ - if (mcnt) { - mcnt--; - STORE_NUMBER(p + 2, mcnt); -#ifdef _LIBC - DEBUG_PRINT3(" Setting %p to %d.\n", p + 2, mcnt); -#else - DEBUG_PRINT3(" Setting 0x%x to %d.\n", p + 2, mcnt); -#endif - goto unconditional_jump; - } - /* If don't have to jump any more, skip over the rest of command. */ - else - p += 4; - break; + PATFETCH(c); - case set_number_at: - { - DEBUG_PRINT1("EXECUTING set_number_at.\n"); + if (c == '*' + || (!(syntax & RE_BK_PLUS_QM) + && (c == '+' || c == '?'))); - EXTRACT_NUMBER_AND_INCR(mcnt, p); - p1 = p + mcnt; - EXTRACT_NUMBER_AND_INCR(mcnt, p); -#ifdef _LIBC - DEBUG_PRINT3(" Setting %p to %d.\n", p1, mcnt); -#else - DEBUG_PRINT3(" Setting 0x%x to %d.\n", p1, mcnt); -#endif - STORE_NUMBER(p1, mcnt); - break; - } + else if (syntax & RE_BK_PLUS_QM && c == '\\') { + if (p == pend) + FREE_STACK_RETURN(REG_EESCAPE); -#if 0 - /* The DEC Alpha C compiler 3.x generates incorrect code for the - test WORDCHAR_P (d - 1) != WORDCHAR_P (d) in the expansion of - AT_WORD_BOUNDARY, so this code is disabled. Expanding the - macro and introducing temporary variables works around the bug. */ + PATFETCH(c1); + if (!(c1 == '+' || c1 == '?')) { + PATUNFETCH; + PATUNFETCH; + break; + } - case wordbound: - DEBUG_PRINT1("EXECUTING wordbound.\n"); - if (AT_WORD_BOUNDARY(d)) - break; - goto fail; + c = c1; + } else { + PATUNFETCH; + break; + } - case notwordbound: - DEBUG_PRINT1("EXECUTING notwordbound.\n"); - if (AT_WORD_BOUNDARY(d)) - goto fail; - break; -#else - case wordbound: - { - boolean prevchar, thischar; + /* If we get here, we found another repeat character. */ + } - DEBUG_PRINT1("EXECUTING wordbound.\n"); - if (AT_STRINGS_BEG(d) || AT_STRINGS_END(d)) - break; + /* Star, etc. applied to an empty pattern is equivalent + to an empty pattern. */ + if (!laststart) + break; - prevchar = WORDCHAR_P(d - 1); - thischar = WORDCHAR_P(d); - if (prevchar != thischar) - break; - goto fail; - } + /* Now we know whether or not zero matches is allowed + and also whether or not two or more matches is allowed. */ + if (many_times_ok) { /* More than one repetition is allowed, so put in at the + end a backward relative jump from `b' to before the next + jump we're going to put in below (which jumps from + laststart to after this jump). - case notwordbound: - { - boolean prevchar, thischar; + But if we are at the `*' in the exact sequence `.*\n', + insert an unconditional jump backwards to the ., + instead of the beginning of the loop. This way we only + push a failure point once, instead of every time + through the loop. */ + assert(p - 1 > pattern); - DEBUG_PRINT1("EXECUTING notwordbound.\n"); - if (AT_STRINGS_BEG(d) || AT_STRINGS_END(d)) - goto fail; + /* Allocate the space for the jump. */ + GET_BUFFER_SPACE(3); - prevchar = WORDCHAR_P(d - 1); - thischar = WORDCHAR_P(d); - if (prevchar != thischar) - goto fail; - break; - } -#endif + /* We know we are not at the first character of the pattern, + because laststart was nonzero. And we've already + incremented `p', by the way, to be the character after + the `*'. Do we have to do something analogous here + for null bytes, because of RE_DOT_NOT_NULL? */ + if (TRANSLATE(*(p - 2)) == TRANSLATE('.') + && zero_times_ok + && p < pend && TRANSLATE(*p) == TRANSLATE('\n') + && !(syntax & RE_DOT_NEWLINE)) { /* We have .*\n. */ + STORE_JUMP(jump, b, laststart); + keep_string_p = true; + } else + /* Anything else. */ + STORE_JUMP(maybe_pop_jump, b, laststart - 3); - case wordbeg: - DEBUG_PRINT1("EXECUTING wordbeg.\n"); - if (WORDCHAR_P(d) && (AT_STRINGS_BEG(d) || !WORDCHAR_P(d - 1))) - break; - goto fail; + /* We've added more stuff to the buffer. */ + b += 3; + } - case wordend: - DEBUG_PRINT1("EXECUTING wordend.\n"); - if (!AT_STRINGS_BEG(d) && WORDCHAR_P(d - 1) - && (!WORDCHAR_P(d) || AT_STRINGS_END(d))) - break; - goto fail; + /* On failure, jump from laststart to b + 3, which will be the + end of the buffer after this jump is inserted. */ + GET_BUFFER_SPACE(3); + INSERT_JUMP(keep_string_p ? on_failure_keep_string_jump + : on_failure_jump, laststart, b + 3); + pending_exact = 0; + b += 3; -#ifdef emacs - case before_dot: - DEBUG_PRINT1("EXECUTING before_dot.\n"); - if (PTR_CHAR_POS((unsigned char *) d) >= point) - goto fail; + if (!zero_times_ok) { + /* At least one repetition is required, so insert a + `dummy_failure_jump' before the initial + `on_failure_jump' instruction of the loop. This + effects a skip over that instruction the first time + we hit that loop. */ + GET_BUFFER_SPACE(3); + INSERT_JUMP(dummy_failure_jump, laststart, + laststart + 6); + b += 3; + } + } break; - case at_dot: - DEBUG_PRINT1("EXECUTING at_dot.\n"); - if (PTR_CHAR_POS((unsigned char *) d) != point) - goto fail; - break; - case after_dot: - DEBUG_PRINT1("EXECUTING after_dot.\n"); - if (PTR_CHAR_POS((unsigned char *) d) <= point) - goto fail; + case '.': + laststart = b; + BUF_PUSH(anychar); break; - case syntaxspec: - DEBUG_PRINT2("EXECUTING syntaxspec %d.\n", mcnt); - mcnt = *p++; - goto matchsyntax; - case wordchar: - DEBUG_PRINT1("EXECUTING Emacs wordchar.\n"); - mcnt = (int) Sword; - matchsyntax: - PREFETCH(); - /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ - d++; - if (SYNTAX(d[-1]) != (enum syntaxcode) mcnt) - goto fail; - SET_REGS_MATCHED(); - break; + case '[': + { + boolean had_char_class = false; - case notsyntaxspec: - DEBUG_PRINT2("EXECUTING notsyntaxspec %d.\n", mcnt); - mcnt = *p++; - goto matchnotsyntax; + if (p == pend) + FREE_STACK_RETURN(REG_EBRACK); - case notwordchar: - DEBUG_PRINT1("EXECUTING Emacs notwordchar.\n"); - mcnt = (int) Sword; - matchnotsyntax: - PREFETCH(); - /* Can't use *d++ here; SYNTAX may be an unsafe macro. */ - d++; - if (SYNTAX(d[-1]) == (enum syntaxcode) mcnt) - goto fail; - SET_REGS_MATCHED(); - break; + /* Ensure that we have enough space to push a charset: the + opcode, the length count, and the bitset; 34 bytes in all. */ + GET_BUFFER_SPACE(34); -#else /* not emacs */ - case wordchar: - DEBUG_PRINT1("EXECUTING non-Emacs wordchar.\n"); - PREFETCH(); - if (!WORDCHAR_P(d)) - goto fail; - SET_REGS_MATCHED(); - d++; - break; + laststart = b; - case notwordchar: - DEBUG_PRINT1("EXECUTING non-Emacs notwordchar.\n"); - PREFETCH(); - if (WORDCHAR_P(d)) - goto fail; - SET_REGS_MATCHED(); - d++; - break; -#endif /* not emacs */ + /* We test `*p == '^' twice, instead of using an if + statement, so we only need one BUF_PUSH. */ + BUF_PUSH(*p == '^' ? charset_not : charset); + if (*p == '^') + p++; - default: - abort(); - } - continue; /* Successfully executed one pattern command; keep going. */ + /* Remember the first position in the bracket expression. */ + p1 = p; + /* Push the number of bytes in the bitmap. */ + BUF_PUSH((1 << BYTEWIDTH) / BYTEWIDTH); - /* We goto here if a matching operation fails. */ - fail: - if (!FAIL_STACK_EMPTY()) { /* A restart point is known. Restore to that state. */ - DEBUG_PRINT1("\nFAIL:\n"); - POP_FAILURE_POINT(d, p, - lowest_active_reg, highest_active_reg, - regstart, regend, reg_info); + /* Clear the whole map. */ + bzero(b, (1 << BYTEWIDTH) / BYTEWIDTH); - /* If this failure point is a dummy, try the next one. */ - if (!p) - goto fail; + /* charset_not matches newline according to a syntax bit. */ + if ((re_opcode_t) b[-2] == charset_not + && (syntax & RE_HAT_LISTS_NOT_NEWLINE)) SET_LIST_BIT('\n'); - /* If we failed to the end of the pattern, don't examine *p. */ - assert(p <= pend); - if (p < pend) { - boolean is_a_jump_n = false; + /* Read in characters and ranges, setting map bits. */ + for (;;) { + if (p == pend) + FREE_STACK_RETURN(REG_EBRACK); - /* If failed to a backwards jump that's part of a repetition - loop, need to pop this failure point and use the next one. */ - switch ((re_opcode_t) * p) { - case jump_n: - is_a_jump_n = true; - case maybe_pop_jump: - case pop_failure_jump: - case jump: - p1 = p + 1; - EXTRACT_NUMBER_AND_INCR(mcnt, p1); - p1 += mcnt; + PATFETCH(c); - if ((is_a_jump_n && (re_opcode_t) * p1 == succeed_n) - || (!is_a_jump_n - && (re_opcode_t) * p1 == on_failure_jump)) - goto fail; - break; - default: - /* do nothing */ ; - } - } + /* \ might escape characters inside [...] and [^...]. */ + if ((syntax & RE_BACKSLASH_ESCAPE_IN_LISTS) && c == '\\') { + if (p == pend) + FREE_STACK_RETURN(REG_EESCAPE); - if (d >= string1 && d <= end1) - dend = end_match_1; - } else - break; /* Matching at this starting point really fails. */ - } /* for (;;) */ + PATFETCH(c1); + SET_LIST_BIT(c1); + continue; + } - if (best_regs_set) - goto restore_best_regs; + /* Could be the end of the bracket expression. If it's + not (i.e., when the bracket expression is `[]' so + far), the ']' character bit gets set way below. */ + if (c == ']' && p != p1 + 1) + break; - FREE_VARIABLES(); + /* Look ahead to see if it's a range when the last thing + was a character class. */ + if (had_char_class && c == '-' && *p != ']') + FREE_STACK_RETURN(REG_ERANGE); - return -1; /* Failure to match. */ -} /* re_match_2 */ - -/* Subroutine definitions for re_match_2. */ + /* Look ahead to see if it's a range when the last thing + was a character: if this is a hyphen not at the + beginning or the end of a list, then it's the range + operator. */ + if (c == '-' && !(p - 2 >= pattern && p[-2] == '[') + && !(p - 3 >= pattern && p[-3] == '[' && p[-2] == '^') + && *p != ']') { + reg_errcode_t ret + = compile_range(&p, pend, translate, syntax, b); + if (ret != REG_NOERROR) + FREE_STACK_RETURN(ret); + } -/* We are passed P pointing to a register number after a start_memory. + else if (p[0] == '-' && p[1] != ']') { /* This handles ranges made up of characters only. */ + reg_errcode_t ret; - Return true if the pattern up to the corresponding stop_memory can - match the empty string, and false otherwise. + /* Move past the `-'. */ + PATFETCH(c1); - If we find the matching stop_memory, sets P to point to one past its number. - Otherwise, sets P to an undefined byte less than or equal to END. + ret = compile_range(&p, pend, translate, syntax, b); + if (ret != REG_NOERROR) + FREE_STACK_RETURN(ret); + } - We don't handle duplicates properly (yet). */ + /* See if we're at the beginning of a possible character + class. */ -static boolean group_match_null_string_p(p, end, reg_info) -unsigned char **p, *end; -register_info_type *reg_info; -{ - int mcnt; + else if (syntax & RE_CHAR_CLASSES && c == '[' && *p == ':') { /* Leave room for the null. */ + char str[CHAR_CLASS_MAX_LENGTH + 1]; - /* Point to after the args to the start_memory. */ - unsigned char *p1 = *p + 2; + PATFETCH(c); + c1 = 0; - while (p1 < end) { - /* Skip over opcodes that can match nothing, and return true or - false, as appropriate, when we get to one that can't, or to the - matching stop_memory. */ + /* If pattern is `[[:'. */ + if (p == pend) + FREE_STACK_RETURN(REG_EBRACK); - switch ((re_opcode_t) * p1) { - /* Could be either a loop or a series of alternatives. */ - case on_failure_jump: - p1++; - EXTRACT_NUMBER_AND_INCR(mcnt, p1); + for (;;) { + PATFETCH(c); + if ((c == ':' && *p == ']') || p == pend) + break; + if (c1 < CHAR_CLASS_MAX_LENGTH) + str[c1++] = c; + else + /* This is in any case an invalid class name. */ + str[0] = '\0'; + } + str[c1] = '\0'; - /* If the next operation is not a jump backwards in the - pattern. */ + /* If isn't a word bracketed by `[:' and `:]': + undo the ending character, the letters, and leave + the leading `:' and `[' (but set bits for them). */ + if (c == ':' && *p == ']') { +#if defined _LIBC || WIDE_CHAR_SUPPORT + boolean is_lower = STREQ(str, "lower"); + boolean is_upper = STREQ(str, "upper"); + wctype_t wt; + int ch; - if (mcnt >= 0) { - /* Go through the on_failure_jumps of the alternatives, - seeing if any of the alternatives cannot match nothing. - The last alternative starts with only a jump, - whereas the rest start with on_failure_jump and end - with a jump, e.g., here is the pattern for `a|b|c': + wt = IS_CHAR_CLASS(str); + if (wt == 0) + FREE_STACK_RETURN(REG_ECTYPE); - /on_failure_jump/0/6/exactn/1/a/jump_past_alt/0/6 - /on_failure_jump/0/6/exactn/1/b/jump_past_alt/0/3 - /exactn/1/c + /* Throw away the ] at the end of the character + class. */ + PATFETCH(c); - So, we have to first go through the first (n-1) - alternatives and then deal with the last one separately. */ + if (p == pend) + FREE_STACK_RETURN(REG_EBRACK); + for (ch = 0; ch < 1 << BYTEWIDTH; ++ch) { +# ifdef _LIBC + if (__iswctype(__btowc(ch), wt)) + SET_LIST_BIT(ch); +# else + if (iswctype(btowc(ch), wt)) + SET_LIST_BIT(ch); +# endif - /* Deal with the first (n-1) alternatives, which start - with an on_failure_jump (see above) that jumps to right - past a jump_past_alt. */ + if (translate && (is_upper || is_lower) + && (ISUPPER(ch) || ISLOWER(ch))) + SET_LIST_BIT(ch); + } - while ((re_opcode_t) p1[mcnt - 3] == jump_past_alt) { - /* `mcnt' holds how many bytes long the alternative - is, including the ending `jump_past_alt' and - its number. */ + had_char_class = true; +#else + int ch; + boolean is_alnum = STREQ(str, "alnum"); + boolean is_alpha = STREQ(str, "alpha"); + boolean is_blank = STREQ(str, "blank"); + boolean is_cntrl = STREQ(str, "cntrl"); + boolean is_digit = STREQ(str, "digit"); + boolean is_graph = STREQ(str, "graph"); + boolean is_lower = STREQ(str, "lower"); + boolean is_print = STREQ(str, "print"); + boolean is_punct = STREQ(str, "punct"); + boolean is_space = STREQ(str, "space"); + boolean is_upper = STREQ(str, "upper"); + boolean is_xdigit = STREQ(str, "xdigit"); - if (!alt_match_null_string_p(p1, p1 + mcnt - 3, - reg_info)) return false; + if (!IS_CHAR_CLASS(str)) + FREE_STACK_RETURN(REG_ECTYPE); - /* Move to right after this alternative, including the - jump_past_alt. */ - p1 += mcnt; + /* Throw away the ] at the end of the character + class. */ + PATFETCH(c); - /* Break if it's the beginning of an n-th alternative - that doesn't begin with an on_failure_jump. */ - if ((re_opcode_t) * p1 != on_failure_jump) - break; + if (p == pend) + FREE_STACK_RETURN(REG_EBRACK); - /* Still have to check that it's not an n-th - alternative that starts with an on_failure_jump. */ - p1++; - EXTRACT_NUMBER_AND_INCR(mcnt, p1); - if ((re_opcode_t) p1[mcnt - 3] != jump_past_alt) { - /* Get to the beginning of the n-th alternative. */ - p1 -= 3; - break; + for (ch = 0; ch < 1 << BYTEWIDTH; ch++) { + /* This was split into 3 if's to + avoid an arbitrary limit in some compiler. */ + if ((is_alnum && ISALNUM(ch)) + || (is_alpha && ISALPHA(ch)) + || (is_blank && ISBLANK(ch)) + || (is_cntrl && ISCNTRL(ch))) + SET_LIST_BIT(ch); + if ((is_digit && ISDIGIT(ch)) + || (is_graph && ISGRAPH(ch)) + || (is_lower && ISLOWER(ch)) + || (is_print && ISPRINT(ch))) + SET_LIST_BIT(ch); + if ((is_punct && ISPUNCT(ch)) + || (is_space && ISSPACE(ch)) + || (is_upper && ISUPPER(ch)) + || (is_xdigit && ISXDIGIT(ch))) + SET_LIST_BIT(ch); + if (translate && (is_upper || is_lower) + && (ISUPPER(ch) || ISLOWER(ch))) + SET_LIST_BIT(ch); + } + had_char_class = true; +#endif /* libc || wctype.h */ + } else { + c1++; + while (c1--) + PATUNFETCH; + SET_LIST_BIT('['); + SET_LIST_BIT(':'); + had_char_class = false; } + } else { + had_char_class = false; + SET_LIST_BIT(c); } + } - /* Deal with the last alternative: go back and get number - of the `jump_past_alt' just before it. `mcnt' contains - the length of the alternative. */ - EXTRACT_NUMBER(mcnt, p1 - 2); + /* Discard any (non)matching list bytes that are all 0 at the + end of the map. Decrease the map-length byte too. */ + while ((int) b[-1] > 0 && b[b[-1] - 1] == 0) + b[-1]--; + b += b[-1]; + } + break; - if (!alt_match_null_string_p(p1, p1 + mcnt, reg_info)) - return false; - p1 += mcnt; /* Get past the n-th alternative. */ - } /* if mcnt > 0 */ - break; + case '(': + if (syntax & RE_NO_BK_PARENS) + goto handle_open; + else + goto normal_char; - case stop_memory: - assert(p1[1] == **p); - *p = p1 + 2; - return true; + case ')': + if (syntax & RE_NO_BK_PARENS) + goto handle_close; + else + goto normal_char; - default: - if (!common_op_match_null_string_p(&p1, end, reg_info)) - return false; - } - } /* while p1 < end */ + case '\n': + if (syntax & RE_NEWLINE_ALT) + goto handle_alt; + else + goto normal_char; - return false; -} /* group_match_null_string_p */ + case '|': + if (syntax & RE_NO_BK_VBAR) + goto handle_alt; + else + goto normal_char; -/* Similar to group_match_null_string_p, but doesn't deal with alternatives: - It expects P to be the first byte of a single alternative and END one - byte past the last. The alternative can contain groups. */ -static boolean alt_match_null_string_p(p, end, reg_info) -unsigned char *p, *end; -register_info_type *reg_info; -{ - int mcnt; - unsigned char *p1 = p; + case '{': + if (syntax & RE_INTERVALS && syntax & RE_NO_BK_BRACES) + goto handle_interval; + else + goto normal_char; - while (p1 < end) { - /* Skip over opcodes that can match nothing, and break when we get - to one that can't. */ - switch ((re_opcode_t) * p1) { - /* It's a loop. */ - case on_failure_jump: - p1++; - EXTRACT_NUMBER_AND_INCR(mcnt, p1); - p1 += mcnt; - break; + case '\\': + if (p == pend) + FREE_STACK_RETURN(REG_EESCAPE); - default: - if (!common_op_match_null_string_p(&p1, end, reg_info)) - return false; - } - } /* while p1 < end */ + /* Do not translate the character after the \, so that we can + distinguish, e.g., \B from \b, even if we normally would + translate, e.g., B to b. */ + PATFETCH_RAW(c); - return true; -} /* alt_match_null_string_p */ + switch (c) { + case '(': + if (syntax & RE_NO_BK_PARENS) + goto normal_backslash; + handle_open: + bufp->re_nsub++; + regnum++; -/* Deals with the ops common to group_match_null_string_p and - alt_match_null_string_p. + if (COMPILE_STACK_FULL) { + RETALLOC(compile_stack.stack, compile_stack.size << 1, + compile_stack_elt_t); + if (compile_stack.stack == NULL) + return REG_ESPACE; - Sets P to one after the op and its arguments, if any. */ + compile_stack.size <<= 1; + } -static boolean common_op_match_null_string_p(p, end, reg_info) -unsigned char **p, *end; -register_info_type *reg_info; -{ - int mcnt; - boolean ret; - int reg_no; - unsigned char *p1 = *p; + /* These are the values to restore when we hit end of this + group. They are all relative offsets, so that if the + whole pattern moves because of realloc, they will still + be valid. */ + COMPILE_STACK_TOP.begalt_offset = begalt - bufp->buffer; + COMPILE_STACK_TOP.fixup_alt_jump + = + fixup_alt_jump ? fixup_alt_jump - bufp->buffer + 1 : 0; + COMPILE_STACK_TOP.laststart_offset = b - bufp->buffer; + COMPILE_STACK_TOP.regnum = regnum; - switch ((re_opcode_t) * p1++) { - case no_op: - case begline: - case endline: - case begbuf: - case endbuf: - case wordbeg: - case wordend: - case wordbound: - case notwordbound: -#ifdef emacs - case before_dot: - case at_dot: - case after_dot: -#endif - break; + /* We will eventually replace the 0 with the number of + groups inner to this one. But do not push a + start_memory for groups beyond the last one we can + represent in the compiled pattern. */ + if (regnum <= MAX_REGNUM) { + COMPILE_STACK_TOP.inner_group_offset = + b - bufp->buffer + 2; + BUF_PUSH_3(start_memory, regnum, 0); + } - case start_memory: - reg_no = *p1; - assert(reg_no > 0 && reg_no <= MAX_REGNUM); - ret = group_match_null_string_p(&p1, end, reg_info); + compile_stack.avail++; - /* Have to set this here in case we're checking a group which - contains a group and a back reference to it. */ + fixup_alt_jump = 0; + laststart = 0; + begalt = b; + /* If we've reached MAX_REGNUM groups, then this open + won't actually generate any code, so we'll have to + clear pending_exact explicitly. */ + pending_exact = 0; + break; - if (REG_MATCH_NULL_STRING_P(reg_info[reg_no]) == - MATCH_NULL_UNSET_VALUE) - REG_MATCH_NULL_STRING_P(reg_info[reg_no]) = ret; - if (!ret) - return false; - break; + case ')': + if (syntax & RE_NO_BK_PARENS) + goto normal_backslash; - /* If this is an optimized succeed_n for zero times, make the jump. */ - case jump: - EXTRACT_NUMBER_AND_INCR(mcnt, p1); - if (mcnt >= 0) - p1 += mcnt; - else - return false; - break; + if (COMPILE_STACK_EMPTY) { + if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) + goto normal_backslash; + else + FREE_STACK_RETURN(REG_ERPAREN); + } - case succeed_n: - /* Get to the number of times to succeed. */ - p1 += 2; - EXTRACT_NUMBER_AND_INCR(mcnt, p1); + handle_close: + if (fixup_alt_jump) { /* Push a dummy failure point at the end of the + alternative for a possible future + `pop_failure_jump' to pop. See comments at + `push_dummy_failure' in `re_match_2'. */ + BUF_PUSH(push_dummy_failure); - if (mcnt == 0) { - p1 -= 4; - EXTRACT_NUMBER_AND_INCR(mcnt, p1); - p1 += mcnt; - } else - return false; - break; + /* We allocated space for this jump when we assigned + to `fixup_alt_jump', in the `handle_alt' case below. */ + STORE_JUMP(jump_past_alt, fixup_alt_jump, b - 1); + } - case duplicate: - if (!REG_MATCH_NULL_STRING_P(reg_info[*p1])) - return false; - break; + /* See similar code for backslashed left paren above. */ + if (COMPILE_STACK_EMPTY) { + if (syntax & RE_UNMATCHED_RIGHT_PAREN_ORD) + goto normal_char; + else + FREE_STACK_RETURN(REG_ERPAREN); + } - case set_number_at: - p1 += 4; + /* Since we just checked for an empty stack above, this + ``can't happen''. */ + assert(compile_stack.avail != 0); + { + /* We don't just want to restore into `regnum', because + later groups should continue to be numbered higher, + as in `(ab)c(de)' -- the second group is #2. */ + regnum_t this_group_regnum; - default: - /* All other opcodes mean we cannot match the empty string. */ - return false; - } + compile_stack.avail--; + begalt = + bufp->buffer + COMPILE_STACK_TOP.begalt_offset; + fixup_alt_jump = + COMPILE_STACK_TOP.fixup_alt_jump ? bufp->buffer + + COMPILE_STACK_TOP.fixup_alt_jump - 1 : 0; + laststart = + bufp->buffer + COMPILE_STACK_TOP.laststart_offset; + this_group_regnum = COMPILE_STACK_TOP.regnum; + /* If we've reached MAX_REGNUM groups, then this open + won't actually generate any code, so we'll have to + clear pending_exact explicitly. */ + pending_exact = 0; - *p = p1; - return true; -} /* common_op_match_null_string_p */ + /* We're at the end of the group, so now we know how many + groups were inside this one. */ + if (this_group_regnum <= MAX_REGNUM) { + unsigned char *inner_group_loc + = + bufp->buffer + + COMPILE_STACK_TOP.inner_group_offset; -/* Return zero if TRANSLATE[S1] and TRANSLATE[S2] are identical for LEN - bytes; nonzero otherwise. */ + *inner_group_loc = regnum - this_group_regnum; + BUF_PUSH_3(stop_memory, this_group_regnum, + regnum - this_group_regnum); + } + } + break; -static int bcmp_translate(s1, s2, len, translate) -const char *s1, *s2; -register int len; -RE_TRANSLATE_TYPE translate; -{ - register const unsigned char *p1 = (const unsigned char *) s1; - register const unsigned char *p2 = (const unsigned char *) s2; - while (len) { - if (translate[*p1++] != translate[*p2++]) - return 1; - len--; - } - return 0; -} - -/* Entry points for GNU code. */ + case '|': /* `\|'. */ + if (syntax & RE_LIMITED_OPS || syntax & RE_NO_BK_VBAR) + goto normal_backslash; + handle_alt: + if (syntax & RE_LIMITED_OPS) + goto normal_char; -/* re_compile_pattern is the GNU regular expression compiler: it - compiles PATTERN (of length SIZE) and puts the result in BUFP. - Returns 0 if the pattern was valid, otherwise an error string. + /* Insert before the previous alternative a jump which + jumps to this alternative if the former fails. */ + GET_BUFFER_SPACE(3); + INSERT_JUMP(on_failure_jump, begalt, b + 6); + pending_exact = 0; + b += 3; - Assumes the `allocated' (and perhaps `buffer') and `translate' fields - are set in BUFP on entry. + /* The alternative before this one has a jump after it + which gets executed if it gets matched. Adjust that + jump so it will jump to this alternative's analogous + jump (put in below, which in turn will jump to the next + (if any) alternative's such jump, etc.). The last such + jump jumps to the correct final destination. A picture: + _____ _____ + | | | | + | v | v + a | b | c - We call regex_compile to do the actual compilation. */ + If we are at `b', then fixup_alt_jump right now points to a + three-byte space after `a'. We'll put in the jump, set + fixup_alt_jump to right after `b', and leave behind three + bytes which we'll fill in when we get to after `c'. */ -const char *re_compile_pattern(pattern, length, bufp) -const char *pattern; -size_t length; -struct re_pattern_buffer *bufp; -{ - reg_errcode_t ret; + if (fixup_alt_jump) + STORE_JUMP(jump_past_alt, fixup_alt_jump, b); - /* GNU code is written to assume at least RE_NREGS registers will be set - (and at least one extra will be -1). */ - bufp->regs_allocated = REGS_UNALLOCATED; + /* Mark and leave space for a jump after this alternative, + to be filled in later either by next alternative or + when know we're at the end of a series of alternatives. */ + fixup_alt_jump = b; + GET_BUFFER_SPACE(3); + b += 3; - /* And GNU code determines whether or not to get register information - by passing null for the REGS argument to re_match, etc., not by - setting no_sub. */ - bufp->no_sub = 0; + laststart = 0; + begalt = b; + break; - /* Match anchors at newline. */ - bufp->newline_anchor = 1; - ret = regex_compile(pattern, length, re_syntax_options, bufp); + case '{': + /* If \{ is a literal. */ + if (!(syntax & RE_INTERVALS) + /* If we're at `\{' and it's not the open-interval + operator. */ + || ((syntax & RE_INTERVALS) + && (syntax & RE_NO_BK_BRACES)) || (p - 2 == pattern + && p == pend)) + goto normal_backslash; - if (!ret) - return NULL; - return gettext(re_error_msgid + re_error_msgid_idx[(int) ret]); -} + handle_interval: + { + /* If got here, then the syntax allows intervals. */ -#ifdef _LIBC -weak_alias(__re_compile_pattern, re_compile_pattern) -#endif - /* Entry points compatible with 4.2 BSD regex library. We don't define - them unless specifically requested. */ -#if defined _REGEX_RE_COMP || defined _LIBC -/* BSD has one and only one pattern buffer. */ -static struct re_pattern_buffer re_comp_buf; + /* At least (most) this many matches must be made. */ + int lower_bound = -1, upper_bound = -1; -char * -#ifdef _LIBC -/* Make these definitions weak in libc, so POSIX programs can redefine - these names if they don't use our functions, and still use - regcomp/regexec below without link errors. */ weak_function -#endif -re_comp(s) -const char *s; -{ - reg_errcode_t ret; + beg_interval = p - 1; - if (!s) { - if (!re_comp_buf.buffer) - return gettext("No previous regular expression"); - return 0; - } + if (p == pend) { + if (!(syntax & RE_INTERVALS) + && (syntax & RE_NO_BK_BRACES)) goto + unfetch_interval; + else + FREE_STACK_RETURN(REG_EBRACE); + } - if (!re_comp_buf.buffer) { - re_comp_buf.buffer = (unsigned char *) malloc(200); - if (re_comp_buf.buffer == NULL) - return (char *) gettext(re_error_msgid - + - re_error_msgid_idx[(int) REG_ESPACE]); - re_comp_buf.allocated = 200; + GET_UNSIGNED_NUMBER(lower_bound); - re_comp_buf.fastmap = (char *) malloc(1 << BYTEWIDTH); - if (re_comp_buf.fastmap == NULL) - return (char *) gettext(re_error_msgid - + - re_error_msgid_idx[(int) REG_ESPACE]); - } + if (c == ',') { + GET_UNSIGNED_NUMBER(upper_bound); + if ((!(syntax & RE_NO_BK_BRACES) && c != '\\') + || ((syntax & RE_NO_BK_BRACES) && c != '}')) + FREE_STACK_RETURN(REG_BADBR); - /* Since `re_exec' always passes NULL for the `regs' argument, we - don't need to initialize the pattern buffer fields which affect it. */ + if (upper_bound < 0) + upper_bound = RE_DUP_MAX; + } else + /* Interval such as `{1}' => match exactly once. */ + upper_bound = lower_bound; - /* Match anchors at newlines. */ - re_comp_buf.newline_anchor = 1; + if (lower_bound < 0 || upper_bound > RE_DUP_MAX + || lower_bound > upper_bound) { + if (!(syntax & RE_INTERVALS) + && (syntax & RE_NO_BK_BRACES)) goto + unfetch_interval; + else + FREE_STACK_RETURN(REG_BADBR); + } - ret = regex_compile(s, strlen(s), re_syntax_options, &re_comp_buf); + if (!(syntax & RE_NO_BK_BRACES)) { + if (c != '\\') + FREE_STACK_RETURN(REG_EBRACE); - if (!ret) - return NULL; + PATFETCH(c); + } - /* Yes, we're discarding `const' here if !HAVE_LIBINTL. */ - return (char *) gettext(re_error_msgid + - re_error_msgid_idx[(int) ret]); -} + if (c != '}') { + if (!(syntax & RE_INTERVALS) + && (syntax & RE_NO_BK_BRACES)) goto + unfetch_interval; + else + FREE_STACK_RETURN(REG_BADBR); + } + /* We just parsed a valid interval. */ -int -#ifdef _LIBC - weak_function -#endif -re_exec(s) -const char *s; -{ - const int len = strlen(s); + /* If it's invalid to have no preceding re. */ + if (!laststart) { + if (syntax & RE_CONTEXT_INVALID_OPS) + FREE_STACK_RETURN(REG_BADRPT); + else if (syntax & RE_CONTEXT_INDEP_OPS) + laststart = b; + else + goto unfetch_interval; + } - return - 0 <= re_search(&re_comp_buf, s, len, 0, len, - (struct re_registers *) 0); -} + /* If the upper bound is zero, don't want to succeed at + all; jump from `laststart' to `b + 3', which will be + the end of the buffer after we insert the jump. */ + if (upper_bound == 0) { + GET_BUFFER_SPACE(3); + INSERT_JUMP(jump, laststart, b + 3); + b += 3; + } -#endif /* _REGEX_RE_COMP */ - -/* POSIX.2 functions. Don't define these for Emacs. */ + /* Otherwise, we have a nontrivial interval. When + we're all done, the pattern will look like: + set_number_at + set_number_at + succeed_n + + jump_n + (The upper bound and `jump_n' are omitted if + `upper_bound' is 1, though.) */ + else { /* If the upper bound is > 1, we need to insert + more at the end of the loop. */ + unsigned nbytes = 10 + (upper_bound > 1) * 10; -#ifndef emacs + GET_BUFFER_SPACE(nbytes); -/* regcomp takes a regular expression as a string and compiles it. + /* Initialize lower bound of the `succeed_n', even + though it will be set during matching by its + attendant `set_number_at' (inserted next), + because `re_compile_fastmap' needs to know. + Jump to the `jump_n' we might insert below. */ + INSERT_JUMP2(succeed_n, laststart, + b + 5 + (upper_bound > 1) * 5, + lower_bound); + b += 5; - PREG is a regex_t *. We do not expect any fields to be initialized, - since POSIX says we shouldn't. Thus, we set + /* Code to initialize the lower bound. Insert + before the `succeed_n'. The `5' is the last two + bytes of this `set_number_at', plus 3 bytes of + the following `succeed_n'. */ + insert_op2(set_number_at, laststart, 5, + lower_bound, b); + b += 5; - `buffer' to the compiled pattern; - `used' to the length of the compiled pattern; - `syntax' to RE_SYNTAX_POSIX_EXTENDED if the - REG_EXTENDED bit in CFLAGS is set; otherwise, to - RE_SYNTAX_POSIX_BASIC; - `newline_anchor' to REG_NEWLINE being set in CFLAGS; - `fastmap' to an allocated space for the fastmap; - `fastmap_accurate' to zero; - `re_nsub' to the number of subexpressions in PATTERN. + if (upper_bound > 1) { /* More than one repetition is allowed, so + append a backward jump to the `succeed_n' + that starts this interval. - PATTERN is the address of the pattern string. + When we've reached this during matching, + we'll have matched the interval once, so + jump back only `upper_bound - 1' times. */ + STORE_JUMP2(jump_n, b, laststart + 5, + upper_bound - 1); + b += 5; - CFLAGS is a series of bits which affect compilation. + /* The location we want to set is the second + parameter of the `jump_n'; that is `b-2' as + an absolute address. `laststart' will be + the `set_number_at' we're about to insert; + `laststart+3' the number to set, the source + for the relative address. But we are + inserting into the middle of the pattern -- + so everything is getting moved up by 5. + Conclusion: (b - 2) - (laststart + 3) + 5, + i.e., b - laststart. - If REG_EXTENDED is set, we use POSIX extended syntax; otherwise, we - use POSIX basic syntax. + We insert this at the beginning of the loop + so that if we fail during matching, we'll + reinitialize the bounds. */ + insert_op2(set_number_at, laststart, + b - laststart, upper_bound - 1, b); + b += 5; + } + } + pending_exact = 0; + beg_interval = NULL; + } + break; - If REG_NEWLINE is set, then . and [^...] don't match newline. - Also, regexec will try a match beginning after every newline. + unfetch_interval: + /* If an invalid interval, match the characters as literals. */ + assert(beg_interval); + p = beg_interval; + beg_interval = NULL; - If REG_ICASE is set, then we considers upper- and lowercase - versions of letters to be equivalent when matching. + /* normal_char and normal_backslash need `c'. */ + PATFETCH(c); - If REG_NOSUB is set, then when PREG is passed to regexec, that - routine will report only success or failure, and nothing about the - registers. + if (!(syntax & RE_NO_BK_BRACES)) { + if (p > pattern && p[-1] == '\\') + goto normal_backslash; + } + goto normal_char; - It returns 0 if it succeeds, nonzero if it doesn't. (See regex.h for - the return codes and their meanings.) */ +#ifdef emacs + /* There is no way to specify the before_dot and after_dot + operators. rms says this is ok. --karl */ + case '=': + BUF_PUSH(at_dot); + break; -int regcomp(preg, pattern, cflags) -regex_t *preg; -const char *pattern; -int cflags; -{ - reg_errcode_t ret; - reg_syntax_t syntax - = (cflags & REG_EXTENDED) ? + case 's': + laststart = b; + PATFETCH(c); + BUF_PUSH_2(syntaxspec, syntax_spec_code[c]); + break; - RE_SYNTAX_POSIX_EXTENDED : RE_SYNTAX_POSIX_BASIC; + case 'S': + laststart = b; + PATFETCH(c); + BUF_PUSH_2(notsyntaxspec, syntax_spec_code[c]); + break; +#endif /* emacs */ - /* regex_compile will allocate the space for the compiled pattern. */ - preg->buffer = 0; - preg->allocated = 0; - preg->used = 0; - /* Try to allocate space for the fastmap. */ - preg->fastmap = (char *) malloc(1 << BYTEWIDTH); + case 'w': + if (syntax & RE_NO_GNU_OPS) + goto normal_char; + laststart = b; + BUF_PUSH(wordchar); + break; - if (cflags & REG_ICASE) { - unsigned i; - preg->translate - = (RE_TRANSLATE_TYPE) malloc(CHAR_SET_SIZE - * sizeof(*(RE_TRANSLATE_TYPE) 0)); - if (preg->translate == NULL) - return (int) REG_ESPACE; + case 'W': + if (syntax & RE_NO_GNU_OPS) + goto normal_char; + laststart = b; + BUF_PUSH(notwordchar); + break; - /* Map uppercase characters to corresponding lowercase ones. */ - for (i = 0; i < CHAR_SET_SIZE; i++) - preg->translate[i] = ISUPPER(i) ? TOLOWER(i) : i; - } else - preg->translate = NULL; - /* If REG_NEWLINE is set, newlines are treated differently. */ - if (cflags & REG_NEWLINE) { /* REG_NEWLINE implies neither . nor [^...] match newline. */ - syntax &= ~RE_DOT_NEWLINE; - syntax |= RE_HAT_LISTS_NOT_NEWLINE; - /* It also changes the matching behavior. */ - preg->newline_anchor = 1; - } else - preg->newline_anchor = 0; + case '<': + if (syntax & RE_NO_GNU_OPS) + goto normal_char; + BUF_PUSH(wordbeg); + break; - preg->no_sub = !!(cflags & REG_NOSUB); + case '>': + if (syntax & RE_NO_GNU_OPS) + goto normal_char; + BUF_PUSH(wordend); + break; - /* POSIX says a null character in the pattern terminates it, so we - can use strlen here in compiling the pattern. */ - ret = regex_compile(pattern, strlen(pattern), syntax, preg); + case 'b': + if (syntax & RE_NO_GNU_OPS) + goto normal_char; + BUF_PUSH(wordbound); + break; - /* POSIX doesn't distinguish between an unmatched open-group and an - unmatched close-group: both are REG_EPAREN. */ - if (ret == REG_ERPAREN) - ret = REG_EPAREN; + case 'B': + if (syntax & RE_NO_GNU_OPS) + goto normal_char; + BUF_PUSH(notwordbound); + break; - if (ret == REG_NOERROR && preg->fastmap) { - /* Compute the fastmap now, since regexec cannot modify the pattern - buffer. */ - if (re_compile_fastmap(preg) == -2) { - /* Some error occurred while computing the fastmap, just forget - about it. */ - free(preg->fastmap); - preg->fastmap = NULL; - } - } + case '`': + if (syntax & RE_NO_GNU_OPS) + goto normal_char; + BUF_PUSH(begbuf); + break; - return (int) ret; -} + case '\'': + if (syntax & RE_NO_GNU_OPS) + goto normal_char; + BUF_PUSH(endbuf); + break; -#ifdef _LIBC -weak_alias(__regcomp, regcomp) -#endif -/* regexec searches for a given pattern, specified by PREG, in the - string STRING. + case '1': + case '2': + case '3': + case '4': + case '5': + case '6': + case '7': + case '8': + case '9': + if (syntax & RE_NO_BK_REFS) + goto normal_char; - If NMATCH is zero or REG_NOSUB was set in the cflags argument to - `regcomp', we ignore PMATCH. Otherwise, we assume PMATCH has at - least NMATCH elements, and we set them to the offsets of the - corresponding matched substrings. + c1 = c - '0'; - EFLAGS specifies `execution flags' which affect matching: if - REG_NOTBOL is set, then ^ does not match at the beginning of the - string; if REG_NOTEOL is set, then $ does not match at the end. + if (c1 > regnum) + FREE_STACK_RETURN(REG_ESUBREG); - We return 0 if we find a match and REG_NOMATCH if not. */ -int regexec(preg, string, nmatch, pmatch, eflags) -const regex_t *preg; -const char *string; -size_t nmatch; -regmatch_t pmatch[]; -int eflags; -{ - int ret; - struct re_registers regs; - regex_t private_preg; - int len = strlen(string); - boolean want_reg_info = !preg->no_sub && nmatch > 0; + /* Can't back reference to a subexpression if inside of it. */ + if (group_in_compile_stack(compile_stack, (regnum_t) c1)) + goto normal_char; - private_preg = *preg; + laststart = b; + BUF_PUSH_2(duplicate, c1); + break; - private_preg.not_bol = !!(eflags & REG_NOTBOL); - private_preg.not_eol = !!(eflags & REG_NOTEOL); - /* The user has told us exactly how many registers to return - information about, via `nmatch'. We have to pass that on to the - matching routines. */ - private_preg.regs_allocated = REGS_FIXED; + case '+': + case '?': + if (syntax & RE_BK_PLUS_QM) + goto handle_plus; + else + goto normal_backslash; - if (want_reg_info) { - regs.num_regs = nmatch; - regs.start = TALLOC(nmatch * 2, regoff_t); - if (regs.start == NULL) - return (int) REG_NOMATCH; - regs.end = regs.start + nmatch; - } + default: + normal_backslash: + /* You might think it would be useful for \ to mean + not to translate; but if we don't translate it + it will never match anything. */ + c = TRANSLATE(c); + goto normal_char; + } + break; - /* Perform the searching operation. */ - ret = re_search(&private_preg, string, len, - /* start: */ 0, /* range: */ len, - want_reg_info ? ®s : (struct re_registers *) 0); - /* Copy the register information to the POSIX structure. */ - if (want_reg_info) { - if (ret >= 0) { - unsigned r; + default: + /* Expects the character in `c'. */ + normal_char: + /* If no exactn currently being built. */ + if (!pending_exact + /* If last exactn not at current position. */ + || pending_exact + *pending_exact + 1 != b + /* We have only one byte following the exactn for the count. */ + || *pending_exact == (1 << BYTEWIDTH) - 1 + /* If followed by a repetition operator. */ + || *p == '*' || *p == '^' || ((syntax & RE_BK_PLUS_QM) + ? *p == '\\' && (p[1] == '+' + || p[1] == + '?') : (*p + == + '+' + || + *p + == + '?')) + || ((syntax & RE_INTERVALS) + && ((syntax & RE_NO_BK_BRACES) + ? *p == '{' : (p[0] == '\\' && p[1] == '{')))) { + /* Start building a new exactn. */ - for (r = 0; r < nmatch; r++) { - pmatch[r].rm_so = regs.start[r]; - pmatch[r].rm_eo = regs.end[r]; + laststart = b; + + BUF_PUSH_2(exactn, 0); + pending_exact = b - 1; } - } - /* If we needed the temporary register info, free the space now. */ - free(regs.start); - } + BUF_PUSH(c); + (*pending_exact)++; + break; + } /* switch (c) */ + } /* while p != pend */ - /* We want zero return to mean success, unlike `re_search'. */ - return ret >= 0 ? (int) REG_NOERROR : (int) REG_NOMATCH; -} -#ifdef _LIBC -weak_alias(__regexec, regexec) -#endif -/* Returns a message corresponding to an error code, ERRCODE, returned - from either regcomp or regexec. We don't use PREG here. */ - size_t regerror(errcode, preg, errbuf, errbuf_size) -int errcode; -const regex_t *preg; -char *errbuf; -size_t errbuf_size; -{ - const char *msg; - size_t msg_size; + /* Through the pattern now. */ - if (errcode < 0 || errcode >= (int) (sizeof(re_error_msgid_idx) - / sizeof(re_error_msgid_idx[0]))) - /* Only error codes returned by the rest of the code should be passed - to this routine. If we are given anything else, or if other regex - code generates an invalid error code, then the program has a bug. - Dump core so we can fix it. */ - abort(); + if (fixup_alt_jump) + STORE_JUMP(jump_past_alt, fixup_alt_jump, b); - msg = gettext(re_error_msgid + re_error_msgid_idx[errcode]); + if (!COMPILE_STACK_EMPTY) + FREE_STACK_RETURN(REG_EPAREN); - msg_size = strlen(msg) + 1; /* Includes the null. */ + /* If we don't want backtracking, force success + the first time we reach the end of the compiled pattern. */ + if (syntax & RE_NO_POSIX_BACKTRACKING) + BUF_PUSH(succeed); - if (errbuf_size != 0) { - if (msg_size > errbuf_size) { -#if defined HAVE_MEMPCPY || defined _LIBC - *((char *) __mempcpy(errbuf, msg, errbuf_size - 1)) = '\0'; -#else - memcpy(errbuf, msg, errbuf_size - 1); - errbuf[errbuf_size - 1] = 0; -#endif - } else - memcpy(errbuf, msg, msg_size); - } + free(compile_stack.stack); - return msg_size; -} + /* We have succeeded; set the length of the buffer. */ + bufp->used = b - bufp->buffer; -#ifdef _LIBC -weak_alias(__regerror, regerror) -#endif -/* Free dynamically allocated space used by PREG. */ -void regfree(preg) -regex_t *preg; -{ - if (preg->buffer != NULL) - free(preg->buffer); - preg->buffer = NULL; +#ifdef DEBUG + if (debug) { + DEBUG_PRINT1("\nCompiled pattern: \n"); + print_compiled_pattern(bufp); + } +#endif /* DEBUG */ - preg->allocated = 0; - preg->used = 0; +#ifndef MATCH_MAY_ALLOCATE + /* Initialize the failure stack to the largest possible stack. This + isn't necessary unless we're trying to avoid calling alloca in + the search and match routines. */ + { + int num_regs = bufp->re_nsub + 1; - if (preg->fastmap != NULL) - free(preg->fastmap); - preg->fastmap = NULL; - preg->fastmap_accurate = 0; + /* Since DOUBLE_FAIL_STACK refuses to double only if the current size + is strictly greater than re_max_failures, the largest possible stack + is 2 * re_max_failures failure points. */ + if (fail_stack.size < (2 * re_max_failures * MAX_FAILURE_ITEMS)) { + fail_stack.size = (2 * re_max_failures * MAX_FAILURE_ITEMS); - if (preg->translate != NULL) - free(preg->translate); - preg->translate = NULL; -} +# ifdef emacs + if (!fail_stack.stack) + fail_stack.stack + = (fail_stack_elt_t *) xmalloc(fail_stack.size + * + sizeof + (fail_stack_elt_t)); + else + fail_stack.stack = + (fail_stack_elt_t *) xrealloc(fail_stack.stack, + (fail_stack.size * + sizeof + (fail_stack_elt_t))); +# else /* not emacs */ + if (!fail_stack.stack) + fail_stack.stack + = (fail_stack_elt_t *) malloc(fail_stack.size + * + sizeof + (fail_stack_elt_t)); + else + fail_stack.stack = + (fail_stack_elt_t *) realloc(fail_stack.stack, + (fail_stack.size * + sizeof + (fail_stack_elt_t))); +# endif /* not emacs */ + } -#ifdef _LIBC -weak_alias(__regfree, regfree) -#endif -#endif /* not emacs */ + regex_grow_registers(num_regs); + } +#endif /* not MATCH_MAY_ALLOCATE */ + + return REG_NOERROR; +} /* regex_compile */ -- cgit v1.2.3