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(-) 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 p