/* regcomp.c */ /* * "A fair jaw-cracker dwarf-language must be." --Samwise Gamgee */ /* This file contains functions for compiling a regular expression. See * also regexec.c which funnily enough, contains functions for executing * a regular expression. * * This file is also copied at build time to ext/re/re_comp.c, where * it's built with -DPERL_EXT_RE_BUILD -DPERL_EXT_RE_DEBUG -DPERL_EXT. * This causes the main functions to be compiled under new names and with * debugging support added, which makes "use re 'debug'" work. */ /* NOTE: this is derived from Henry Spencer's regexp code, and should not * confused with the original package (see point 3 below). Thanks, Henry! */ /* Additional note: this code is very heavily munged from Henry's version * in places. In some spots I've traded clarity for efficiency, so don't * blame Henry for some of the lack of readability. */ /* The names of the functions have been changed from regcomp and * regexec to pregcomp and pregexec in order to avoid conflicts * with the POSIX routines of the same names. */ #ifdef PERL_EXT_RE_BUILD /* need to replace pregcomp et al, so enable that */ # ifndef PERL_IN_XSUB_RE # define PERL_IN_XSUB_RE # endif /* need access to debugger hooks */ # if defined(PERL_EXT_RE_DEBUG) && !defined(DEBUGGING) # define DEBUGGING # endif #endif #ifdef PERL_IN_XSUB_RE /* We *really* need to overwrite these symbols: */ # define Perl_pregcomp my_regcomp # define Perl_regdump my_regdump # define Perl_regprop my_regprop # define Perl_pregfree my_regfree # define Perl_re_intuit_string my_re_intuit_string /* *These* symbols are masked to allow static link. */ # define Perl_regnext my_regnext # define Perl_save_re_context my_save_re_context # define Perl_reginitcolors my_reginitcolors # define PERL_NO_GET_CONTEXT #endif /* * pregcomp and pregexec -- regsub and regerror are not used in perl * * Copyright (c) 1986 by University of Toronto. * Written by Henry Spencer. Not derived from licensed software. * * Permission is granted to anyone to use this software for any * purpose on any computer system, and to redistribute it freely, * subject to the following restrictions: * * 1. The author is not responsible for the consequences of use of * this software, no matter how awful, even if they arise * from defects in it. * * 2. The origin of this software must not be misrepresented, either * by explicit claim or by omission. * * 3. Altered versions must be plainly marked as such, and must not * be misrepresented as being the original software. * * **** Alterations to Henry's code are... **** **** Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, **** 2000, 2001, 2002, 2003, 2004, 2005, by Larry Wall and others **** **** You may distribute under the terms of either the GNU General Public **** License or the Artistic License, as specified in the README file. * * Beware that some of this code is subtly aware of the way operator * precedence is structured in regular expressions. Serious changes in * regular-expression syntax might require a total rethink. */ #include "EXTERN.h" #define PERL_IN_REGCOMP_C #include "perl.h" #ifndef PERL_IN_XSUB_RE # include "INTERN.h" #endif #define REG_COMP_C #include "regcomp.h" #ifdef op #undef op #endif /* op */ #ifdef MSDOS # if defined(BUGGY_MSC6) /* MSC 6.00A breaks on op/regexp.t test 85 unless we turn this off */ # pragma optimize("a",off) /* But MSC 6.00A is happy with 'w', for aliases only across function calls*/ # pragma optimize("w",on ) # endif /* BUGGY_MSC6 */ #endif /* MSDOS */ #ifndef STATIC #define STATIC static #endif typedef struct RExC_state_t { U32 flags; /* are we folding, multilining? */ char *precomp; /* uncompiled string. */ regexp *rx; char *start; /* Start of input for compile */ char *end; /* End of input for compile */ char *parse; /* Input-scan pointer. */ I32 whilem_seen; /* number of WHILEM in this expr */ regnode *emit_start; /* Start of emitted-code area */ regnode *emit; /* Code-emit pointer; ®dummy = don't = compiling */ I32 naughty; /* How bad is this pattern? */ I32 sawback; /* Did we see \1, ...? */ U32 seen; I32 size; /* Code size. */ I32 npar; /* () count. */ I32 extralen; I32 seen_zerolen; I32 seen_evals; I32 utf8; #if ADD_TO_REGEXEC char *starttry; /* -Dr: where regtry was called. */ #define RExC_starttry (pRExC_state->starttry) #endif } RExC_state_t; #define RExC_flags (pRExC_state->flags) #define RExC_precomp (pRExC_state->precomp) #define RExC_rx (pRExC_state->rx) #define RExC_start (pRExC_state->start) #define RExC_end (pRExC_state->end) #define RExC_parse (pRExC_state->parse) #define RExC_whilem_seen (pRExC_state->whilem_seen) #define RExC_offsets (pRExC_state->rx->offsets) /* I am not like the others */ #define RExC_emit (pRExC_state->emit) #define RExC_emit_start (pRExC_state->emit_start) #define RExC_naughty (pRExC_state->naughty) #define RExC_sawback (pRExC_state->sawback) #define RExC_seen (pRExC_state->seen) #define RExC_size (pRExC_state->size) #define RExC_npar (pRExC_state->npar) #define RExC_extralen (pRExC_state->extralen) #define RExC_seen_zerolen (pRExC_state->seen_zerolen) #define RExC_seen_evals (pRExC_state->seen_evals) #define RExC_utf8 (pRExC_state->utf8) #define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?') #define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \ ((*s) == '{' && regcurly(s))) #ifdef SPSTART #undef SPSTART /* dratted cpp namespace... */ #endif /* * Flags to be passed up and down. */ #define WORST 0 /* Worst case. */ #define HASWIDTH 0x1 /* Known to match non-null strings. */ #define SIMPLE 0x2 /* Simple enough to be STAR/PLUS operand. */ #define SPSTART 0x4 /* Starts with * or +. */ #define TRYAGAIN 0x8 /* Weeded out a declaration. */ /* Length of a variant. */ typedef struct scan_data_t { I32 len_min; I32 len_delta; I32 pos_min; I32 pos_delta; SV *last_found; I32 last_end; /* min value, <0 unless valid. */ I32 last_start_min; I32 last_start_max; SV **longest; /* Either &l_fixed, or &l_float. */ SV *longest_fixed; I32 offset_fixed; SV *longest_float; I32 offset_float_min; I32 offset_float_max; I32 flags; I32 whilem_c; I32 *last_closep; struct regnode_charclass_class *start_class; } scan_data_t; /* * Forward declarations for pregcomp()'s friends. */ static const scan_data_t zero_scan_data = { 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0}; #define SF_BEFORE_EOL (SF_BEFORE_SEOL|SF_BEFORE_MEOL) #define SF_BEFORE_SEOL 0x1 #define SF_BEFORE_MEOL 0x2 #define SF_FIX_BEFORE_EOL (SF_FIX_BEFORE_SEOL|SF_FIX_BEFORE_MEOL) #define SF_FL_BEFORE_EOL (SF_FL_BEFORE_SEOL|SF_FL_BEFORE_MEOL) #ifdef NO_UNARY_PLUS # define SF_FIX_SHIFT_EOL (0+2) # define SF_FL_SHIFT_EOL (0+4) #else # define SF_FIX_SHIFT_EOL (+2) # define SF_FL_SHIFT_EOL (+4) #endif #define SF_FIX_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FIX_SHIFT_EOL) #define SF_FIX_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FIX_SHIFT_EOL) #define SF_FL_BEFORE_SEOL (SF_BEFORE_SEOL << SF_FL_SHIFT_EOL) #define SF_FL_BEFORE_MEOL (SF_BEFORE_MEOL << SF_FL_SHIFT_EOL) /* 0x20 */ #define SF_IS_INF 0x40 #define SF_HAS_PAR 0x80 #define SF_IN_PAR 0x100 #define SF_HAS_EVAL 0x200 #define SCF_DO_SUBSTR 0x400 #define SCF_DO_STCLASS_AND 0x0800 #define SCF_DO_STCLASS_OR 0x1000 #define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR) #define SCF_WHILEM_VISITED_POS 0x2000 #define UTF (RExC_utf8 != 0) #define LOC ((RExC_flags & PMf_LOCALE) != 0) #define FOLD ((RExC_flags & PMf_FOLD) != 0) #define OOB_UNICODE 12345678 #define OOB_NAMEDCLASS -1 #define CHR_SVLEN(sv) (UTF ? sv_len_utf8(sv) : SvCUR(sv)) #define CHR_DIST(a,b) (UTF ? utf8_distance(a,b) : a - b) /* length of regex to show in messages that don't mark a position within */ #define RegexLengthToShowInErrorMessages 127 /* * If MARKER[12] are adjusted, be sure to adjust the constants at the top * of t/op/regmesg.t, the tests in t/op/re_tests, and those in * op/pragma/warn/regcomp. */ #define MARKER1 "<-- HERE" /* marker as it appears in the description */ #define MARKER2 " <-- HERE " /* marker as it appears within the regex */ #define REPORT_LOCATION " in regex; marked by " MARKER1 " in m/%.*s" MARKER2 "%s/" /* * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given * arg. Show regex, up to a maximum length. If it's too long, chop and add * "...". */ #define FAIL(msg) STMT_START { \ const char *ellipses = ""; \ IV len = RExC_end - RExC_precomp; \ \ if (!SIZE_ONLY) \ SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \ if (len > RegexLengthToShowInErrorMessages) { \ /* chop 10 shorter than the max, to ensure meaning of "..." */ \ len = RegexLengthToShowInErrorMessages - 10; \ ellipses = "..."; \ } \ Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \ msg, (int)len, RExC_precomp, ellipses); \ } STMT_END /* * Calls SAVEDESTRUCTOR_X if needed, then calls Perl_croak with the given * args. Show regex, up to a maximum length. If it's too long, chop and add * "...". */ #define FAIL2(pat,msg) STMT_START { \ const char *ellipses = ""; \ IV len = RExC_end - RExC_precomp; \ \ if (!SIZE_ONLY) \ SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \ if (len > RegexLengthToShowInErrorMessages) { \ /* chop 10 shorter than the max, to ensure meaning of "..." */ \ len = RegexLengthToShowInErrorMessages - 10; \ ellipses = "..."; \ } \ S_re_croak2(aTHX_ pat, " in regex m/%.*s%s/", \ msg, (int)len, RExC_precomp, ellipses); \ } STMT_END /* * Simple_vFAIL -- like FAIL, but marks the current location in the scan */ #define Simple_vFAIL(m) STMT_START { \ const IV offset = RExC_parse - RExC_precomp; \ Perl_croak(aTHX_ "%s" REPORT_LOCATION, \ m, (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END /* * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL() */ #define vFAIL(m) STMT_START { \ if (!SIZE_ONLY) \ SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \ Simple_vFAIL(m); \ } STMT_END /* * Like Simple_vFAIL(), but accepts two arguments. */ #define Simple_vFAIL2(m,a1) STMT_START { \ const IV offset = RExC_parse - RExC_precomp; \ S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, \ (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END /* * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL2(). */ #define vFAIL2(m,a1) STMT_START { \ if (!SIZE_ONLY) \ SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \ Simple_vFAIL2(m, a1); \ } STMT_END /* * Like Simple_vFAIL(), but accepts three arguments. */ #define Simple_vFAIL3(m, a1, a2) STMT_START { \ const IV offset = RExC_parse - RExC_precomp; \ S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, \ (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END /* * Calls SAVEDESTRUCTOR_X if needed, then Simple_vFAIL3(). */ #define vFAIL3(m,a1,a2) STMT_START { \ if (!SIZE_ONLY) \ SAVEDESTRUCTOR_X(clear_re,(void*)RExC_rx); \ Simple_vFAIL3(m, a1, a2); \ } STMT_END /* * Like Simple_vFAIL(), but accepts four arguments. */ #define Simple_vFAIL4(m, a1, a2, a3) STMT_START { \ const IV offset = RExC_parse - RExC_precomp; \ S_re_croak2(aTHX_ m, REPORT_LOCATION, a1, a2, a3, \ (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END #define vWARN(loc,m) STMT_START { \ const IV offset = loc - RExC_precomp; \ Perl_warner(aTHX_ packWARN(WARN_REGEXP), "%s" REPORT_LOCATION, \ m, (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END #define vWARNdep(loc,m) STMT_START { \ const IV offset = loc - RExC_precomp; \ Perl_warner(aTHX_ packWARN2(WARN_DEPRECATED, WARN_REGEXP), \ "%s" REPORT_LOCATION, \ m, (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END #define vWARN2(loc, m, a1) STMT_START { \ const IV offset = loc - RExC_precomp; \ Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ a1, (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END #define vWARN3(loc, m, a1, a2) STMT_START { \ const IV offset = loc - RExC_precomp; \ Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ a1, a2, (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END #define vWARN4(loc, m, a1, a2, a3) STMT_START { \ const IV offset = loc - RExC_precomp; \ Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ a1, a2, a3, (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END #define vWARN5(loc, m, a1, a2, a3, a4) STMT_START { \ const IV offset = loc - RExC_precomp; \ Perl_warner(aTHX_ packWARN(WARN_REGEXP), m REPORT_LOCATION, \ a1, a2, a3, a4, (int)offset, RExC_precomp, RExC_precomp + offset); \ } STMT_END /* Allow for side effects in s */ #define REGC(c,s) STMT_START { \ if (!SIZE_ONLY) *(s) = (c); else (void)(s); \ } STMT_END /* Macros for recording node offsets. 20001227 mjd@plover.com * Nodes are numbered 1, 2, 3, 4. Node #n's position is recorded in * element 2*n-1 of the array. Element #2n holds the byte length node #n. * Element 0 holds the number n. */ #define MJD_OFFSET_DEBUG(x) /* #define MJD_OFFSET_DEBUG(x) DEBUG_r(Perl_warn_nocontext x) */ #define Set_Node_Offset_To_R(node,byte) STMT_START { \ if (! SIZE_ONLY) { \ MJD_OFFSET_DEBUG(("** (%d) offset of node %d is %d.\n", \ __LINE__, (node), (byte))); \ if((node) < 0) { \ Perl_croak(aTHX_ "value of node is %d in Offset macro", node); \ } else { \ RExC_offsets[2*(node)-1] = (byte); \ } \ } \ } STMT_END #define Set_Node_Offset(node,byte) \ Set_Node_Offset_To_R((node)-RExC_emit_start, (byte)-RExC_start) #define Set_Cur_Node_Offset Set_Node_Offset(RExC_emit, RExC_parse) #define Set_Node_Length_To_R(node,len) STMT_START { \ if (! SIZE_ONLY) { \ MJD_OFFSET_DEBUG(("** (%d) size of node %d is %d.\n", \ __LINE__, (node), (len))); \ if((node) < 0) { \ Perl_croak(aTHX_ "value of node is %d in Length macro", node); \ } else { \ RExC_offsets[2*(node)] = (len); \ } \ } \ } STMT_END #define Set_Node_Length(node,len) \ Set_Node_Length_To_R((node)-RExC_emit_start, len) #define Set_Cur_Node_Length(len) Set_Node_Length(RExC_emit, len) #define Set_Node_Cur_Length(node) \ Set_Node_Length(node, RExC_parse - parse_start) /* Get offsets and lengths */ #define Node_Offset(n) (RExC_offsets[2*((n)-RExC_emit_start)-1]) #define Node_Length(n) (RExC_offsets[2*((n)-RExC_emit_start)]) static void clear_re(pTHX_ void *r); /* Mark that we cannot extend a found fixed substring at this point. Updata the longest found anchored substring and the longest found floating substrings if needed. */ STATIC void S_scan_commit(pTHX_ RExC_state_t *pRExC_state, scan_data_t *data) ###### { ###### const STRLEN l = CHR_SVLEN(data->last_found); ###### const STRLEN old_l = CHR_SVLEN(*data->longest); ###### if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) { ###### SvSetMagicSV(*data->longest, data->last_found); ###### if (*data->longest == data->longest_fixed) { ###### data->offset_fixed = l ? data->last_start_min : data->pos_min; ###### if (data->flags & SF_BEFORE_EOL) ###### data->flags |= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL); else ###### data->flags &= ~SF_FIX_BEFORE_EOL; } else { ###### data->offset_float_min = l ? data->last_start_min : data->pos_min; ###### data->offset_float_max = (l ? data->last_start_max : data->pos_min + data->pos_delta); ###### if ((U32)data->offset_float_max > (U32)I32_MAX) ###### data->offset_float_max = I32_MAX; ###### if (data->flags & SF_BEFORE_EOL) ###### data->flags |= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL); else ###### data->flags &= ~SF_FL_BEFORE_EOL; } } ###### SvCUR_set(data->last_found, 0); { ###### SV * const sv = data->last_found; ###### MAGIC * const mg = ###### SvUTF8(sv) && SvMAGICAL(sv) ? mg_find(sv, PERL_MAGIC_utf8) : NULL; ###### if (mg && mg->mg_len > 0) ###### mg->mg_len = 0; } ###### data->last_end = -1; ###### data->flags &= ~SF_BEFORE_EOL; } /* Can match anything (initialization) */ STATIC void S_cl_anything(pTHX_ RExC_state_t *pRExC_state, struct regnode_charclass_class *cl) ###### { ###### ANYOF_CLASS_ZERO(cl); ###### ANYOF_BITMAP_SETALL(cl); ###### cl->flags = ANYOF_EOS|ANYOF_UNICODE_ALL; ###### if (LOC) ###### cl->flags |= ANYOF_LOCALE; } /* Can match anything (initialization) */ STATIC int S_cl_is_anything(pTHX_ const struct regnode_charclass_class *cl) ###### { ###### int value; ###### for (value = 0; value <= ANYOF_MAX; value += 2) ###### if (ANYOF_CLASS_TEST(cl, value) && ANYOF_CLASS_TEST(cl, value + 1)) ###### return 1; ###### if (!(cl->flags & ANYOF_UNICODE_ALL)) ###### return 0; ###### if (!ANYOF_BITMAP_TESTALLSET(cl)) ###### return 0; ###### return 1; } /* Can match anything (initialization) */ STATIC void S_cl_init(pTHX_ RExC_state_t *pRExC_state, struct regnode_charclass_class *cl) ###### { ###### Zero(cl, 1, struct regnode_charclass_class); ###### cl->type = ANYOF; ###### cl_anything(pRExC_state, cl); } STATIC void S_cl_init_zero(pTHX_ RExC_state_t *pRExC_state, struct regnode_charclass_class *cl) ###### { ###### Zero(cl, 1, struct regnode_charclass_class); ###### cl->type = ANYOF; ###### cl_anything(pRExC_state, cl); ###### if (LOC) ###### cl->flags |= ANYOF_LOCALE; } /* 'And' a given class with another one. Can create false positives */ /* We assume that cl is not inverted */ STATIC void S_cl_and(pTHX_ struct regnode_charclass_class *cl, const struct regnode_charclass_class *and_with) ###### { ###### if (!(and_with->flags & ANYOF_CLASS) && !(cl->flags & ANYOF_CLASS) && (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) && !(and_with->flags & ANYOF_FOLD) && !(cl->flags & ANYOF_FOLD)) { ###### int i; ###### if (and_with->flags & ANYOF_INVERT) ###### for (i = 0; i < ANYOF_BITMAP_SIZE; i++) ###### cl->bitmap[i] &= ~and_with->bitmap[i]; else ###### for (i = 0; i < ANYOF_BITMAP_SIZE; i++) ###### cl->bitmap[i] &= and_with->bitmap[i]; } /* XXXX: logic is complicated otherwise, leave it along for a moment. */ ###### if (!(and_with->flags & ANYOF_EOS)) ###### cl->flags &= ~ANYOF_EOS; ###### if (cl->flags & ANYOF_UNICODE_ALL && and_with->flags & ANYOF_UNICODE && !(and_with->flags & ANYOF_INVERT)) { ###### cl->flags &= ~ANYOF_UNICODE_ALL; ###### cl->flags |= ANYOF_UNICODE; ###### ARG_SET(cl, ARG(and_with)); } ###### if (!(and_with->flags & ANYOF_UNICODE_ALL) && !(and_with->flags & ANYOF_INVERT)) ###### cl->flags &= ~ANYOF_UNICODE_ALL; ###### if (!(and_with->flags & (ANYOF_UNICODE|ANYOF_UNICODE_ALL)) && !(and_with->flags & ANYOF_INVERT)) ###### cl->flags &= ~ANYOF_UNICODE; } /* 'OR' a given class with another one. Can create false positives */ /* We assume that cl is not inverted */ STATIC void S_cl_or(pTHX_ RExC_state_t *pRExC_state, struct regnode_charclass_class *cl, const struct regnode_charclass_class *or_with) ###### { ###### if (or_with->flags & ANYOF_INVERT) { /* We do not use * (B1 | CL1) | (!B2 & !CL2) = (B1 | !B2 & !CL2) | (CL1 | (!B2 & !CL2)) * <= (B1 | !B2) | (CL1 | !CL2) * which is wasteful if CL2 is small, but we ignore CL2: * (B1 | CL1) | (!B2 & !CL2) <= (B1 | CL1) | !B2 = (B1 | !B2) | CL1 * XXXX Can we handle case-fold? Unclear: * (OK1(i) | OK1(i')) | !(OK1(i) | OK1(i')) = * (OK1(i) | OK1(i')) | (!OK1(i) & !OK1(i')) */ ###### if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) && !(or_with->flags & ANYOF_FOLD) && !(cl->flags & ANYOF_FOLD) ) { ###### int i; ###### for (i = 0; i < ANYOF_BITMAP_SIZE; i++) ###### cl->bitmap[i] |= ~or_with->bitmap[i]; } /* XXXX: logic is complicated otherwise */ else { ###### cl_anything(pRExC_state, cl); } } else { /* (B1 | CL1) | (B2 | CL2) = (B1 | B2) | (CL1 | CL2)) */ ###### if ( (or_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE) && (!(or_with->flags & ANYOF_FOLD) || (cl->flags & ANYOF_FOLD)) ) { ###### int i; /* OR char bitmap and class bitmap separately */ ###### for (i = 0; i < ANYOF_BITMAP_SIZE; i++) ###### cl->bitmap[i] |= or_with->bitmap[i]; ###### if (or_with->flags & ANYOF_CLASS) { ###### for (i = 0; i < ANYOF_CLASSBITMAP_SIZE; i++) ###### cl->classflags[i] |= or_with->classflags[i]; ###### cl->flags |= ANYOF_CLASS; } } else { /* XXXX: logic is complicated, leave it along for a moment. */ ###### cl_anything(pRExC_state, cl); } } ###### if (or_with->flags & ANYOF_EOS) ###### cl->flags |= ANYOF_EOS; ###### if (cl->flags & ANYOF_UNICODE && or_with->flags & ANYOF_UNICODE && ARG(cl) != ARG(or_with)) { ###### cl->flags |= ANYOF_UNICODE_ALL; ###### cl->flags &= ~ANYOF_UNICODE; } ###### if (or_with->flags & ANYOF_UNICODE_ALL) { ###### cl->flags |= ANYOF_UNICODE_ALL; ###### cl->flags &= ~ANYOF_UNICODE; } } /* make_trie(startbranch,first,last,tail,flags) startbranch: the first branch in the whole branch sequence first : start branch of sequence of branch-exact nodes. May be the same as startbranch last : Thing following the last branch. May be the same as tail. tail : item following the branch sequence flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/ Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node. A trie is an N'ary tree where the branches are determined by digital decomposition of the key. IE, at the root node you look up the 1st character and follow that branch repeat until you find the end of the branches. Nodes can be marked as "accepting" meaning they represent a complete word. Eg: /he|she|his|hers/ would convert into the following structure. Numbers represent states, letters following numbers represent valid transitions on the letter from that state, if the number is in square brackets it represents an accepting state, otherwise it will be in parenthesis. +-h->+-e->[3]-+-r->(8)-+-s->[9] | | | (2) | | (1) +-i->(6)-+-s->[7] | +-s->(3)-+-h->(4)-+-e->[5] Accept Word Mapping: 3=>1 (he),5=>2 (she), 7=>3 (his), 9=>4 (hers) This shows that when matching against the string 'hers' we will begin at state 1 read 'h' and move to state 2, read 'e' and move to state 3 which is accepting, then read 'r' and go to state 8 followed by 's' which takes us to state 9 which is also accepting. Thus we know that we can match both 'he' and 'hers' with a single traverse. We store a mapping from accepting to state to which word was matched, and then when we have multiple possibilities we try to complete the rest of the regex in the order in which they occured in the alternation. The only prior NFA like behaviour that would be changed by the TRIE support is the silent ignoring of duplicate alternations which are of the form: / (DUPE|DUPE) X? (?{ ... }) Y /x Thus EVAL blocks follwing a trie may be called a different number of times with and without the optimisation. With the optimisations dupes will be silently ignored. This inconsistant behaviour of EVAL type nodes is well established as the following demonstrates: 'words'=~/(word|word|word)(?{ print $1 })[xyz]/ which prints out 'word' three times, but 'words'=~/(word|word|word)(?{ print $1 })S/ which doesnt print it out at all. This is due to other optimisations kicking in. Example of what happens on a structural level: The regexp /(ac|ad|ab)+/ will produce the folowing debug output: 1: CURLYM[1] {1,32767}(18) 5: BRANCH(8) 6: EXACT (16) 8: BRANCH(11) 9: EXACT (16) 11: BRANCH(14) 12: EXACT (16) 16: SUCCEED(0) 17: NOTHING(18) 18: END(0) This would be optimizable with startbranch=5, first=5, last=16, tail=16 and should turn into: 1: CURLYM[1] {1,32767}(18) 5: TRIE(16) [Words:3 Chars Stored:6 Unique Chars:4 States:5 NCP:1] 16: SUCCEED(0) 17: NOTHING(18) 18: END(0) Cases where tail != last would be like /(?foo|bar)baz/: 1: BRANCH(4) 2: EXACT (8) 4: BRANCH(7) 5: EXACT (8) 7: TAIL(8) 8: EXACT (10) 10: END(0) which would be optimizable with startbranch=1, first=1, last=7, tail=8 and would end up looking like: 1: TRIE(8) [Words:2 Chars Stored:6 Unique Chars:5 States:7 NCP:1] 7: TAIL(8) 8: EXACT (10) 10: END(0) */ #define TRIE_DEBUG_CHAR \ DEBUG_TRIE_COMPILE_r({ \ SV *tmp; \ if ( UTF ) { \ tmp = newSVpvn( "", 0 ); \ pv_uni_display( tmp, uc, len, 60, UNI_DISPLAY_REGEX ); \ } else { \ tmp = Perl_newSVpvf_nocontext( "%c", (int)uvc ); \ } \ av_push( trie->revcharmap, tmp ); \ }) #define TRIE_READ_CHAR STMT_START { \ if ( UTF ) { \ if ( folder ) { \ if ( foldlen > 0 ) { \ uvc = utf8n_to_uvuni( scan, UTF8_MAXLEN, &len, uniflags ); \ foldlen -= len; \ scan += len; \ len = 0; \ } else { \ uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\ uvc = to_uni_fold( uvc, foldbuf, &foldlen ); \ foldlen -= UNISKIP( uvc ); \ scan = foldbuf + UNISKIP( uvc ); \ } \ } else { \ uvc = utf8n_to_uvuni( (const U8*)uc, UTF8_MAXLEN, &len, uniflags);\ } \ } else { \ uvc = (U32)*uc; \ len = 1; \ } \ } STMT_END #define TRIE_LIST_ITEM(state,idx) (trie->states[state].trans.list)[ idx ] #define TRIE_LIST_CUR(state) ( TRIE_LIST_ITEM( state, 0 ).forid ) #define TRIE_LIST_LEN(state) ( TRIE_LIST_ITEM( state, 0 ).newstate ) #define TRIE_LIST_USED(idx) ( trie->states[state].trans.list ? (TRIE_LIST_CUR( idx ) - 1) : 0 ) #define TRIE_LIST_PUSH(state,fid,ns) STMT_START { \ if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \ TRIE_LIST_LEN( state ) *= 2; \ Renew( trie->states[ state ].trans.list, \ TRIE_LIST_LEN( state ), reg_trie_trans_le ); \ } \ TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).forid = fid; \ TRIE_LIST_ITEM( state, TRIE_LIST_CUR( state ) ).newstate = ns; \ TRIE_LIST_CUR( state )++; \ } STMT_END #define TRIE_LIST_NEW(state) STMT_START { \ Newz( 1023, trie->states[ state ].trans.list, \ 4, reg_trie_trans_le ); \ TRIE_LIST_CUR( state ) = 1; \ TRIE_LIST_LEN( state ) = 4; \ } STMT_END STATIC I32 S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 flags) ###### { dVAR; /* first pass, loop through and scan words */ ###### reg_trie_data *trie; ###### regnode *cur; ###### const U32 uniflags = ckWARN(WARN_UTF8) ? 0 : UTF8_ALLOW_ANY; ###### STRLEN len = 0; ###### UV uvc = 0; ###### U16 curword = 0; ###### U32 next_alloc = 0; /* we just use folder as a flag in utf8 */ ###### const U8 * const folder = ( flags == EXACTF ? PL_fold : ( flags == EXACTFL ? PL_fold_locale : NULL ) ###### ); ###### const U32 data_slot = add_data( pRExC_state, 1, "t" ); ###### SV *re_trie_maxbuff; ###### GET_RE_DEBUG_FLAGS_DECL; ###### Newz( 848200, trie, 1, reg_trie_data ); ###### trie->refcount = 1; ###### RExC_rx->data->data[ data_slot ] = (void*)trie; ###### Newz( 848201, trie->charmap, 256, U16 ); DEBUG_r({ trie->words = newAV(); trie->revcharmap = newAV(); ###### }); ###### re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1); ###### if (!SvIOK(re_trie_maxbuff)) { ###### sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); } /* -- First loop and Setup -- We first traverse the branches and scan each word to determine if it contains widechars, and how many unique chars there are, this is important as we have to build a table with at least as many columns as we have unique chars. We use an array of integers to represent the character codes 0..255 (trie->charmap) and we use a an HV* to store unicode characters. We use the native representation of the character value as the key and IV's for the coded index. *TODO* If we keep track of how many times each character is used we can remap the columns so that the table compression later on is more efficient in terms of memory by ensuring most common value is in the middle and the least common are on the outside. IMO this would be better than a most to least common mapping as theres a decent chance the most common letter will share a node with the least common, meaning the node will not be compressable. With a middle is most common approach the worst case is when we have the least common nodes twice. */ ###### for ( cur = first ; cur < last ; cur = regnext( cur ) ) { ###### regnode *noper = NEXTOPER( cur ); ###### const U8 *uc = (U8*)STRING( noper ); ###### const U8 * const e = uc + STR_LEN( noper ); ###### STRLEN foldlen = 0; ###### U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; ###### const U8 *scan = (U8*)NULL; ###### for ( ; uc < e ; uc += len ) { ###### trie->charcount++; ###### TRIE_READ_CHAR; ###### if ( uvc < 256 ) { ###### if ( !trie->charmap[ uvc ] ) { ###### trie->charmap[ uvc ]=( ++trie->uniquecharcount ); ###### if ( folder ) ###### trie->charmap[ folder[ uvc ] ] = trie->charmap[ uvc ]; ###### TRIE_DEBUG_CHAR; } } else { ###### SV** svpp; ###### if ( !trie->widecharmap ) ###### trie->widecharmap = newHV(); ###### svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 1 ); ###### if ( !svpp ) ###### Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc ); ###### if ( !SvTRUE( *svpp ) ) { ###### sv_setiv( *svpp, ++trie->uniquecharcount ); ###### TRIE_DEBUG_CHAR; } } } ###### trie->wordcount++; } /* end first pass */ DEBUG_TRIE_COMPILE_r( PerlIO_printf( Perl_debug_log, "TRIE(%s): W:%d C:%d Uq:%d \n", ( trie->widecharmap ? "UTF8" : "NATIVE" ), trie->wordcount, (int)trie->charcount, trie->uniquecharcount ) ###### ); /* We now know what we are dealing with in terms of unique chars and string sizes so we can calculate how much memory a naive representation using a flat table will take. If it's over a reasonable limit (as specified by ${^RE_TRIE_MAXBUF}) we use a more memory conservative but potentially much slower representation using an array of lists. At the end we convert both representations into the same compressed form that will be used in regexec.c for matching with. The latter is a form that cannot be used to construct with but has memory properties similar to the list form and access properties similar to the table form making it both suitable for fast searches and small enough that its feasable to store for the duration of a program. See the comment in the code where the compressed table is produced inplace from the flat tabe representation for an explanation of how the compression works. */ ###### if ( (IV)( ( trie->charcount + 1 ) * trie->uniquecharcount + 1) > SvIV(re_trie_maxbuff) ) { /* Second Pass -- Array Of Lists Representation Each state will be represented by a list of charid:state records (reg_trie_trans_le) the first such element holds the CUR and LEN points of the allocated array. (See defines above). We build the initial structure using the lists, and then convert it into the compressed table form which allows faster lookups (but cant be modified once converted). */ ###### STRLEN transcount = 1; ###### Newz( 848204, trie->states, trie->charcount + 2, reg_trie_state ); ###### TRIE_LIST_NEW(1); ###### next_alloc = 2; ###### for ( cur = first ; cur < last ; cur = regnext( cur ) ) { ###### regnode *noper = NEXTOPER( cur ); ###### U8 *uc = (U8*)STRING( noper ); ###### const U8 * const e = uc + STR_LEN( noper ); ###### U32 state = 1; /* required init */ ###### U16 charid = 0; /* sanity init */ ###### U8 *scan = (U8*)NULL; /* sanity init */ ###### STRLEN foldlen = 0; /* required init */ ###### U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; ###### for ( ; uc < e ; uc += len ) { ###### TRIE_READ_CHAR; ###### if ( uvc < 256 ) { ###### charid = trie->charmap[ uvc ]; } else { ###### SV** svpp=(SV**)NULL; ###### svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 0); ###### if ( !svpp ) { ###### charid = 0; } else { ###### charid=(U16)SvIV( *svpp ); } } ###### if ( charid ) { ###### U16 check; ###### U32 newstate = 0; ###### charid--; ###### if ( !trie->states[ state ].trans.list ) { ###### TRIE_LIST_NEW( state ); } ###### for ( check = 1; check <= TRIE_LIST_USED( state ); check++ ) { ###### if ( TRIE_LIST_ITEM( state, check ).forid == charid ) { ###### newstate = TRIE_LIST_ITEM( state, check ).newstate; ###### break; } } ###### if ( ! newstate ) { ###### newstate = next_alloc++; ###### TRIE_LIST_PUSH( state, charid, newstate ); ###### transcount++; } ###### state = newstate; } else { ###### Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc ); } /* charid is now 0 if we dont know the char read, or nonzero if we do */ } ###### if ( !trie->states[ state ].wordnum ) { /* we havent inserted this word into the structure yet. */ ###### trie->states[ state ].wordnum = ++curword; DEBUG_r({ /* store the word for dumping */ SV* tmp = newSVpvn( STRING( noper ), STR_LEN( noper ) ); if ( UTF ) SvUTF8_on( tmp ); av_push( trie->words, tmp ); ###### }); } else { /* Its a dupe. So ignore it. */ } } /* end second pass */ ###### trie->laststate = next_alloc; ###### Renew( trie->states, next_alloc, reg_trie_state ); DEBUG_TRIE_COMPILE_MORE_r({ U32 state; /* print out the table precompression. */ PerlIO_printf( Perl_debug_log, "\nState :Word | Transition Data\n" ); PerlIO_printf( Perl_debug_log, "------:-----+-----------------" ); for( state=1 ; state < next_alloc ; state ++ ) { U16 charid; PerlIO_printf( Perl_debug_log, "\n %04"UVXf" :", (UV)state ); if ( ! trie->states[ state ].wordnum ) { PerlIO_printf( Perl_debug_log, "%5s| ",""); } else { PerlIO_printf( Perl_debug_log, "W%04x| ", trie->states[ state ].wordnum ); } for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) { SV **tmp = av_fetch( trie->revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0); PerlIO_printf( Perl_debug_log, "%s:%3X=%04"UVXf" | ", SvPV_nolen_const( *tmp ), TRIE_LIST_ITEM(state,charid).forid, (UV)TRIE_LIST_ITEM(state,charid).newstate ); } } PerlIO_printf( Perl_debug_log, "\n\n" ); ###### }); ###### Newz( 848203, trie->trans, transcount ,reg_trie_trans ); { ###### U32 state; ###### U32 tp = 0; ###### U32 zp = 0; ###### for( state=1 ; state < next_alloc ; state ++ ) { ###### U32 base=0; /* DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log, "tp: %d zp: %d ",tp,zp) ); */ ###### if (trie->states[state].trans.list) { ###### U16 minid=TRIE_LIST_ITEM( state, 1).forid; ###### U16 maxid=minid; ###### U16 idx; ###### for( idx = 2 ; idx <= TRIE_LIST_USED( state ) ; idx++ ) { ###### if ( TRIE_LIST_ITEM( state, idx).forid < minid ) { ###### minid=TRIE_LIST_ITEM( state, idx).forid; ###### } else if ( TRIE_LIST_ITEM( state, idx).forid > maxid ) { ###### maxid=TRIE_LIST_ITEM( state, idx).forid; } } ###### if ( transcount < tp + maxid - minid + 1) { ###### transcount *= 2; ###### Renew( trie->trans, transcount, reg_trie_trans ); ###### Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans ); } ###### base = trie->uniquecharcount + tp - minid; ###### if ( maxid == minid ) { ###### U32 set = 0; ###### for ( ; zp < tp ; zp++ ) { ###### if ( ! trie->trans[ zp ].next ) { ###### base = trie->uniquecharcount + zp - minid; ###### trie->trans[ zp ].next = TRIE_LIST_ITEM( state, 1).newstate; ###### trie->trans[ zp ].check = state; ###### set = 1; ###### break; } } ###### if ( !set ) { ###### trie->trans[ tp ].next = TRIE_LIST_ITEM( state, 1).newstate; ###### trie->trans[ tp ].check = state; ###### tp++; ###### zp = tp; } } else { ###### for ( idx=1; idx <= TRIE_LIST_USED( state ) ; idx++ ) { ###### U32 tid = base - trie->uniquecharcount + TRIE_LIST_ITEM( state, idx ).forid; ###### trie->trans[ tid ].next = TRIE_LIST_ITEM( state, idx ).newstate; ###### trie->trans[ tid ].check = state; } ###### tp += ( maxid - minid + 1 ); } ###### Safefree(trie->states[ state ].trans.list); } /* DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log, " base: %d\n",base); ); */ ###### trie->states[ state ].trans.base=base; } ###### trie->lasttrans = tp + 1; } } else { /* Second Pass -- Flat Table Representation. we dont use the 0 slot of either trans[] or states[] so we add 1 to each. We know that we will need Charcount+1 trans at most to store the data (one row per char at worst case) So we preallocate both structures assuming worst case. We then construct the trie using only the .next slots of the entry structs. We use the .check field of the first entry of the node temporarily to make compression both faster and easier by keeping track of how many non zero fields are in the node. Since trans are numbered from 1 any 0 pointer in the table is a FAIL transition. There are two terms at use here: state as a TRIE_NODEIDX() which is a number representing the first entry of the node, and state as a TRIE_NODENUM() which is the trans number. state 1 is TRIE_NODEIDX(1) and TRIE_NODENUM(1), state 2 is TRIE_NODEIDX(2) and TRIE_NODENUM(3) if there are 2 entrys per node. eg: A B A B 1. 2 4 1. 3 7 2. 0 3 3. 0 5 3. 0 0 5. 0 0 4. 0 0 7. 0 0 The table is internally in the right hand, idx form. However as we also have to deal with the states array which is indexed by nodenum we have to use TRIE_NODENUM() to convert. */ Newz( 848203, trie->trans, ( trie->charcount + 1 ) * trie->uniquecharcount + 1, ###### reg_trie_trans ); ###### Newz( 848204, trie->states, trie->charcount + 2, reg_trie_state ); ###### next_alloc = trie->uniquecharcount + 1; ###### for ( cur = first ; cur < last ; cur = regnext( cur ) ) { ###### regnode *noper = NEXTOPER( cur ); ###### const U8 *uc = (U8*)STRING( noper ); ###### const U8 * const e = uc + STR_LEN( noper ); ###### U32 state = 1; /* required init */ ###### U16 charid = 0; /* sanity init */ ###### U32 accept_state = 0; /* sanity init */ ###### U8 *scan = (U8*)NULL; /* sanity init */ ###### STRLEN foldlen = 0; /* required init */ ###### U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ]; ###### for ( ; uc < e ; uc += len ) { ###### TRIE_READ_CHAR; ###### if ( uvc < 256 ) { ###### charid = trie->charmap[ uvc ]; } else { ###### SV** svpp=(SV**)NULL; ###### svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 0); ###### if ( !svpp ) { ###### charid = 0; } else { ###### charid=(U16)SvIV( *svpp ); } } ###### if ( charid ) { ###### charid--; ###### if ( !trie->trans[ state + charid ].next ) { ###### trie->trans[ state + charid ].next = next_alloc; ###### trie->trans[ state ].check++; ###### next_alloc += trie->uniquecharcount; } ###### state = trie->trans[ state + charid ].next; } else { ###### Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc ); } /* charid is now 0 if we dont know the char read, or nonzero if we do */ } ###### accept_state = TRIE_NODENUM( state ); ###### if ( !trie->states[ accept_state ].wordnum ) { /* we havent inserted this word into the structure yet. */ ###### trie->states[ accept_state ].wordnum = ++curword; DEBUG_r({ /* store the word for dumping */ SV* tmp = newSVpvn( STRING( noper ), STR_LEN( noper ) ); if ( UTF ) SvUTF8_on( tmp ); av_push( trie->words, tmp ); ###### }); } else { /* Its a dupe. So ignore it. */ } } /* end second pass */ DEBUG_TRIE_COMPILE_MORE_r({ /* print out the table precompression so that we can do a visual check that they are identical. */ U32 state; U16 charid; PerlIO_printf( Perl_debug_log, "\nChar : " ); for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { SV **tmp = av_fetch( trie->revcharmap, charid, 0); if ( tmp ) { PerlIO_printf( Perl_debug_log, "%4.4s ", SvPV_nolen_const( *tmp ) ); } } PerlIO_printf( Perl_debug_log, "\nState+-" ); for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) { PerlIO_printf( Perl_debug_log, "%4s-", "----" ); } PerlIO_printf( Perl_debug_log, "\n" ); for( state=1 ; state < next_alloc ; state += trie->uniquecharcount ) { PerlIO_printf( Perl_debug_log, "%04"UVXf" : ", (UV)TRIE_NODENUM( state ) ); for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) { PerlIO_printf( Perl_debug_log, "%04"UVXf" ", (UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next ) ); } if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) { PerlIO_printf( Perl_debug_log, " (%04"UVXf")\n", (UV)trie->trans[ state ].check ); } else { PerlIO_printf( Perl_debug_log, " (%04"UVXf") W%04X\n", (UV)trie->trans[ state ].check, trie->states[ TRIE_NODENUM( state ) ].wordnum ); } } PerlIO_printf( Perl_debug_log, "\n\n" ); ###### }); { /* * Inplace compress the table.* For sparse data sets the table constructed by the trie algorithm will be mostly 0/FAIL transitions or to put it another way mostly empty. (Note that leaf nodes will not contain any transitions.) This algorithm compresses the tables by eliminating most such transitions, at the cost of a modest bit of extra work during lookup: - Each states[] entry contains a .base field which indicates the index in the state[] array wheres its transition data is stored. - If .base is 0 there are no valid transitions from that node. - If .base is nonzero then charid is added to it to find an entry in the trans array. -If trans[states[state].base+charid].check!=state then the transition is taken to be a 0/Fail transition. Thus if there are fail transitions at the front of the node then the .base offset will point somewhere inside the previous nodes data (or maybe even into a node even earlier), but the .check field determines if the transition is valid. The following process inplace converts the table to the compressed table: We first do not compress the root node 1,and mark its all its .check pointers as 1 and set its .base pointer as 1 as well. This allows to do a DFA construction from the compressed table later, and ensures that any .base pointers we calculate later are greater than 0. - We set 'pos' to indicate the first entry of the second node. - We then iterate over the columns of the node, finding the first and last used entry at l and m. We then copy l..m into pos..(pos+m-l), and set the .check pointers accordingly, and advance pos appropriately and repreat for the next node. Note that when we copy the next pointers we have to convert them from the original NODEIDX form to NODENUM form as the former is not valid post compression. - If a node has no transitions used we mark its base as 0 and do not advance the pos pointer. - If a node only has one transition we use a second pointer into the structure to fill in allocated fail transitions from other states. This pointer is independent of the main pointer and scans forward looking for null transitions that are allocated to a state. When it finds one it writes the single transition into the "hole". If the pointer doesnt find one the single transition is appeneded as normal. - Once compressed we can Renew/realloc the structures to release the excess space. See "Table-Compression Methods" in sec 3.9 of the Red Dragon, specifically Fig 3.47 and the associated pseudocode. demq */ ###### const U32 laststate = TRIE_NODENUM( next_alloc ); ###### U32 state, charid; ###### U32 pos = 0, zp=0; ###### trie->laststate = laststate; ###### for ( state = 1 ; state < laststate ; state++ ) { ###### U8 flag = 0; ###### const U32 stateidx = TRIE_NODEIDX( state ); ###### const U32 o_used = trie->trans[ stateidx ].check; ###### U32 used = trie->trans[ stateidx ].check; ###### trie->trans[ stateidx ].check = 0; ###### for ( charid = 0 ; used && charid < trie->uniquecharcount ; charid++ ) { ###### if ( flag || trie->trans[ stateidx + charid ].next ) { ###### if ( trie->trans[ stateidx + charid ].next ) { ###### if (o_used == 1) { ###### for ( ; zp < pos ; zp++ ) { ###### if ( ! trie->trans[ zp ].next ) { ###### break; } } ###### trie->states[ state ].trans.base = zp + trie->uniquecharcount - charid ; ###### trie->trans[ zp ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next ); ###### trie->trans[ zp ].check = state; ###### if ( ++zp > pos ) pos = zp; ###### break; } ###### used--; } ###### if ( !flag ) { ###### flag = 1; ###### trie->states[ state ].trans.base = pos + trie->uniquecharcount - charid ; } ###### trie->trans[ pos ].next = SAFE_TRIE_NODENUM( trie->trans[ stateidx + charid ].next ); ###### trie->trans[ pos ].check = state; ###### pos++; } } } ###### trie->lasttrans = pos + 1; ###### Renew( trie->states, laststate + 1, reg_trie_state); DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log, " Alloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n", (int)( ( trie->charcount + 1 ) * trie->uniquecharcount + 1 ), (IV)next_alloc, (IV)pos, ( ( next_alloc - pos ) * 100 ) / (double)next_alloc ); ###### ); } /* end table compress */ } /* resize the trans array to remove unused space */ ###### Renew( trie->trans, trie->lasttrans, reg_trie_trans); DEBUG_TRIE_COMPILE_r({ U32 state; /* Now we print it out again, in a slightly different form as there is additional info we want to be able to see when its compressed. They are close enough for visual comparison though. */ PerlIO_printf( Perl_debug_log, "\nChar : %-6s%-6s%-4s ","Match","Base","Ofs" ); for( state = 0 ; state < trie->uniquecharcount ; state++ ) { SV **tmp = av_fetch( trie->revcharmap, state, 0); if ( tmp ) { PerlIO_printf( Perl_debug_log, "%4.4s ", SvPV_nolen_const( *tmp ) ); } } PerlIO_printf( Perl_debug_log, "\n-----:-----------------------"); for( state = 0 ; state < trie->uniquecharcount ; state++ ) PerlIO_printf( Perl_debug_log, "-----"); PerlIO_printf( Perl_debug_log, "\n"); for( state = 1 ; state < trie->laststate ; state++ ) { const U32 base = trie->states[ state ].trans.base; PerlIO_printf( Perl_debug_log, "#%04"UVXf" ", (UV)state); if ( trie->states[ state ].wordnum ) { PerlIO_printf( Perl_debug_log, " W%04X", trie->states[ state ].wordnum ); } else { PerlIO_printf( Perl_debug_log, "%6s", "" ); } PerlIO_printf( Perl_debug_log, " @%04"UVXf" ", (UV)base ); if ( base ) { U32 ofs = 0; while( ( base + ofs < trie->uniquecharcount ) || ( base + ofs - trie->uniquecharcount < trie->lasttrans && trie->trans[ base + ofs - trie->uniquecharcount ].check != state)) ofs++; PerlIO_printf( Perl_debug_log, "+%02"UVXf"[ ", (UV)ofs); for ( ofs = 0 ; ofs < trie->uniquecharcount ; ofs++ ) { if ( ( base + ofs >= trie->uniquecharcount ) && ( base + ofs - trie->uniquecharcount < trie->lasttrans ) && trie->trans[ base + ofs - trie->uniquecharcount ].check == state ) { PerlIO_printf( Perl_debug_log, "%04"UVXf" ", (UV)trie->trans[ base + ofs - trie->uniquecharcount ].next ); } else { PerlIO_printf( Perl_debug_log, "%4s "," 0" ); } } PerlIO_printf( Perl_debug_log, "]"); } PerlIO_printf( Perl_debug_log, "\n" ); } ###### }); { /* now finally we "stitch in" the new TRIE node This means we convert either the first branch or the first Exact, depending on whether the thing following (in 'last') is a branch or not and whther first is the startbranch (ie is it a sub part of the alternation or is it the whole thing.) Assuming its a sub part we conver the EXACT otherwise we convert the whole branch sequence, including the first. */ ###### regnode *convert; ###### if ( first == startbranch && OP( last ) != BRANCH ) { ###### convert = first; } else { ###### convert = NEXTOPER( first ); ###### NEXT_OFF( first ) = (U16)(last - first); } ###### OP( convert ) = TRIE + (U8)( flags - EXACT ); ###### NEXT_OFF( convert ) = (U16)(tail - convert); ###### ARG_SET( convert, data_slot ); /* tells us if we need to handle accept buffers specially */ ###### convert->flags = ( RExC_seen_evals ? 1 : 0 ); /* needed for dumping*/ DEBUG_r({ regnode *optimize = convert + NODE_STEP_REGNODE + regarglen[ TRIE ]; /* We now need to mark all of the space originally used by the branches as optimized away. This keeps the dumpuntil from throwing a wobbly as it doesnt use regnext() to traverse the opcodes. */ while( optimize < last ) { OP( optimize ) = OPTIMIZED; optimize++; } ###### }); } /* end node insert */ ###### return 1; } /* * There are strange code-generation bugs caused on sparc64 by gcc-2.95.2. * These need to be revisited when a newer toolchain becomes available. */ #if defined(__sparc64__) && defined(__GNUC__) # if __GNUC__ < 2 || (__GNUC__ == 2 && __GNUC_MINOR__ < 96) # undef SPARC64_GCC_WORKAROUND # define SPARC64_GCC_WORKAROUND 1 # endif #endif /* REx optimizer. Converts nodes into quickier variants "in place". Finds fixed substrings. */ /* Stops at toplevel WHILEM as well as at "last". At end *scanp is set to the position after last scanned or to NULL. */ STATIC I32 S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp, I32 *deltap, regnode *last, scan_data_t *data, U32 flags, U32 depth) /* scanp: Start here (read-write). */ /* deltap: Write maxlen-minlen here. */ /* last: Stop before this one. */ ###### { ###### I32 min = 0, pars = 0, code; ###### regnode *scan = *scanp, *next; ###### I32 delta = 0; ###### int is_inf = (flags & SCF_DO_SUBSTR) && (data->flags & SF_IS_INF); ###### int is_inf_internal = 0; /* The studied chunk is infinite */ ###### I32 is_par = OP(scan) == OPEN ? ARG(scan) : 0; ###### scan_data_t data_fake; ###### struct regnode_charclass_class and_with; /* Valid if flags & SCF_DO_STCLASS_OR */ ###### SV *re_trie_maxbuff = NULL; ###### GET_RE_DEBUG_FLAGS_DECL; ###### while (scan && OP(scan) != END && scan < last) { /* Peephole optimizer: */ DEBUG_OPTIMISE_r({ SV *mysv=sv_newmortal(); regprop( mysv, scan); PerlIO_printf(Perl_debug_log, "%*speep: %s (0x%08"UVXf")\n", (int)depth*2, "", SvPV_nolen_const(mysv), PTR2UV(scan)); ###### }); ###### if (PL_regkind[(U8)OP(scan)] == EXACT) { /* Merge several consecutive EXACTish nodes into one. */ ###### regnode *n = regnext(scan); ###### U32 stringok = 1; #ifdef DEBUGGING ###### regnode *stop = scan; #endif ###### next = scan + NODE_SZ_STR(scan); /* Skip NOTHING, merge EXACT*. */ ###### while (n && ( PL_regkind[(U8)OP(n)] == NOTHING || (stringok && (OP(n) == OP(scan)))) && NEXT_OFF(n) && NEXT_OFF(scan) + NEXT_OFF(n) < I16_MAX) { ###### if (OP(n) == TAIL || n > next) ###### stringok = 0; ###### if (PL_regkind[(U8)OP(n)] == NOTHING) { ###### NEXT_OFF(scan) += NEXT_OFF(n); ###### next = n + NODE_STEP_REGNODE; #ifdef DEBUGGING ###### if (stringok) ###### stop = n; #endif ###### n = regnext(n); } ###### else if (stringok) { ###### const int oldl = STR_LEN(scan); ###### regnode *nnext = regnext(n); ###### if (oldl + STR_LEN(n) > U8_MAX) ###### break; ###### NEXT_OFF(scan) += NEXT_OFF(n); ###### STR_LEN(scan) += STR_LEN(n); ###### next = n + NODE_SZ_STR(n); /* Now we can overwrite *n : */ ###### Move(STRING(n), STRING(scan) + oldl, STR_LEN(n), char); #ifdef DEBUGGING ###### stop = next - 1; #endif ###### n = nnext; } } ###### if (UTF && ( OP(scan) == EXACTF ) && ( STR_LEN(scan) >= 6 ) ) { /* Two problematic code points in Unicode casefolding of EXACT nodes: U+0390 - GREEK SMALL LETTER IOTA WITH DIALYTIKA AND TONOS U+03B0 - GREEK SMALL LETTER UPSILON WITH DIALYTIKA AND TONOS which casefold to Unicode UTF-8 U+03B9 U+0308 U+0301 0xCE 0xB9 0xCC 0x88 0xCC 0x81 U+03C5 U+0308 U+0301 0xCF 0x85 0xCC 0x88 0xCC 0x81 This means that in case-insensitive matching (or "loose matching", as Unicode calls it), an EXACTF of length six (the UTF-8 encoded byte length of the above casefolded versions) can match a target string of length two (the byte length of UTF-8 encoded U+0390 or U+03B0). This would rather mess up the minimum length computation. What we'll do is to look for the tail four bytes, and then peek at the preceding two bytes to see whether we need to decrease the minimum length by four (six minus two). Thanks to the design of UTF-8, there cannot be false matches: A sequence of valid UTF-8 bytes cannot be a subsequence of another valid sequence of UTF-8 bytes. */ ###### char *s0 = STRING(scan), *s, *t; ###### char *s1 = s0 + STR_LEN(scan) - 1, *s2 = s1 - 4; ###### const char * const t0 = "\xcc\x88\xcc\x81"; ###### const char * const t1 = t0 + 3; ###### for (s = s0 + 2; s < s2 && (t = ninstr(s, s1, t0, t1)); s = t + 4) { ###### if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) || ((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF)) ###### min -= 4; } } #ifdef DEBUGGING /* Allow dumping */ ###### n = scan + NODE_SZ_STR(scan); ###### while (n <= stop) { ###### if (PL_regkind[(U8)OP(n)] != NOTHING || OP(n) == NOTHING) { ###### OP(n) = OPTIMIZED; ###### NEXT_OFF(n) = 0; } ###### n++; } #endif } /* Follow the next-chain of the current node and optimize away all the NOTHINGs from it. */ ###### if (OP(scan) != CURLYX) { ###### const int max = (reg_off_by_arg[OP(scan)] ? I32_MAX /* I32 may be smaller than U16 on CRAYs! */ ###### : (I32_MAX < U16_MAX ? I32_MAX : U16_MAX)); ###### int off = (reg_off_by_arg[OP(scan)] ? ARG(scan) : NEXT_OFF(scan)); ###### int noff; ###### regnode *n = scan; /* Skip NOTHING and LONGJMP. */ ###### while ((n = regnext(n)) && ((PL_regkind[(U8)OP(n)] == NOTHING && (noff = NEXT_OFF(n))) || ((OP(n) == LONGJMP) && (noff = ARG(n)))) && off + noff < max) ###### off += noff; ###### if (reg_off_by_arg[OP(scan)]) ###### ARG(scan) = off; else ###### NEXT_OFF(scan) = off; } /* The principal pseudo-switch. Cannot be a switch, since we look into several different things. */ ###### if (OP(scan) == BRANCH || OP(scan) == BRANCHJ || OP(scan) == IFTHEN || OP(scan) == SUSPEND) { ###### next = regnext(scan); ###### code = OP(scan); /* demq: the op(next)==code check is to see if we have "branch-branch" AFAICT */ ###### if (OP(next) == code || code == IFTHEN || code == SUSPEND) { ###### I32 max1 = 0, min1 = I32_MAX, num = 0; ###### struct regnode_charclass_class accum; ###### regnode *startbranch=scan; ###### if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */ ###### scan_commit(pRExC_state, data); /* Cannot merge strings after this. */ ###### if (flags & SCF_DO_STCLASS) ###### cl_init_zero(pRExC_state, &accum); ###### while (OP(scan) == code) { ###### I32 deltanext, minnext, f = 0, fake; ###### struct regnode_charclass_class this_class; ###### num++; ###### data_fake.flags = 0; ###### if (data) { ###### data_fake.whilem_c = data->whilem_c; ###### data_fake.last_closep = data->last_closep; } else ###### data_fake.last_closep = &fake; ###### next = regnext(scan); ###### scan = NEXTOPER(scan); ###### if (code != BRANCH) ###### scan = NEXTOPER(scan); ###### if (flags & SCF_DO_STCLASS) { ###### cl_init(pRExC_state, &this_class); ###### data_fake.start_class = &this_class; ###### f = SCF_DO_STCLASS_AND; } ###### if (flags & SCF_WHILEM_VISITED_POS) ###### f |= SCF_WHILEM_VISITED_POS; /* we suppose the run is continuous, last=next...*/ ###### minnext = study_chunk(pRExC_state, &scan, &deltanext, next, &data_fake, f,depth+1); ###### if (min1 > minnext) ###### min1 = minnext; ###### if (max1 < minnext + deltanext) ###### max1 = minnext + deltanext; ###### if (deltanext == I32_MAX) ###### is_inf = is_inf_internal = 1; ###### scan = next; ###### if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) ###### pars++; ###### if (data && (data_fake.flags & SF_HAS_EVAL)) ###### data->flags |= SF_HAS_EVAL; ###### if (data) ###### data->whilem_c = data_fake.whilem_c; ###### if (flags & SCF_DO_STCLASS) ###### cl_or(pRExC_state, &accum, &this_class); ###### if (code == SUSPEND) ###### break; } ###### if (code == IFTHEN && num < 2) /* Empty ELSE branch */ ###### min1 = 0; ###### if (flags & SCF_DO_SUBSTR) { ###### data->pos_min += min1; ###### data->pos_delta += max1 - min1; ###### if (max1 != min1 || is_inf) ###### data->longest = &(data->longest_float); } ###### min += min1; ###### delta += max1 - min1; ###### if (flags & SCF_DO_STCLASS_OR) { ###### cl_or(pRExC_state, data->start_class, &accum); ###### if (min1) { ###### cl_and(data->start_class, &and_with); ###### flags &= ~SCF_DO_STCLASS; } } ###### else if (flags & SCF_DO_STCLASS_AND) { ###### if (min1) { ###### cl_and(data->start_class, &accum); ###### flags &= ~SCF_DO_STCLASS; } else { /* Switch to OR mode: cache the old value of * data->start_class */ StructCopy(data->start_class, &and_with, ###### struct regnode_charclass_class); ###### flags &= ~SCF_DO_STCLASS_AND; StructCopy(&accum, data->start_class, ###### struct regnode_charclass_class); ###### flags |= SCF_DO_STCLASS_OR; ###### data->start_class->flags |= ANYOF_EOS; } } /* demq. Assuming this was/is a branch we are dealing with: 'scan' now points at the item that follows the branch sequence, whatever it is. We now start at the beginning of the sequence and look for subsequences of BRANCH->EXACT=>X BRANCH->EXACT=>X which would be constructed from a pattern like /A|LIST|OF|WORDS/ If we can find such a subseqence we need to turn the first element into a trie and then add the subsequent branch exact strings to the trie. We have two cases 1. patterns where the whole set of branch can be converted to a trie, 2. patterns where only a subset of the alternations can be converted to a trie. In case 1 we can replace the whole set with a single regop for the trie. In case 2 we need to keep the start and end branchs so 'BRANCH EXACT; BRANCH EXACT; BRANCH X' becomes BRANCH TRIE; BRANCH X; Hypthetically when we know the regex isnt anchored we can turn a case 1 into a DFA and let it rip... Every time it finds a match it would just call its tail, no WHILEM/CURLY needed. */ ###### if (DO_TRIE) { ###### if (!re_trie_maxbuff) { ###### re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1); ###### if (!SvIOK(re_trie_maxbuff)) ###### sv_setiv(re_trie_maxbuff, RE_TRIE_MAXBUF_INIT); } ###### if ( SvIV(re_trie_maxbuff)>=0 && OP( startbranch )==BRANCH ) { ###### regnode *cur; ###### regnode *first = (regnode *)NULL; ###### regnode *last = (regnode *)NULL; ###### regnode *tail = scan; ###### U8 optype = 0; ###### U32 count=0; #ifdef DEBUGGING ###### SV *mysv = sv_newmortal(); /* for dumping */ #endif /* var tail is used because there may be a TAIL regop in the way. Ie, the exacts will point to the thing following the TAIL, but the last branch will point at the TAIL. So we advance tail. If we have nested (?:) we may have to move through several tails. */ ###### while ( OP( tail ) == TAIL ) { /* this is the TAIL generated by (?:) */ ###### tail = regnext( tail ); } DEBUG_OPTIMISE_r({ regprop( mysv, tail ); PerlIO_printf( Perl_debug_log, "%*s%s%s%s\n", (int)depth * 2 + 2, "", "Tail node is:", SvPV_nolen_const( mysv ), (RExC_seen_evals) ? "[EVAL]" : "" ); ###### }); /* step through the branches, cur represents each branch, noper is the first thing to be matched as part of that branch and noper_next is the regnext() of that node. if noper is an EXACT and noper_next is the same as scan (our current position in the regex) then the EXACT branch is a possible optimization target. Once we have two or more consequetive such branches we can create a trie of the EXACT's contents and stich it in place. If the sequence represents all of the branches we eliminate the whole thing and replace it with a single TRIE. If it is a subsequence then we need to stitch it in. This means the first branch has to remain, and needs to be repointed at the item on the branch chain following the last branch optimized. This could be either a BRANCH, in which case the subsequence is internal, or it could be the item following the branch sequence in which case the subsequence is at the end. */ /* dont use tail as the end marker for this traverse */ ###### for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) { ###### regnode *noper = NEXTOPER( cur ); ###### regnode *noper_next = regnext( noper ); DEBUG_OPTIMISE_r({ regprop( mysv, cur); PerlIO_printf( Perl_debug_log, "%*s%s", (int)depth * 2 + 2," ", SvPV_nolen_const( mysv ) ); regprop( mysv, noper); PerlIO_printf( Perl_debug_log, " -> %s", SvPV_nolen_const(mysv)); if ( noper_next ) { regprop( mysv, noper_next ); PerlIO_printf( Perl_debug_log,"\t=> %s\t", SvPV_nolen_const(mysv)); } PerlIO_printf( Perl_debug_log, "0x%p,0x%p,0x%p)\n", first, last, cur ); ###### }); ###### if ( ( first ? OP( noper ) == optype : PL_regkind[ (U8)OP( noper ) ] == EXACT ) && noper_next == tail && count %s\n", SvPV_nolen_const( mysv ) ); } ###### ); ###### last = cur; DEBUG_OPTIMISE_r({ regprop( mysv, cur); PerlIO_printf( Perl_debug_log, "%*s%s", (int)depth * 2 + 2, "N:", SvPV_nolen_const( mysv ) ); regprop( mysv, noper ); PerlIO_printf( Perl_debug_log, " -> %s\n", SvPV_nolen_const( mysv ) ); ###### }); } } else { ###### if ( last ) { DEBUG_OPTIMISE_r( PerlIO_printf( Perl_debug_log, "%*s%s\n", (int)depth * 2 + 2, "E:", "**END**" ); ###### ); ###### make_trie( pRExC_state, startbranch, first, cur, tail, optype ); } ###### if ( PL_regkind[ (U8)OP( noper ) ] == EXACT && noper_next == tail ) { ###### count = 1; ###### first = cur; ###### optype = OP( noper ); } else { ###### count = 0; ###### first = NULL; ###### optype = 0; } ###### last = NULL; } } DEBUG_OPTIMISE_r({ regprop( mysv, cur); PerlIO_printf( Perl_debug_log, "%*s%s\t(0x%p,0x%p,0x%p)\n", (int)depth * 2 + 2, " ", SvPV_nolen_const( mysv ), first, last, cur); ###### }); ###### if ( last ) { DEBUG_OPTIMISE_r( PerlIO_printf( Perl_debug_log, "%*s%s\n", (int)depth * 2 + 2, "E:", "==END==" ); ###### ); ###### make_trie( pRExC_state, startbranch, first, scan, tail, optype ); } } } } ###### else if ( code == BRANCHJ ) { /* single branch is optimized. */ ###### scan = NEXTOPER(NEXTOPER(scan)); } else /* single branch is optimized. */ ###### scan = NEXTOPER(scan); ###### continue; } ###### else if (OP(scan) == EXACT) { ###### I32 l = STR_LEN(scan); ###### UV uc = *((U8*)STRING(scan)); ###### if (UTF) { ###### const U8 * const s = (U8*)STRING(scan); ###### l = utf8_length(s, s + l); ###### uc = utf8_to_uvchr(s, NULL); } ###### min += l; ###### if (flags & SCF_DO_SUBSTR) { /* Update longest substr. */ /* The code below prefers earlier match for fixed offset, later match for variable offset. */ ###### if (data->last_end == -1) { /* Update the start info. */ ###### data->last_start_min = data->pos_min; ###### data->last_start_max = is_inf ? I32_MAX : data->pos_min + data->pos_delta; } ###### sv_catpvn(data->last_found, STRING(scan), STR_LEN(scan)); { ###### SV * sv = data->last_found; ###### MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ? ###### mg_find(sv, PERL_MAGIC_utf8) : NULL; ###### if (mg && mg->mg_len >= 0) ###### mg->mg_len += utf8_length((U8*)STRING(scan), (U8*)STRING(scan)+STR_LEN(scan)); } ###### if (UTF) ###### SvUTF8_on(data->last_found); ###### data->last_end = data->pos_min + l; ###### data->pos_min += l; /* As in the first entry. */ ###### data->flags &= ~SF_BEFORE_EOL; } ###### if (flags & SCF_DO_STCLASS_AND) { /* Check whether it is compatible with what we know already! */ ###### int compat = 1; ###### if (uc >= 0x100 || (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE)) && !ANYOF_BITMAP_TEST(data->start_class, uc) && (!(data->start_class->flags & ANYOF_FOLD) || !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc]))) ) ###### compat = 0; ###### ANYOF_CLASS_ZERO(data->start_class); ###### ANYOF_BITMAP_ZERO(data->start_class); ###### if (compat) ###### ANYOF_BITMAP_SET(data->start_class, uc); ###### data->start_class->flags &= ~ANYOF_EOS; ###### if (uc < 0x100) ###### data->start_class->flags &= ~ANYOF_UNICODE_ALL; } ###### else if (flags & SCF_DO_STCLASS_OR) { /* false positive possible if the class is case-folded */ ###### if (uc < 0x100) ###### ANYOF_BITMAP_SET(data->start_class, uc); else ###### data->start_class->flags |= ANYOF_UNICODE_ALL; ###### data->start_class->flags &= ~ANYOF_EOS; ###### cl_and(data->start_class, &and_with); } ###### flags &= ~SCF_DO_STCLASS; } ###### else if (PL_regkind[(U8)OP(scan)] == EXACT) { /* But OP != EXACT! */ ###### I32 l = STR_LEN(scan); ###### UV uc = *((U8*)STRING(scan)); /* Search for fixed substrings supports EXACT only. */ ###### if (flags & SCF_DO_SUBSTR) ###### scan_commit(pRExC_state, data); ###### if (UTF) { ###### U8 *s = (U8 *)STRING(scan); ###### l = utf8_length(s, s + l); ###### uc = utf8_to_uvchr(s, NULL); } ###### min += l; ###### if (data && (flags & SCF_DO_SUBSTR)) ###### data->pos_min += l; ###### if (flags & SCF_DO_STCLASS_AND) { /* Check whether it is compatible with what we know already! */ ###### int compat = 1; ###### if (uc >= 0x100 || (!(data->start_class->flags & (ANYOF_CLASS | ANYOF_LOCALE)) && !ANYOF_BITMAP_TEST(data->start_class, uc) && !ANYOF_BITMAP_TEST(data->start_class, PL_fold[uc]))) ###### compat = 0; ###### ANYOF_CLASS_ZERO(data->start_class); ###### ANYOF_BITMAP_ZERO(data->start_class); ###### if (compat) { ###### ANYOF_BITMAP_SET(data->start_class, uc); ###### data->start_class->flags &= ~ANYOF_EOS; ###### data->start_class->flags |= ANYOF_FOLD; ###### if (OP(scan) == EXACTFL) ###### data->start_class->flags |= ANYOF_LOCALE; } } ###### else if (flags & SCF_DO_STCLASS_OR) { ###### if (data->start_class->flags & ANYOF_FOLD) { /* false positive possible if the class is case-folded. Assume that the locale settings are the same... */ ###### if (uc < 0x100) ###### ANYOF_BITMAP_SET(data->start_class, uc); ###### data->start_class->flags &= ~ANYOF_EOS; } ###### cl_and(data->start_class, &and_with); } ###### flags &= ~SCF_DO_STCLASS; } ###### else if (strchr((const char*)PL_varies,OP(scan))) { ###### I32 mincount, maxcount, minnext, deltanext, fl = 0; ###### I32 f = flags, pos_before = 0; ###### regnode *oscan = scan; ###### struct regnode_charclass_class this_class; ###### struct regnode_charclass_class *oclass = NULL; ###### I32 next_is_eval = 0; ###### switch (PL_regkind[(U8)OP(scan)]) { case WHILEM: /* End of (?:...)* . */ ###### scan = NEXTOPER(scan); ###### goto finish; case PLUS: ###### if (flags & (SCF_DO_SUBSTR | SCF_DO_STCLASS)) { ###### next = NEXTOPER(scan); ###### if (OP(next) == EXACT || (flags & SCF_DO_STCLASS)) { ###### mincount = 1; ###### maxcount = REG_INFTY; ###### next = regnext(scan); ###### scan = NEXTOPER(scan); ###### goto do_curly; } } ###### if (flags & SCF_DO_SUBSTR) ###### data->pos_min++; ###### min++; /* Fall through. */ case STAR: ###### if (flags & SCF_DO_STCLASS) { ###### mincount = 0; ###### maxcount = REG_INFTY; ###### next = regnext(scan); ###### scan = NEXTOPER(scan); ###### goto do_curly; } ###### is_inf = is_inf_internal = 1; ###### scan = regnext(scan); ###### if (flags & SCF_DO_SUBSTR) { ###### scan_commit(pRExC_state, data); /* Cannot extend fixed substrings */ ###### data->longest = &(data->longest_float); } ###### goto optimize_curly_tail; case CURLY: ###### mincount = ARG1(scan); ###### maxcount = ARG2(scan); ###### next = regnext(scan); ###### if (OP(scan) == CURLYX) { ###### I32 lp = (data ? *(data->last_closep) : 0); ###### scan->flags = ((lp <= U8_MAX) ? (U8)lp : U8_MAX); } ###### scan = NEXTOPER(scan) + EXTRA_STEP_2ARGS; ###### next_is_eval = (OP(scan) == EVAL); do_curly: ###### if (flags & SCF_DO_SUBSTR) { ###### if (mincount == 0) scan_commit(pRExC_state,data); /* Cannot extend fixed substrings */ ###### pos_before = data->pos_min; } ###### if (data) { ###### fl = data->flags; ###### data->flags &= ~(SF_HAS_PAR|SF_IN_PAR|SF_HAS_EVAL); ###### if (is_inf) ###### data->flags |= SF_IS_INF; } ###### if (flags & SCF_DO_STCLASS) { ###### cl_init(pRExC_state, &this_class); ###### oclass = data->start_class; ###### data->start_class = &this_class; ###### f |= SCF_DO_STCLASS_AND; ###### f &= ~SCF_DO_STCLASS_OR; } /* These are the cases when once a subexpression fails at a particular position, it cannot succeed even after backtracking at the enclosing scope. XXXX what if minimal match and we are at the initial run of {n,m}? */ ###### if ((mincount != maxcount - 1) && (maxcount != REG_INFTY)) ###### f &= ~SCF_WHILEM_VISITED_POS; /* This will finish on WHILEM, setting scan, or on NULL: */ ###### minnext = study_chunk(pRExC_state, &scan, &deltanext, last, data, (mincount == 0 ? (f & ~SCF_DO_SUBSTR) : f),depth+1); ###### if (flags & SCF_DO_STCLASS) ###### data->start_class = oclass; ###### if (mincount == 0 || minnext == 0) { ###### if (flags & SCF_DO_STCLASS_OR) { ###### cl_or(pRExC_state, data->start_class, &this_class); } ###### else if (flags & SCF_DO_STCLASS_AND) { /* Switch to OR mode: cache the old value of * data->start_class */ StructCopy(data->start_class, &and_with, ###### struct regnode_charclass_class); ###### flags &= ~SCF_DO_STCLASS_AND; StructCopy(&this_class, data->start_class, ###### struct regnode_charclass_class); ###### flags |= SCF_DO_STCLASS_OR; ###### data->start_class->flags |= ANYOF_EOS; } } else { /* Non-zero len */ ###### if (flags & SCF_DO_STCLASS_OR) { ###### cl_or(pRExC_state, data->start_class, &this_class); ###### cl_and(data->start_class, &and_with); } ###### else if (flags & SCF_DO_STCLASS_AND) ###### cl_and(data->start_class, &this_class); ###### flags &= ~SCF_DO_STCLASS; } ###### if (!scan) /* It was not CURLYX, but CURLY. */ ###### scan = next; ###### if (ckWARN(WARN_REGEXP) /* ? quantifier ok, except for (?{ ... }) */ && (next_is_eval || !(mincount == 0 && maxcount == 1)) && (minnext == 0) && (deltanext == 0) && data && !(data->flags & (SF_HAS_PAR|SF_IN_PAR)) && maxcount <= REG_INFTY/3) /* Complement check for big count */ { vWARN(RExC_parse, ###### "Quantifier unexpected on zero-length expression"); } ###### min += minnext * mincount; ###### is_inf_internal |= ((maxcount == REG_INFTY && (minnext + deltanext) > 0) || deltanext == I32_MAX); ###### is_inf |= is_inf_internal; ###### delta += (minnext + deltanext) * maxcount - minnext * mincount; /* Try powerful optimization CURLYX => CURLYN. */ ###### if ( OP(oscan) == CURLYX && data && data->flags & SF_IN_PAR && !(data->flags & SF_HAS_EVAL) && !deltanext && minnext == 1 ) { /* Try to optimize to CURLYN. */ ###### regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; ###### regnode *nxt1 = nxt; #ifdef DEBUGGING ###### regnode *nxt2; #endif /* Skip open. */ ###### nxt = regnext(nxt); ###### if (!strchr((const char*)PL_simple,OP(nxt)) && !(PL_regkind[(U8)OP(nxt)] == EXACT && STR_LEN(nxt) == 1)) ###### goto nogo; #ifdef DEBUGGING ###### nxt2 = nxt; #endif ###### nxt = regnext(nxt); ###### if (OP(nxt) != CLOSE) ###### goto nogo; /* Now we know that nxt2 is the only contents: */ ###### oscan->flags = (U8)ARG(nxt); ###### OP(oscan) = CURLYN; ###### OP(nxt1) = NOTHING; /* was OPEN. */ #ifdef DEBUGGING ###### OP(nxt1 + 1) = OPTIMIZED; /* was count. */ ###### NEXT_OFF(nxt1+ 1) = 0; /* just for consistancy. */ ###### NEXT_OFF(nxt2) = 0; /* just for consistancy with CURLY. */ ###### OP(nxt) = OPTIMIZED; /* was CLOSE. */ ###### OP(nxt + 1) = OPTIMIZED; /* was count. */ ###### NEXT_OFF(nxt+ 1) = 0; /* just for consistancy. */ #endif } nogo: /* Try optimization CURLYX => CURLYM. */ ###### if ( OP(oscan) == CURLYX && data && !(data->flags & SF_HAS_PAR) && !(data->flags & SF_HAS_EVAL) && !deltanext /* atom is fixed width */ && minnext != 0 /* CURLYM can't handle zero width */ ) { /* XXXX How to optimize if data == 0? */ /* Optimize to a simpler form. */ ###### regnode *nxt = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN */ ###### regnode *nxt2; ###### OP(oscan) = CURLYM; ###### while ( (nxt2 = regnext(nxt)) /* skip over embedded stuff*/ && (OP(nxt2) != WHILEM)) ###### nxt = nxt2; ###### OP(nxt2) = SUCCEED; /* Whas WHILEM */ /* Need to optimize away parenths. */ ###### if (data->flags & SF_IN_PAR) { /* Set the parenth number. */ ###### regnode *nxt1 = NEXTOPER(oscan) + EXTRA_STEP_2ARGS; /* OPEN*/ ###### if (OP(nxt) != CLOSE) ###### FAIL("Panic opt close"); ###### oscan->flags = (U8)ARG(nxt); ###### OP(nxt1) = OPTIMIZED; /* was OPEN. */ ###### OP(nxt) = OPTIMIZED; /* was CLOSE. */ #ifdef DEBUGGING ###### OP(nxt1 + 1) = OPTIMIZED; /* was count. */ ###### OP(nxt + 1) = OPTIMIZED; /* was count. */ ###### NEXT_OFF(nxt1 + 1) = 0; /* just for consistancy. */ ###### NEXT_OFF(nxt + 1) = 0; /* just for consistancy. */ #endif #if 0 while ( nxt1 && (OP(nxt1) != WHILEM)) { regnode *nnxt = regnext(nxt1); if (nnxt == nxt) { if (reg_off_by_arg[OP(nxt1)]) ARG_SET(nxt1, nxt2 - nxt1); else if (nxt2 - nxt1 < U16_MAX) NEXT_OFF(nxt1) = nxt2 - nxt1; else OP(nxt) = NOTHING; /* Cannot beautify */ } nxt1 = nnxt; } #endif /* Optimize again: */ ###### study_chunk(pRExC_state, &nxt1, &deltanext, nxt, NULL, 0,depth+1); } else ###### oscan->flags = 0; } ###### else if ((OP(oscan) == CURLYX) && (flags & SCF_WHILEM_VISITED_POS) /* See the comment on a similar expression above. However, this time it not a subexpression we care about, but the expression itself. */ && (maxcount == REG_INFTY) && data && ++data->whilem_c < 16) { /* This stays as CURLYX, we can put the count/of pair. */ /* Find WHILEM (as in regexec.c) */ ###### regnode *nxt = oscan + NEXT_OFF(oscan); ###### if (OP(PREVOPER(nxt)) == NOTHING) /* LONGJMP */ ###### nxt += ARG(nxt); ###### PREVOPER(nxt)->flags = (U8)(data->whilem_c | (RExC_whilem_seen << 4)); /* On WHILEM */ } ###### if (data && fl & (SF_HAS_PAR|SF_IN_PAR)) ###### pars++; ###### if (flags & SCF_DO_SUBSTR) { ###### SV *last_str = Nullsv; ###### int counted = mincount != 0; ###### if (data->last_end > 0 && mincount != 0) { /* Ends with a string. */ #if defined(SPARC64_GCC_WORKAROUND) I32 b = 0; STRLEN l = 0; const char *s = NULL; I32 old = 0; if (pos_before >= data->last_start_min) b = pos_before; else b = data->last_start_min; l = 0; s = SvPV_const(data->last_found, l); old = b - data->last_start_min; #else ###### I32 b = pos_before >= data->last_start_min ###### ? pos_before : data->last_start_min; ###### STRLEN l; ###### const char *s = SvPV_const(data->last_found, l); ###### I32 old = b - data->last_start_min; #endif ###### if (UTF) ###### old = utf8_hop((U8*)s, old) - (U8*)s; ###### l -= old; /* Get the added string: */ ###### last_str = newSVpvn(s + old, l); ###### if (UTF) ###### SvUTF8_on(last_str); ###### if (deltanext == 0 && pos_before == b) { /* What was added is a constant string */ ###### if (mincount > 1) { ###### SvGROW(last_str, (mincount * l) + 1); ###### repeatcpy(SvPVX(last_str) + l, SvPVX_const(last_str), l, mincount - 1); ###### SvCUR_set(last_str, SvCUR(last_str) * mincount); /* Add additional parts. */ SvCUR_set(data->last_found, ###### SvCUR(data->last_found) - l); ###### sv_catsv(data->last_found, last_str); { ###### SV * sv = data->last_found; ###### MAGIC *mg = SvUTF8(sv) && SvMAGICAL(sv) ? ###### mg_find(sv, PERL_MAGIC_utf8) : NULL; ###### if (mg && mg->mg_len >= 0) ###### mg->mg_len += CHR_SVLEN(last_str); } ###### data->last_end += l * (mincount - 1); } } else { /* start offset must point into the last copy */ ###### data->last_start_min += minnext * (mincount - 1); ###### data->last_start_max += is_inf ? I32_MAX : (maxcount - 1) * (minnext + data->pos_delta); } } /* It is counted once already... */ ###### data->pos_min += minnext * (mincount - counted); ###### data->pos_delta += - counted * deltanext + (minnext + deltanext) * maxcount - minnext * mincount; ###### if (mincount != maxcount) { /* Cannot extend fixed substrings found inside the group. */ ###### scan_commit(pRExC_state,data); ###### if (mincount && last_str) { ###### sv_setsv(data->last_found, last_str); ###### data->last_end = data->pos_min; ###### data->last_start_min = data->pos_min - CHR_SVLEN(last_str); ###### data->last_start_max = is_inf ? I32_MAX : data->pos_min + data->pos_delta - CHR_SVLEN(last_str); } ###### data->longest = &(data->longest_float); } ###### SvREFCNT_dec(last_str); } ###### if (data && (fl & SF_HAS_EVAL)) ###### data->flags |= SF_HAS_EVAL; optimize_curly_tail: ###### if (OP(oscan) != CURLYX) { ###### while (PL_regkind[(U8)OP(next = regnext(oscan))] == NOTHING && NEXT_OFF(next)) ###### NEXT_OFF(oscan) += NEXT_OFF(next); } ###### continue; default: /* REF and CLUMP only? */ ###### if (flags & SCF_DO_SUBSTR) { ###### scan_commit(pRExC_state,data); /* Cannot expect anything... */ ###### data->longest = &(data->longest_float); } ###### is_inf = is_inf_internal = 1; ###### if (flags & SCF_DO_STCLASS_OR) ###### cl_anything(pRExC_state, data->start_class); ###### flags &= ~SCF_DO_STCLASS; ###### break; } } ###### else if (strchr((const char*)PL_simple,OP(scan))) { ###### int value = 0; ###### if (flags & SCF_DO_SUBSTR) { ###### scan_commit(pRExC_state,data); ###### data->pos_min++; } ###### min++; ###### if (flags & SCF_DO_STCLASS) { ###### data->start_class->flags &= ~ANYOF_EOS; /* No match on empty */ /* Some of the logic below assumes that switching locale on will only add false positives. */ ###### switch (PL_regkind[(U8)OP(scan)]) { case SANY: default: do_default: /* Perl_croak(aTHX_ "panic: unexpected simple REx opcode %d", OP(scan)); */ ###### if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ ###### cl_anything(pRExC_state, data->start_class); ###### break; case REG_ANY: ###### if (OP(scan) == SANY) ###### goto do_default; ###### if (flags & SCF_DO_STCLASS_OR) { /* Everything but \n */ ###### value = (ANYOF_BITMAP_TEST(data->start_class,'\n') || (data->start_class->flags & ANYOF_CLASS)); ###### cl_anything(pRExC_state, data->start_class); } ###### if (flags & SCF_DO_STCLASS_AND || !value) ###### ANYOF_BITMAP_CLEAR(data->start_class,'\n'); ###### break; case ANYOF: ###### if (flags & SCF_DO_STCLASS_AND) ###### cl_and(data->start_class, (struct regnode_charclass_class*)scan); else ###### cl_or(pRExC_state, data->start_class, (struct regnode_charclass_class*)scan); ###### break; case ALNUM: ###### if (flags & SCF_DO_STCLASS_AND) { ###### if (!(data->start_class->flags & ANYOF_LOCALE)) { ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM); ###### for (value = 0; value < 256; value++) ###### if (!isALNUM(value)) ###### ANYOF_BITMAP_CLEAR(data->start_class, value); } } else { ###### if (data->start_class->flags & ANYOF_LOCALE) ###### ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM); else { ###### for (value = 0; value < 256; value++) ###### if (isALNUM(value)) ###### ANYOF_BITMAP_SET(data->start_class, value); } } ###### break; case ALNUML: ###### if (flags & SCF_DO_STCLASS_AND) { ###### if (data->start_class->flags & ANYOF_LOCALE) ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NALNUM); } else { ###### ANYOF_CLASS_SET(data->start_class,ANYOF_ALNUM); ###### data->start_class->flags |= ANYOF_LOCALE; } ###### break; case NALNUM: ###### if (flags & SCF_DO_STCLASS_AND) { ###### if (!(data->start_class->flags & ANYOF_LOCALE)) { ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM); ###### for (value = 0; value < 256; value++) ###### if (isALNUM(value)) ###### ANYOF_BITMAP_CLEAR(data->start_class, value); } } else { ###### if (data->start_class->flags & ANYOF_LOCALE) ###### ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM); else { ###### for (value = 0; value < 256; value++) ###### if (!isALNUM(value)) ###### ANYOF_BITMAP_SET(data->start_class, value); } } ###### break; case NALNUML: ###### if (flags & SCF_DO_STCLASS_AND) { ###### if (data->start_class->flags & ANYOF_LOCALE) ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_ALNUM); } else { ###### data->start_class->flags |= ANYOF_LOCALE; ###### ANYOF_CLASS_SET(data->start_class,ANYOF_NALNUM); } ###### break; case SPACE: ###### if (flags & SCF_DO_STCLASS_AND) { ###### if (!(data->start_class->flags & ANYOF_LOCALE)) { ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE); ###### for (value = 0; value < 256; value++) ###### if (!isSPACE(value)) ###### ANYOF_BITMAP_CLEAR(data->start_class, value); } } else { ###### if (data->start_class->flags & ANYOF_LOCALE) ###### ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE); else { ###### for (value = 0; value < 256; value++) ###### if (isSPACE(value)) ###### ANYOF_BITMAP_SET(data->start_class, value); } } ###### break; case SPACEL: ###### if (flags & SCF_DO_STCLASS_AND) { ###### if (data->start_class->flags & ANYOF_LOCALE) ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NSPACE); } else { ###### data->start_class->flags |= ANYOF_LOCALE; ###### ANYOF_CLASS_SET(data->start_class,ANYOF_SPACE); } ###### break; case NSPACE: ###### if (flags & SCF_DO_STCLASS_AND) { ###### if (!(data->start_class->flags & ANYOF_LOCALE)) { ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE); ###### for (value = 0; value < 256; value++) ###### if (isSPACE(value)) ###### ANYOF_BITMAP_CLEAR(data->start_class, value); } } else { ###### if (data->start_class->flags & ANYOF_LOCALE) ###### ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE); else { ###### for (value = 0; value < 256; value++) ###### if (!isSPACE(value)) ###### ANYOF_BITMAP_SET(data->start_class, value); } } ###### break; case NSPACEL: ###### if (flags & SCF_DO_STCLASS_AND) { ###### if (data->start_class->flags & ANYOF_LOCALE) { ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_SPACE); ###### for (value = 0; value < 256; value++) ###### if (!isSPACE(value)) ###### ANYOF_BITMAP_CLEAR(data->start_class, value); } } else { ###### data->start_class->flags |= ANYOF_LOCALE; ###### ANYOF_CLASS_SET(data->start_class,ANYOF_NSPACE); } ###### break; case DIGIT: ###### if (flags & SCF_DO_STCLASS_AND) { ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_NDIGIT); ###### for (value = 0; value < 256; value++) ###### if (!isDIGIT(value)) ###### ANYOF_BITMAP_CLEAR(data->start_class, value); } else { ###### if (data->start_class->flags & ANYOF_LOCALE) ###### ANYOF_CLASS_SET(data->start_class,ANYOF_DIGIT); else { ###### for (value = 0; value < 256; value++) ###### if (isDIGIT(value)) ###### ANYOF_BITMAP_SET(data->start_class, value); } } ###### break; case NDIGIT: ###### if (flags & SCF_DO_STCLASS_AND) { ###### ANYOF_CLASS_CLEAR(data->start_class,ANYOF_DIGIT); ###### for (value = 0; value < 256; value++) ###### if (isDIGIT(value)) ###### ANYOF_BITMAP_CLEAR(data->start_class, value); } else { ###### if (data->start_class->flags & ANYOF_LOCALE) ###### ANYOF_CLASS_SET(data->start_class,ANYOF_NDIGIT); else { ###### for (value = 0; value < 256; value++) ###### if (!isDIGIT(value)) ###### ANYOF_BITMAP_SET(data->start_class, value); } } ###### break; } ###### if (flags & SCF_DO_STCLASS_OR) ###### cl_and(data->start_class, &and_with); ###### flags &= ~SCF_DO_STCLASS; } } ###### else if (PL_regkind[(U8)OP(scan)] == EOL && flags & SCF_DO_SUBSTR) { ###### data->flags |= (OP(scan) == MEOL ? SF_BEFORE_MEOL : SF_BEFORE_SEOL); } ###### else if ( PL_regkind[(U8)OP(scan)] == BRANCHJ /* Lookbehind, or need to calculate parens/evals/stclass: */ && (scan->flags || data || (flags & SCF_DO_STCLASS)) && (OP(scan) == IFMATCH || OP(scan) == UNLESSM)) { /* Lookahead/lookbehind */ ###### I32 deltanext, minnext, fake = 0; ###### regnode *nscan; ###### struct regnode_charclass_class intrnl; ###### int f = 0; ###### data_fake.flags = 0; ###### if (data) { ###### data_fake.whilem_c = data->whilem_c; ###### data_fake.last_closep = data->last_closep; } else ###### data_fake.last_closep = &fake; ###### if ( flags & SCF_DO_STCLASS && !scan->flags && OP(scan) == IFMATCH ) { /* Lookahead */ ###### cl_init(pRExC_state, &intrnl); ###### data_fake.start_class = &intrnl; ###### f |= SCF_DO_STCLASS_AND; } ###### if (flags & SCF_WHILEM_VISITED_POS) ###### f |= SCF_WHILEM_VISITED_POS; ###### next = regnext(scan); ###### nscan = NEXTOPER(NEXTOPER(scan)); ###### minnext = study_chunk(pRExC_state, &nscan, &deltanext, last, &data_fake, f,depth+1); ###### if (scan->flags) { ###### if (deltanext) { ###### vFAIL("Variable length lookbehind not implemented"); } ###### else if (minnext > U8_MAX) { ###### vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX); } ###### scan->flags = (U8)minnext; } ###### if (data && data_fake.flags & (SF_HAS_PAR|SF_IN_PAR)) ###### pars++; ###### if (data && (data_fake.flags & SF_HAS_EVAL)) ###### data->flags |= SF_HAS_EVAL; ###### if (data) ###### data->whilem_c = data_fake.whilem_c; ###### if (f & SCF_DO_STCLASS_AND) { ###### const int was = (data->start_class->flags & ANYOF_EOS); ###### cl_and(data->start_class, &intrnl); ###### if (was) ###### data->start_class->flags |= ANYOF_EOS; } } ###### else if (OP(scan) == OPEN) { ###### pars++; } ###### else if (OP(scan) == CLOSE) { ###### if ((I32)ARG(scan) == is_par) { ###### next = regnext(scan); ###### if ( next && (OP(next) != WHILEM) && next < last) ###### is_par = 0; /* Disable optimization */ } ###### if (data) ###### *(data->last_closep) = ARG(scan); } ###### else if (OP(scan) == EVAL) { ###### if (data) ###### data->flags |= SF_HAS_EVAL; } ###### else if (OP(scan) == LOGICAL && scan->flags == 2) { /* Embedded follows */ ###### if (flags & SCF_DO_SUBSTR) { ###### scan_commit(pRExC_state,data); ###### data->longest = &(data->longest_float); } ###### is_inf = is_inf_internal = 1; ###### if (flags & SCF_DO_STCLASS_OR) /* Allow everything */ ###### cl_anything(pRExC_state, data->start_class); ###### flags &= ~SCF_DO_STCLASS; } /* Else: zero-length, ignore. */ ###### scan = regnext(scan); } finish: ###### *scanp = scan; ###### *deltap = is_inf_internal ? I32_MAX : delta; ###### if (flags & SCF_DO_SUBSTR && is_inf) ###### data->pos_delta = I32_MAX - data->pos_min; ###### if (is_par > U8_MAX) ###### is_par = 0; ###### if (is_par && pars==1 && data) { ###### data->flags |= SF_IN_PAR; ###### data->flags &= ~SF_HAS_PAR; } ###### else if (pars && data) { ###### data->flags |= SF_HAS_PAR; ###### data->flags &= ~SF_IN_PAR; } ###### if (flags & SCF_DO_STCLASS_OR) ###### cl_and(data->start_class, &and_with); ###### return min; } STATIC I32 S_add_data(pTHX_ RExC_state_t *pRExC_state, I32 n, const char *s) ###### { ###### if (RExC_rx->data) { Renewc(RExC_rx->data, sizeof(*RExC_rx->data) + sizeof(void*) * (RExC_rx->data->count + n - 1), ###### char, struct reg_data); ###### Renew(RExC_rx->data->what, RExC_rx->data->count + n, U8); ###### RExC_rx->data->count += n; } else { Newc(1207, RExC_rx->data, sizeof(*RExC_rx->data) + sizeof(void*) * (n - 1), ###### char, struct reg_data); ###### New(1208, RExC_rx->data->what, n, U8); ###### RExC_rx->data->count = n; } ###### Copy(s, RExC_rx->data->what + RExC_rx->data->count - n, n, U8); ###### return RExC_rx->data->count - n; } void Perl_reginitcolors(pTHX) ###### { ###### const char * const s = PerlEnv_getenv("PERL_RE_COLORS"); ###### if (s) { ###### char *t = savepv(s); ###### int i = 0; ###### PL_colors[0] = t; ###### while (++i < 6) { ###### t = strchr(t, '\t'); ###### if (t) { ###### *t = '\0'; ###### PL_colors[i] = ++t; } else ###### PL_colors[i] = t = (char *)""; } } else { ###### int i = 0; ###### while (i < 6) ###### PL_colors[i++] = (char *)""; } ###### PL_colorset = 1; } /* - pregcomp - compile a regular expression into internal code * * We can't allocate space until we know how big the compiled form will be, * but we can't compile it (and thus know how big it is) until we've got a * place to put the code. So we cheat: we compile it twice, once with code * generation turned off and size counting turned on, and once "for real". * This also means that we don't allocate space until we are sure that the * thing really will compile successfully, and we never have to move the * code and thus invalidate pointers into it. (Note that it has to be in * one piece because free() must be able to free it all.) [NB: not true in perl] * * Beware that the optimization-preparation code in here knows about some * of the structure of the compiled regexp. [I'll say.] */ regexp * Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm) ###### { ###### register regexp *r; ###### regnode *scan; ###### regnode *first; ###### I32 flags; ###### I32 minlen = 0; ###### I32 sawplus = 0; ###### I32 sawopen = 0; ###### scan_data_t data; ###### RExC_state_t RExC_state; ###### RExC_state_t *pRExC_state = &RExC_state; ###### GET_RE_DEBUG_FLAGS_DECL; ###### if (exp == NULL) ###### FAIL("NULL regexp argument"); ###### RExC_utf8 = pm->op_pmdynflags & PMdf_CMP_UTF8; ###### RExC_precomp = exp; ###### DEBUG_r(if (!PL_colorset) reginitcolors()); DEBUG_COMPILE_r({ PerlIO_printf(Perl_debug_log, "%sCompiling REx%s \"%s%*s%s\"\n", PL_colors[4],PL_colors[5],PL_colors[0], (int)(xend - exp), RExC_precomp, PL_colors[1]); ###### }); ###### RExC_flags = pm->op_pmflags; ###### RExC_sawback = 0; ###### RExC_seen = 0; ###### RExC_seen_zerolen = *exp == '^' ? -1 : 0; ###### RExC_seen_evals = 0; ###### RExC_extralen = 0; /* First pass: determine size, legality. */ ###### RExC_parse = exp; ###### RExC_start = exp; ###### RExC_end = xend; ###### RExC_naughty = 0; ###### RExC_npar = 1; ###### RExC_size = 0L; ###### RExC_emit = &PL_regdummy; ###### RExC_whilem_seen = 0; #if 0 /* REGC() is (currently) a NOP at the first pass. * Clever compilers notice this and complain. --jhi */ REGC((U8)REG_MAGIC, (char*)RExC_emit); #endif ###### if (reg(pRExC_state, 0, &flags) == NULL) { ###### RExC_precomp = Nullch; ###### return(NULL); } ###### DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "size %"IVdf" ", (IV)RExC_size)); /* Small enough for pointer-storage convention? If extralen==0, this means that we will not need long jumps. */ ###### if (RExC_size >= 0x10000L && RExC_extralen) ###### RExC_size += RExC_extralen; else ###### RExC_extralen = 0; ###### if (RExC_whilem_seen > 15) ###### RExC_whilem_seen = 15; /* Allocate space and initialize. */ Newc(1001, r, sizeof(regexp) + (unsigned)RExC_size * sizeof(regnode), ###### char, regexp); ###### if (r == NULL) ###### FAIL("Regexp out of space"); #ifdef DEBUGGING /* avoid reading uninitialized memory in DEBUGGING code in study_chunk() */ ###### Zero(r, sizeof(regexp) + (unsigned)RExC_size * sizeof(regnode), char); #endif ###### r->refcnt = 1; ###### r->prelen = xend - exp; ###### r->precomp = savepvn(RExC_precomp, r->prelen); ###### r->subbeg = NULL; #ifdef PERL_OLD_COPY_ON_WRITE r->saved_copy = Nullsv; #endif ###### r->reganch = pm->op_pmflags & PMf_COMPILETIME; ###### r->nparens = RExC_npar - 1; /* set early to validate backrefs */ ###### r->substrs = 0; /* Useful during FAIL. */ ###### r->startp = 0; /* Useful during FAIL. */ ###### r->endp = 0; /* Useful during FAIL. */ ###### Newz(1304, r->offsets, 2*RExC_size+1, U32); /* MJD 20001228 */ ###### if (r->offsets) { ###### r->offsets[0] = RExC_size; } DEBUG_OFFSETS_r(PerlIO_printf(Perl_debug_log, "%s %"UVuf" bytes for offset annotations.\n", r->offsets ? "Got" : "Couldn't get", ###### (UV)((2*RExC_size+1) * sizeof(U32)))); ###### RExC_rx = r; /* Second pass: emit code. */ ###### RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */ ###### RExC_parse = exp; ###### RExC_end = xend; ###### RExC_naughty = 0; ###### RExC_npar = 1; ###### RExC_emit_start = r->program; ###### RExC_emit = r->program; /* Store the count of eval-groups for security checks: */ ###### RExC_emit->next_off = (U16)((RExC_seen_evals > U16_MAX) ? U16_MAX : RExC_seen_evals); ###### REGC((U8)REG_MAGIC, (char*) RExC_emit++); ###### r->data = 0; ###### if (reg(pRExC_state, 0, &flags) == NULL) ###### return(NULL); /* Dig out information for optimizations. */ ###### r->reganch = pm->op_pmflags & PMf_COMPILETIME; /* Again? */ ###### pm->op_pmflags = RExC_flags; ###### if (UTF) ###### r->reganch |= ROPT_UTF8; /* Unicode in it? */ ###### r->regstclass = NULL; ###### if (RExC_naughty >= 10) /* Probably an expensive pattern. */ ###### r->reganch |= ROPT_NAUGHTY; ###### scan = r->program + 1; /* First BRANCH. */ /* XXXX To minimize changes to RE engine we always allocate 3-units-long substrs field. */ ###### Newz(1004, r->substrs, 1, struct reg_substr_data); ###### StructCopy(&zero_scan_data, &data, scan_data_t); /* XXXX Should not we check for something else? Usually it is OPEN1... */ ###### if (OP(scan) != BRANCH) { /* Only one top-level choice. */ ###### I32 fake; ###### STRLEN longest_float_length, longest_fixed_length; ###### struct regnode_charclass_class ch_class; ###### int stclass_flag; ###### I32 last_close = 0; ###### first = scan; /* Skip introductions and multiplicators >= 1. */ ###### while ((OP(first) == OPEN && (sawopen = 1)) || /* An OR of *one* alternative - should not happen now. */ (OP(first) == BRANCH && OP(regnext(first)) != BRANCH) || (OP(first) == PLUS) || (OP(first) == MINMOD) || /* An {n,m} with n>0 */ (PL_regkind[(U8)OP(first)] == CURLY && ARG1(first) > 0) ) { ###### if (OP(first) == PLUS) ###### sawplus = 1; else ###### first += regarglen[(U8)OP(first)]; ###### first = NEXTOPER(first); } /* Starting-point info. */ again: ###### if (PL_regkind[(U8)OP(first)] == EXACT) { ###### if (OP(first) == EXACT) ; /* Empty, get anchored substr later. */ ###### else if ((OP(first) == EXACTF || OP(first) == EXACTFL)) ###### r->regstclass = first; } ###### else if (strchr((const char*)PL_simple,OP(first))) ###### r->regstclass = first; ###### else if (PL_regkind[(U8)OP(first)] == BOUND || PL_regkind[(U8)OP(first)] == NBOUND) ###### r->regstclass = first; ###### else if (PL_regkind[(U8)OP(first)] == BOL) { ###### r->reganch |= (OP(first) == MBOL ? ROPT_ANCH_MBOL : (OP(first) == SBOL ? ROPT_ANCH_SBOL : ROPT_ANCH_BOL)); ###### first = NEXTOPER(first); ###### goto again; } ###### else if (OP(first) == GPOS) { ###### r->reganch |= ROPT_ANCH_GPOS; ###### first = NEXTOPER(first); ###### goto again; } ###### else if (!sawopen && (OP(first) == STAR && PL_regkind[(U8)OP(NEXTOPER(first))] == REG_ANY) && !(r->reganch & ROPT_ANCH) ) { /* turn .* into ^.* with an implied $*=1 */ ###### const int type = (OP(NEXTOPER(first)) == REG_ANY) ? ROPT_ANCH_MBOL ###### : ROPT_ANCH_SBOL; ###### r->reganch |= type | ROPT_IMPLICIT; ###### first = NEXTOPER(first); ###### goto again; } ###### if (sawplus && (!sawopen || !RExC_sawback) && !(RExC_seen & REG_SEEN_EVAL)) /* May examine pos and $& */ /* x+ must match at the 1st pos of run of x's */ ###### r->reganch |= ROPT_SKIP; /* Scan is after the zeroth branch, first is atomic matcher. */ DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n", ###### (IV)(first - scan + 1))); /* * If there's something expensive in the r.e., find the * longest literal string that must appear and make it the * regmust. Resolve ties in favor of later strings, since * the regstart check works with the beginning of the r.e. * and avoiding duplication strengthens checking. Not a * strong reason, but sufficient in the absence of others. * [Now we resolve ties in favor of the earlier string if * it happens that c_offset_min has been invalidated, since the * earlier string may buy us something the later one won't.] */ ###### minlen = 0; ###### data.longest_fixed = newSVpvn("",0); ###### data.longest_float = newSVpvn("",0); ###### data.last_found = newSVpvn("",0); ###### data.longest = &(data.longest_fixed); ###### first = scan; ###### if (!r->regstclass) { ###### cl_init(pRExC_state, &ch_class); ###### data.start_class = &ch_class; ###### stclass_flag = SCF_DO_STCLASS_AND; } else /* XXXX Check for BOUND? */ ###### stclass_flag = 0; ###### data.last_closep = &last_close; ###### minlen = study_chunk(pRExC_state, &first, &fake, scan + RExC_size, /* Up to end */ &data, SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0); ###### if ( RExC_npar == 1 && data.longest == &(data.longest_fixed) && data.last_start_min == 0 && data.last_end > 0 && !RExC_seen_zerolen && (!(RExC_seen & REG_SEEN_GPOS) || (r->reganch & ROPT_ANCH_GPOS))) ###### r->reganch |= ROPT_CHECK_ALL; ###### scan_commit(pRExC_state, &data); ###### SvREFCNT_dec(data.last_found); ###### longest_float_length = CHR_SVLEN(data.longest_float); ###### if (longest_float_length || (data.flags & SF_FL_BEFORE_EOL && (!(data.flags & SF_FL_BEFORE_MEOL) || (RExC_flags & PMf_MULTILINE)))) { ###### int t; ###### if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */ && data.offset_fixed == data.offset_float_min && SvCUR(data.longest_fixed) == SvCUR(data.longest_float)) ###### goto remove_float; /* As in (a)+. */ ###### if (SvUTF8(data.longest_float)) { ###### r->float_utf8 = data.longest_float; ###### r->float_substr = Nullsv; } else { ###### r->float_substr = data.longest_float; ###### r->float_utf8 = Nullsv; } ###### r->float_min_offset = data.offset_float_min; ###### r->float_max_offset = data.offset_float_max; ###### t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */ && (!(data.flags & SF_FL_BEFORE_MEOL) || (RExC_flags & PMf_MULTILINE))); ###### fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0); } else { remove_float: ###### r->float_substr = r->float_utf8 = Nullsv; ###### SvREFCNT_dec(data.longest_float); ###### longest_float_length = 0; } ###### longest_fixed_length = CHR_SVLEN(data.longest_fixed); ###### if (longest_fixed_length || (data.flags & SF_FIX_BEFORE_EOL /* Cannot have SEOL and MULTI */ && (!(data.flags & SF_FIX_BEFORE_MEOL) || (RExC_flags & PMf_MULTILINE)))) { ###### int t; ###### if (SvUTF8(data.longest_fixed)) { ###### r->anchored_utf8 = data.longest_fixed; ###### r->anchored_substr = Nullsv; } else { ###### r->anchored_substr = data.longest_fixed; ###### r->anchored_utf8 = Nullsv; } ###### r->anch