X-Git-Url: https://perl5.git.perl.org/perl5.git/blobdiff_plain/49d7dfbcef7527d25e8c34643f831ef2416923a3..5a449a8e553425ace4bea1ea13c16c03c6c5bf4c:/regcomp.h diff --git a/regcomp.h b/regcomp.h index 3e3f223..48e270f 100644 --- a/regcomp.h +++ b/regcomp.h @@ -1,7 +1,7 @@ /* regcomp.h * * Copyright (C) 1991, 1992, 1993, 1994, 1995, 1996, 1997, 1998, 1999, - * 2000, 2001, 2002, 2003, 2005, 2006 by Larry Wall and others + * 2000, 2001, 2002, 2003, 2005, 2006, 2007, 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. @@ -15,9 +15,28 @@ typedef OP OP_4tree; /* Will be redefined later. */ /* Convert branch sequences to more efficient trie ops? */ #define PERL_ENABLE_TRIE_OPTIMISATION 1 -/* Be really agressive about optimising patterns with trie sequences? */ +/* Be really aggressive about optimising patterns with trie sequences? */ #define PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION 1 +/* Use old style unicode mappings for perl and posix character classes + * + * NOTE: Enabling this essentially breaks character class matching against unicode + * strings, so that POSIX char classes match when they shouldn't, and \d matches + * way more than 10 characters, and sometimes a charclass and its complement either + * both match or neither match. + * NOTE: Disabling this will cause various backwards compatibility issues to rear + * their head, and tests to fail. However it will make the charclass behaviour + * consistent regardless of internal string type, and make character class inversions + * consistent. The tests that fail in the regex engine are basically broken tests. + * + * Personally I think 5.12 should disable this for sure. Its a bit more debatable for + * 5.10, so for now im leaving it enabled. + * XXX: It is now enabled for 5.11/5.12 + * + * -demerphq + */ +#define PERL_LEGACY_UNICODE_CHARCLASS_MAPPINGS 1 + /* Should the optimiser take positive assertions into account? */ #define PERL_ENABLE_POSITIVE_ASSERTION_STUDY 0 @@ -178,22 +197,23 @@ struct regnode_2 { #define ANYOF_BITMAP_SIZE 32 /* 256 b/(8 b/B) */ -#define ANYOF_CLASSBITMAP_SIZE 4 /* up to 40 (8*5) named classes */ +#define ANYOF_CLASSBITMAP_SIZE 4 /* up to 32 (8*4) named classes */ /* also used by trie */ struct regnode_charclass { U8 flags; U8 type; U16 next_off; - U32 arg1; + U32 arg1; /* used as ptr in S_regclass */ char bitmap[ANYOF_BITMAP_SIZE]; /* only compile-time */ }; -struct regnode_charclass_class { /* has [[:blah:]] classes */ - U8 flags; /* should have ANYOF_CLASS here */ +/* has runtime (locale) \d, \w, ..., [:posix:] classes */ +struct regnode_charclass_class { + U8 flags; /* ANYOF_CLASS bit must go here */ U8 type; U16 next_off; - U32 arg1; + U32 arg1; /* used as ptr in S_regclass */ char bitmap[ANYOF_BITMAP_SIZE]; /* both compile-time */ char classflags[ANYOF_CLASSBITMAP_SIZE]; /* and run-time */ }; @@ -252,6 +272,9 @@ struct regnode_charclass_class { /* has [[:blah:]] classes */ #undef STRING #define OP(p) ((p)->type) +#define FLAGS(p) ((p)->flags) /* Caution: Doesn't apply to all \ + regnode types. For some, it's the \ + character set of the regnode */ #define OPERAND(p) (((struct regnode_string *)p)->string) #define MASK(p) ((char*)OPERAND(p)) #define STR_LEN(p) (((struct regnode_string *)p)->str_len) @@ -287,25 +310,62 @@ struct regnode_charclass_class { /* has [[:blah:]] classes */ #define SIZE_ONLY (RExC_emit == &PL_regdummy) -/* Flags for node->flags of ANYOF */ +/* Flags for node->flags of several of the node types */ +#define USE_UNI 0x01 + +/* Flags for node->flags of ANYOF. These are in short supply, so some games + * are done to share them, as described below. If necessary, the ANYOF_LOCALE + * and ANYOF_CLASS bits could be shared with a space penalty for locale nodes + * (and the code at the time this comment was written, is written so that all + * that is necessary to make the change would be to redefine the ANYOF_CLASS + * define). Once the planned change to compile all the above-latin1 code points + * is done, then the UNICODE_ALL bit can be freed up. If flags need to be + * added that are applicable to the synthetic start class only, with some work, + * they could be put in the next-node field, or in an unused bit of the + * classflags field. */ + +#define ANYOF_LOCALE 0x01 + +/* The fold is calculated and stored in the bitmap where possible at compile + * time. However there are two cases where it isn't possible. These share + * this bit: 1) under locale, where the actual folding varies depending on + * what the locale is at the time of execution; and 2) where the folding is + * specified in a swash, not the bitmap, such as characters which aren't + * specified in the bitmap, or properties that aren't looked at at compile time + */ +#define ANYOF_LOC_NONBITMAP_FOLD 0x02 + +#define ANYOF_INVERT 0x04 + +/* EOS, meaning that it can match an empty string too, is used for the + * synthetic start class (ssc) only. It can share the INVERT bit, as the ssc + * is never inverted. The bit just needs to be turned off before regexec.c + * gets a hold of it so that regexec.c doesn't think it's inverted, but this + * happens automatically, as if the ssc can match an EOS, the ssc is discarded, + * and never passed to regexec.c */ +#define ANYOF_EOS ANYOF_INVERT -#define ANYOF_CLASS 0x08 /* has [[:blah:]] classes */ -#define ANYOF_INVERT 0x04 -#define ANYOF_FOLD 0x02 -#define ANYOF_LOCALE 0x01 +/* CLASS is never set unless LOCALE is too: has runtime \d, \w, [:posix:], ... + * The non-locale ones are resolved at compile-time */ +#define ANYOF_CLASS 0x08 +#define ANYOF_LARGE ANYOF_CLASS /* Same; name retained for back compat */ -/* Used for regstclass only */ -#define ANYOF_EOS 0x10 /* Can match an empty string too */ +/* Can match something outside the bitmap that is expressible only in utf8 */ +#define ANYOF_UTF8 0x10 -/* There is a character or a range past 0xff */ -#define ANYOF_UNICODE 0x20 -#define ANYOF_UNICODE_ALL 0x40 /* Can match any char past 0xff */ +/* Can match something outside the bitmap that isn't in utf8 */ +#define ANYOF_NONBITMAP_NON_UTF8 0x20 -/* size of node is large (includes class pointer) */ -#define ANYOF_LARGE 0x80 +/* Set if the bitmap doesn't fully represent what this node can match */ +#define ANYOF_NONBITMAP (ANYOF_UTF8|ANYOF_NONBITMAP_NON_UTF8) +#define ANYOF_UNICODE ANYOF_NONBITMAP /* old name, for back compat */ -/* Are there any runtime flags on in this node? */ -#define ANYOF_RUNTIME(s) (ANYOF_FLAGS(s) & 0x0f) +/* Matches every code point 0x100 and above*/ +#define ANYOF_UNICODE_ALL 0x40 + +/* Match all Latin1 characters that aren't ASCII when the target string is not + * in utf8. */ +#define ANYOF_NON_UTF8_LATIN1_ALL 0x80 #define ANYOF_FLAGS_ALL 0xff @@ -317,9 +377,9 @@ struct regnode_charclass_class { /* has [[:blah:]] classes */ #define ANYOF_NALNUM 1 #define ANYOF_SPACE 2 /* \s */ #define ANYOF_NSPACE 3 -#define ANYOF_DIGIT 4 +#define ANYOF_DIGIT 4 /* \d */ #define ANYOF_NDIGIT 5 -#define ANYOF_ALNUMC 6 /* isalnum(3), utf8::IsAlnum, ALNUMC */ +#define ANYOF_ALNUMC 6 /* [[:alnum:]] isalnum(3), utf8::IsAlnum, ALNUMC */ #define ANYOF_NALNUMC 7 #define ANYOF_ALPHA 8 #define ANYOF_NALPHA 9 @@ -389,12 +449,34 @@ struct regnode_charclass_class { /* has [[:blah:]] classes */ #define ANYOF_BITMAP_CLEARALL(p) \ Zero (ANYOF_BITMAP(p), ANYOF_BITMAP_SIZE) /* Check that all 256 bits are all set. Used in S_cl_is_anything() */ -#define ANYOF_BITMAP_TESTALLSET(p) \ +#define ANYOF_BITMAP_TESTALLSET(p) /* Assumes sizeof(p) == 32 */ \ memEQ (ANYOF_BITMAP(p), "\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377\377", ANYOF_BITMAP_SIZE) #define ANYOF_SKIP ((ANYOF_SIZE - 1)/sizeof(regnode)) #define ANYOF_CLASS_SKIP ((ANYOF_CLASS_SIZE - 1)/sizeof(regnode)) -#define ANYOF_CLASS_ADD_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP) + +/* The class bit can be set to the locale one if necessary to save bits at the + * expense of having locale ANYOF nodes always have a class bit map, and hence + * take up extra space. This allows convenient changing it as development + * proceeds on this */ +#if ANYOF_CLASS == ANYOF_LOCALE +# undef ANYOF_CLASS_ADD_SKIP +# define ANYOF_ADD_LOC_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP) + + /* Quicker way to see if there are actually any tests. This is because + * currently the set of tests can be empty even when the class bitmap is + * allocated */ +# if ANYOF_CLASSBITMAP_SIZE != 4 +# error ANYOF_CLASSBITMAP_SIZE is expected to be 4 +# endif +# define ANYOF_CLASS_TEST_ANY_SET(p) /* assumes sizeof(p) = 4 */ \ + memNE (((struct regnode_charclass_class*)(p))->classflags, \ + "\0\0\0\0", ANYOF_CLASSBITMAP_SIZE) +#else +# define ANYOF_CLASS_ADD_SKIP (ANYOF_CLASS_SKIP - ANYOF_SKIP) +# undef ANYOF_ADD_LOC_SKIP +# define ANYOF_CLASS_TEST_ANY_SET(p) (ANYOF_FLAGS(p) & ANYOF_CLASS) +#endif /* @@ -428,49 +510,21 @@ START_EXTERN_C #include "regnodes.h" #endif -/* The following have no fixed length. U8 so we can do strchr() on it. */ -#ifndef DOINIT -EXTCONST U8 PL_varies[]; -#else -EXTCONST U8 PL_varies[] = { - BRANCH, BACK, STAR, PLUS, CURLY, CURLYX, REF, REFF, REFFL, - WHILEM, CURLYM, CURLYN, BRANCHJ, IFTHEN, SUSPEND, CLUMP, - NREF, NREFF, NREFFL, - 0 -}; -#endif - -/* The following always have a length of 1. U8 we can do strchr() on it. */ -/* (Note that length 1 means "one character" under UTF8, not "one octet".) */ -#ifndef DOINIT -EXTCONST U8 PL_simple[]; -#else -EXTCONST U8 PL_simple[] = { - REG_ANY, SANY, CANY, - ANYOF, - ALNUM, ALNUML, - NALNUM, NALNUML, - SPACE, SPACEL, - NSPACE, NSPACEL, - DIGIT, NDIGIT, - VERTWS, NVERTWS, - HORIZWS, NHORIZWS, - 0 -}; -#endif - #ifndef PLUGGABLE_RE_EXTENSION #ifndef DOINIT EXTCONST regexp_engine PL_core_reg_engine; #else /* DOINIT */ EXTCONST regexp_engine PL_core_reg_engine = { Perl_re_compile, - Perl_regexec_flags, + Perl_regexec_flags, Perl_re_intuit_start, Perl_re_intuit_string, - Perl_regfree_internal, - Perl_reg_numbered_buff_get, - Perl_reg_named_buff_get, + Perl_regfree_internal, + Perl_reg_numbered_buff_fetch, + Perl_reg_numbered_buff_store, + Perl_reg_numbered_buff_length, + Perl_reg_named_buff, + Perl_reg_named_buff_iter, Perl_reg_qr_package, #if defined(USE_ITHREADS) Perl_regdupe_internal @@ -485,11 +539,12 @@ END_EXTERN_C /* .what is a character array with one character for each member of .data * The character describes the function of the corresponding .data item: + * a - AV for paren_name_list under DEBUGGING * f - start-class data for regstclass optimization * n - Root of op tree for (?{EVAL}) item * o - Start op for (?{EVAL}) item * p - Pad for (?{EVAL}) item - * s - swash for unicode-style character class, and the multicharacter + * s - swash for Unicode-style character class, and the multicharacter * strings resulting from casefolding the single-character entries * in the character class * t - trie struct @@ -525,7 +580,10 @@ struct reg_data { #define check_offset_max substrs->data[2].max_offset #define check_end_shift substrs->data[2].end_shift - +#define RX_ANCHORED_SUBSTR(rx) (((struct regexp *)SvANY(rx))->anchored_substr) +#define RX_ANCHORED_UTF8(rx) (((struct regexp *)SvANY(rx))->anchored_utf8) +#define RX_FLOAT_SUBSTR(rx) (((struct regexp *)SvANY(rx))->float_substr) +#define RX_FLOAT_UTF8(rx) (((struct regexp *)SvANY(rx))->float_utf8) /* trie related stuff */ @@ -561,6 +619,15 @@ struct _reg_trie_state { } trans; }; +/* info per word; indexed by wordnum */ +typedef struct { + U16 prev; /* previous word in acceptance chain; eg in + * zzz|abc|ab/ after matching the chars abc, the + * accepted word is #2, and the previous accepted + * word is #3 */ + U32 len; /* how many chars long is this word? */ + U32 accept; /* accept state for this word */ +} reg_trie_wordinfo; typedef struct _reg_trie_state reg_trie_state; @@ -578,15 +645,14 @@ struct _reg_trie_data { reg_trie_state *states; /* state data */ reg_trie_trans *trans; /* array of transition elements */ char *bitmap; /* stclass bitmap */ - U32 *wordlen; /* array of lengths of words */ U16 *jump; /* optional 1 indexed array of offsets before tail for the node following a given word. */ - U16 *nextword; /* optional 1 indexed array to support linked list - of duplicate wordnums */ + reg_trie_wordinfo *wordinfo; /* array of info per word */ U16 uniquecharcount; /* unique chars in trie (width of trans table) */ U32 startstate; /* initial state - used for common prefix optimisation */ STRLEN minlen; /* minimum length of words in trie - build/opt only? */ STRLEN maxlen; /* maximum length of words in trie - build/opt only? */ + U32 prefixlen; /* #chars in common prefix */ U32 statecount; /* Build only - number of states in the states array (including the unused zero state) */ U32 wordcount; /* Build only */ @@ -618,7 +684,7 @@ struct _reg_ac_data { }; typedef struct _reg_ac_data reg_ac_data; -/* ANY_BIT doesnt use the structure, so we can borrow it here. +/* ANY_BIT doesn't use the structure, so we can borrow it here. This is simpler than refactoring all of it as wed end up with three different sets... */ @@ -688,6 +754,7 @@ re.pm, especially to the documentation. #define RE_DEBUG_COMPILE_OPTIMISE 0x000002 #define RE_DEBUG_COMPILE_TRIE 0x000004 #define RE_DEBUG_COMPILE_DUMP 0x000008 +#define RE_DEBUG_COMPILE_FLAGS 0x000010 /* Execute */ #define RE_DEBUG_EXECUTE_MASK 0x00FF00 @@ -703,6 +770,7 @@ re.pm, especially to the documentation. #define RE_DEBUG_EXTRA_STATE 0x080000 #define RE_DEBUG_EXTRA_OPTIMISE 0x100000 #define RE_DEBUG_EXTRA_BUFFERS 0x400000 +#define RE_DEBUG_EXTRA_GPOS 0x800000 /* combined */ #define RE_DEBUG_EXTRA_STACK 0x280000 @@ -720,7 +788,8 @@ re.pm, especially to the documentation. if (re_debug_flags & RE_DEBUG_COMPILE_DUMP) x ) #define DEBUG_TRIE_COMPILE_r(x) DEBUG_r( \ if (re_debug_flags & RE_DEBUG_COMPILE_TRIE) x ) - +#define DEBUG_FLAGS_r(x) DEBUG_r( \ + if (re_debug_flags & RE_DEBUG_COMPILE_FLAGS) x ) /* Execute */ #define DEBUG_EXECUTE_r(x) DEBUG_r( \ if (re_debug_flags & RE_DEBUG_EXECUTE_MASK) x ) @@ -757,11 +826,15 @@ re.pm, especially to the documentation. #define DEBUG_TRIE_r(x) DEBUG_r( \ if (re_debug_flags & (RE_DEBUG_COMPILE_TRIE \ | RE_DEBUG_EXECUTE_TRIE )) x ) +#define DEBUG_GPOS_r(x) DEBUG_r( \ + if (re_debug_flags & RE_DEBUG_EXTRA_GPOS) x ) /* initialization */ -/* get_sv() can return NULL during global destruction. */ +/* get_sv() can return NULL during global destruction. re_debug_flags can get + * clobbered by a longjmp, so must be initialized */ #define GET_RE_DEBUG_FLAGS DEBUG_r({ \ SV * re_debug_flags_sv = NULL; \ + re_debug_flags = 0; \ re_debug_flags_sv = get_sv(RE_DEBUG_FLAGS, 1); \ if (re_debug_flags_sv) { \ if (!SvIOK(re_debug_flags_sv)) \ @@ -772,26 +845,26 @@ re.pm, especially to the documentation. #ifdef DEBUGGING -#define GET_RE_DEBUG_FLAGS_DECL IV re_debug_flags = 0; GET_RE_DEBUG_FLAGS; +#define GET_RE_DEBUG_FLAGS_DECL VOL IV re_debug_flags = 0; GET_RE_DEBUG_FLAGS; #define RE_PV_COLOR_DECL(rpv,rlen,isuni,dsv,pv,l,m,c1,c2) \ const char * const rpv = \ pv_pretty((dsv), (pv), (l), (m), \ PL_colors[(c1)],PL_colors[(c2)], \ - PERL_PV_ESCAPE_RE |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \ + PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ); \ const int rlen = SvCUR(dsv) #define RE_SV_ESCAPE(rpv,isuni,dsv,sv,m) \ const char * const rpv = \ pv_pretty((dsv), (SvPV_nolen_const(sv)), (SvCUR(sv)), (m), \ PL_colors[(c1)],PL_colors[(c2)], \ - PERL_PV_ESCAPE_RE |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ) + PERL_PV_ESCAPE_RE|PERL_PV_ESCAPE_NONASCII |((isuni) ? PERL_PV_ESCAPE_UNI : 0) ) #define RE_PV_QUOTED_DECL(rpv,isuni,dsv,pv,l,m) \ const char * const rpv = \ pv_pretty((dsv), (pv), (l), (m), \ PL_colors[0], PL_colors[1], \ - ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_PRETTY_ELIPSES | \ + ( PERL_PV_PRETTY_QUOTE | PERL_PV_ESCAPE_RE | PERL_PV_ESCAPE_NONASCII | PERL_PV_PRETTY_ELLIPSES | \ ((isuni) ? PERL_PV_ESCAPE_UNI : 0)) \ ) @@ -809,5 +882,12 @@ re.pm, especially to the documentation. #endif /* DEBUG RELATED DEFINES */ - - +/* + * Local variables: + * c-indentation-style: bsd + * c-basic-offset: 4 + * indent-tabs-mode: t + * End: + * + * ex: set ts=8 sts=4 sw=4 noet: + */