typedef struct RExC_state_t {
U32 flags; /* are we folding, multilining? */
char *precomp; /* uncompiled string. */
- regexp *rx;
+ regexp *rx; /* perl core regexp structure */
+ regexp_internal *rxi; /* internal data for regexp object pprivate field */
char *start; /* Start of input for compile */
char *end; /* End of input for compile */
char *parse; /* Input-scan pointer. */
I32 sawback; /* Did we see \1, ...? */
U32 seen;
I32 size; /* Code size. */
- I32 npar; /* () count. */
+ I32 npar; /* Capture buffer count, (OPEN). */
+ I32 cpar; /* Capture buffer count, (CLOSE). */
+ I32 nestroot; /* root parens we are in - used by accept */
I32 extralen;
I32 seen_zerolen;
I32 seen_evals;
+ regnode **open_parens; /* pointers to open parens */
+ regnode **close_parens; /* pointers to close parens */
+ regnode *opend; /* END node in program */
I32 utf8;
+ HV *charnames; /* cache of named sequences */
+ HV *paren_names; /* Paren names */
+ regnode **recurse; /* Recurse regops */
+ I32 recurse_count; /* Number of recurse regops */
#if ADD_TO_REGEXEC
char *starttry; /* -Dr: where regtry was called. */
#define RExC_starttry (pRExC_state->starttry)
#define RExC_flags (pRExC_state->flags)
#define RExC_precomp (pRExC_state->precomp)
#define RExC_rx (pRExC_state->rx)
+#define RExC_rxi (pRExC_state->rxi)
#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_offsets (pRExC_state->rxi->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_seen (pRExC_state->seen)
#define RExC_size (pRExC_state->size)
#define RExC_npar (pRExC_state->npar)
+#define RExC_cpar (pRExC_state->cpar)
+#define RExC_nestroot (pRExC_state->nestroot)
#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 RExC_charnames (pRExC_state->charnames)
+#define RExC_open_parens (pRExC_state->open_parens)
+#define RExC_close_parens (pRExC_state->close_parens)
+#define RExC_opend (pRExC_state->opend)
+#define RExC_paren_names (pRExC_state->paren_names)
+#define RExC_recurse (pRExC_state->recurse)
+#define RExC_recurse_count (pRExC_state->recurse_count)
#define ISMULT1(c) ((c) == '*' || (c) == '+' || (c) == '?')
#define ISMULT2(s) ((*s) == '*' || (*s) == '+' || (*s) == '?' || \
/* whether trie related optimizations are enabled */
#if PERL_ENABLE_EXTENDED_TRIE_OPTIMISATION
#define TRIE_STUDY_OPT
+#define FULL_TRIE_STUDY
#define TRIE_STCLASS
#endif
-/* Length of a variant. */
+
+
+
+#define PBYTE(u8str,paren) ((U8*)(u8str))[(paren) >> 3]
+#define PBITVAL(paren) (1 << ((paren) & 7))
+#define PAREN_TEST(u8str,paren) ( PBYTE(u8str,paren) & PBITVAL(paren))
+#define PAREN_SET(u8str,paren) PBYTE(u8str,paren) |= PBITVAL(paren)
+#define PAREN_UNSET(u8str,paren) PBYTE(u8str,paren) &= (~PBITVAL(paren))
+
+
+/* About scan_data_t.
+
+ During optimisation we recurse through the regexp program performing
+ various inplace (keyhole style) optimisations. In addition study_chunk
+ and scan_commit populate this data structure with information about
+ what strings MUST appear in the pattern. We look for the longest
+ string that must appear for at a fixed location, and we look for the
+ longest string that may appear at a floating location. So for instance
+ in the pattern:
+
+ /FOO[xX]A.*B[xX]BAR/
+
+ Both 'FOO' and 'A' are fixed strings. Both 'B' and 'BAR' are floating
+ strings (because they follow a .* construct). study_chunk will identify
+ both FOO and BAR as being the longest fixed and floating strings respectively.
+
+ The strings can be composites, for instance
+
+ /(f)(o)(o)/
+
+ will result in a composite fixed substring 'foo'.
+
+ For each string some basic information is maintained:
+
+ - offset or min_offset
+ This is the position the string must appear at, or not before.
+ It also implicitly (when combined with minlenp) tells us how many
+ character must match before the string we are searching.
+ Likewise when combined with minlenp and the length of the string
+ tells us how many characters must appear after the string we have
+ found.
+
+ - max_offset
+ Only used for floating strings. This is the rightmost point that
+ the string can appear at. Ifset to I32 max it indicates that the
+ string can occur infinitely far to the right.
+
+ - minlenp
+ A pointer to the minimum length of the pattern that the string
+ was found inside. This is important as in the case of positive
+ lookahead or positive lookbehind we can have multiple patterns
+ involved. Consider
+
+ /(?=FOO).*F/
+
+ The minimum length of the pattern overall is 3, the minimum length
+ of the lookahead part is 3, but the minimum length of the part that
+ will actually match is 1. So 'FOO's minimum length is 3, but the
+ minimum length for the F is 1. This is important as the minimum length
+ is used to determine offsets in front of and behind the string being
+ looked for. Since strings can be composites this is the length of the
+ pattern at the time it was commited with a scan_commit. Note that
+ the length is calculated by study_chunk, so that the minimum lengths
+ are not known until the full pattern has been compiled, thus the
+ pointer to the value.
+
+ - lookbehind
+
+ In the case of lookbehind the string being searched for can be
+ offset past the start point of the final matching string.
+ If this value was just blithely removed from the min_offset it would
+ invalidate some of the calculations for how many chars must match
+ before or after (as they are derived from min_offset and minlen and
+ the length of the string being searched for).
+ When the final pattern is compiled and the data is moved from the
+ scan_data_t structure into the regexp structure the information
+ about lookbehind is factored in, with the information that would
+ have been lost precalculated in the end_shift field for the
+ associated string.
+
+ The fields pos_min and pos_delta are used to store the minimum offset
+ and the delta to the maximum offset at the current point in the pattern.
+
+*/
typedef struct scan_data_t {
- I32 len_min;
- I32 len_delta;
+ /*I32 len_min; unused */
+ /*I32 len_delta; unused */
I32 pos_min;
I32 pos_delta;
SV *last_found;
- I32 last_end; /* min value, <0 unless valid. */
+ 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;
+ SV **longest; /* Either &l_fixed, or &l_float. */
+ SV *longest_fixed; /* longest fixed string found in pattern */
+ I32 offset_fixed; /* offset where it starts */
+ I32 *minlen_fixed; /* pointer to the minlen relevent to the string */
+ I32 lookbehind_fixed; /* is the position of the string modfied by LB */
+ SV *longest_float; /* longest floating string found in pattern */
+ I32 offset_float_min; /* earliest point in string it can appear */
+ I32 offset_float_max; /* latest point in string it can appear */
+ I32 *minlen_float; /* pointer to the minlen relevent to the string */
+ I32 lookbehind_float; /* is the position of the string modified by LB */
I32 flags;
I32 whilem_c;
I32 *last_closep;
*/
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};
+ { 0, 0, 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 0x0001
#define SCF_DO_STCLASS (SCF_DO_STCLASS_AND|SCF_DO_STCLASS_OR)
#define SCF_WHILEM_VISITED_POS 0x2000
-#define SCF_EXACT_TRIE 0x4000 /* should re study once we are done? */
+#define SCF_TRIE_RESTUDY 0x4000 /* Do restudy? */
+#define SCF_SEEN_ACCEPT 0x8000
#define UTF (RExC_utf8 != 0)
-#define LOC ((RExC_flags & PMf_LOCALE) != 0)
-#define FOLD ((RExC_flags & PMf_FOLD) != 0)
+#define LOC ((RExC_flags & RXf_PMf_LOCALE) != 0)
+#define FOLD ((RExC_flags & RXf_PMf_FOLD) != 0)
#define OOB_UNICODE 12345678
#define OOB_NAMEDCLASS -1
* arg. Show regex, up to a maximum length. If it's too long, chop and add
* "...".
*/
-#define FAIL(msg) STMT_START { \
+#define _FAIL(code) STMT_START { \
const char *ellipses = ""; \
IV len = RExC_end - RExC_precomp; \
\
len = RegexLengthToShowInErrorMessages - 10; \
ellipses = "..."; \
} \
- Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
- msg, (int)len, RExC_precomp, ellipses); \
+ code; \
} STMT_END
+#define FAIL(msg) _FAIL( \
+ Perl_croak(aTHX_ "%s in regex m/%.*s%s/", \
+ msg, (int)len, RExC_precomp, ellipses))
+
+#define FAIL2(msg,arg) _FAIL( \
+ Perl_croak(aTHX_ msg " in regex m/%.*s%s/", \
+ arg, (int)len, RExC_precomp, ellipses))
+
/*
* Simple_vFAIL -- like FAIL, but marks the current location in the scan
*/
#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), (int)(byte))); \
+ __LINE__, (int)(node), (int)(byte))); \
if((node) < 0) { \
Perl_croak(aTHX_ "value of node is %d in Offset macro", (int)(node)); \
} else { \
#define EXPERIMENTAL_INPLACESCAN
#endif
+#define DEBUG_STUDYDATA(data,depth) \
+DEBUG_OPTIMISE_MORE_r(if(data){ \
+ PerlIO_printf(Perl_debug_log, \
+ "%*s"/* Len:%"IVdf"/%"IVdf" */"Pos:%"IVdf"/%"IVdf \
+ " Flags: %"IVdf" Whilem_c: %"IVdf" Lcp: %"IVdf" ", \
+ (int)(depth)*2, "", \
+ (IV)((data)->pos_min), \
+ (IV)((data)->pos_delta), \
+ (IV)((data)->flags), \
+ (IV)((data)->whilem_c), \
+ (IV)((data)->last_closep ? *((data)->last_closep) : -1) \
+ ); \
+ if ((data)->last_found) \
+ PerlIO_printf(Perl_debug_log, \
+ "Last:'%s' %"IVdf":%"IVdf"/%"IVdf" %sFixed:'%s' @ %"IVdf \
+ " %sFloat: '%s' @ %"IVdf"/%"IVdf"", \
+ SvPVX_const((data)->last_found), \
+ (IV)((data)->last_end), \
+ (IV)((data)->last_start_min), \
+ (IV)((data)->last_start_max), \
+ ((data)->longest && \
+ (data)->longest==&((data)->longest_fixed)) ? "*" : "", \
+ SvPVX_const((data)->longest_fixed), \
+ (IV)((data)->offset_fixed), \
+ ((data)->longest && \
+ (data)->longest==&((data)->longest_float)) ? "*" : "", \
+ SvPVX_const((data)->longest_float), \
+ (IV)((data)->offset_float_min), \
+ (IV)((data)->offset_float_max) \
+ ); \
+ PerlIO_printf(Perl_debug_log,"\n"); \
+});
+
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
+ Update the longest found anchored substring and the longest found
floating substrings if needed. */
STATIC void
-S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data)
+S_scan_commit(pTHX_ const RExC_state_t *pRExC_state, scan_data_t *data, I32 *minlenp)
{
const STRLEN l = CHR_SVLEN(data->last_found);
const STRLEN old_l = CHR_SVLEN(*data->longest);
+ GET_RE_DEBUG_FLAGS_DECL;
if ((l >= old_l) && ((l > old_l) || (data->flags & SF_BEFORE_EOL))) {
SvSetMagicSV(*data->longest, data->last_found);
|= ((data->flags & SF_BEFORE_EOL) << SF_FIX_SHIFT_EOL);
else
data->flags &= ~SF_FIX_BEFORE_EOL;
+ data->minlen_fixed=minlenp;
+ data->lookbehind_fixed=0;
}
else {
data->offset_float_min = l ? data->last_start_min : data->pos_min;
|= ((data->flags & SF_BEFORE_EOL) << SF_FL_SHIFT_EOL);
else
data->flags &= ~SF_FL_BEFORE_EOL;
+ data->minlen_float=minlenp;
+ data->lookbehind_float=0;
}
}
SvCUR_set(data->last_found, 0);
}
data->last_end = -1;
data->flags &= ~SF_BEFORE_EOL;
+ DEBUG_STUDYDATA(data,0);
}
/* Can match anything (initialization) */
return 1;
if (!(cl->flags & ANYOF_UNICODE_ALL))
return 0;
- if (!ANYOF_BITMAP_TESTALLSET(cl))
+ if (!ANYOF_BITMAP_TESTALLSET((const void*)cl))
return 0;
return 1;
}
S_cl_and(struct regnode_charclass_class *cl,
const struct regnode_charclass_class *and_with)
{
+
+ assert(and_with->type == ANYOF);
if (!(and_with->flags & ANYOF_CLASS)
&& !(cl->flags & ANYOF_CLASS)
&& (and_with->flags & ANYOF_LOCALE) == (cl->flags & ANYOF_LOCALE)
}
}
-/*
-
- make_trie(startbranch,first,last,tail,flags,depth)
- 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)/
- depth : indent depth
-
-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 <ac>(16)
- 8: BRANCH(11)
- 9: EXACT <ad>(16)
- 11: BRANCH(14)
- 12: EXACT <ab>(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]
- <ac>
- <ad>
- <ab>
- 16: SUCCEED(0)
- 17: NOTHING(18)
- 18: END(0)
-
-Cases where tail != last would be like /(?foo|bar)baz/:
-
- 1: BRANCH(4)
- 2: EXACT <foo>(8)
- 4: BRANCH(7)
- 5: EXACT <bar>(8)
- 7: TAIL(8)
- 8: EXACT <baz>(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]
- <foo>
- <bar>
- 7: TAIL(8)
- 8: EXACT <baz>(10)
- 10: END(0)
-
- d = uvuni_to_utf8_flags(d, uv, 0);
-
-is the recommended Unicode-aware way of saying
-
- *(d++) = uv;
-*/
-
-#define TRIE_STORE_REVCHAR \
- STMT_START { \
- SV *tmp = Perl_newSVpvf_nocontext( "%c", (int)uvc ); \
- av_push( TRIE_REVCHARMAP(trie), tmp ); \
- } STMT_END
-
-#define TRIE_READ_CHAR STMT_START { \
- wordlen++; \
- 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 { \
- Newxz( trie->states[ state ].trans.list, \
- 4, reg_trie_trans_le ); \
- TRIE_LIST_CUR( state ) = 1; \
- TRIE_LIST_LEN( state ) = 4; \
-} STMT_END
-
-#define TRIE_HANDLE_WORD(state) STMT_START { \
- if ( !trie->states[ state ].wordnum ) { \
- /* we haven't inserted this word into the structure yet. */ \
- if (trie->wordlen) \
- trie->wordlen[ curword ] = wordlen; \
- trie->states[ state ].wordnum = ++curword; \
- DEBUG_r({ \
- /* store the word for dumping */ \
- SV* tmp; \
- if (OP(noper) != NOTHING) \
- tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
- else \
- tmp = newSVpvn( "", 0 ); \
- if ( UTF ) SvUTF8_on( tmp ); \
- av_push( trie->words, tmp ); \
- }); \
- } else { \
- NOOP; /* It's a dupe. So ignore it. */ \
- } \
-} STMT_END
#ifdef DEBUGGING
/*
- dump_trie(trie)
- dump_trie_interim_list(trie,next_alloc)
- dump_trie_interim_table(trie,next_alloc)
+ dump_trie(trie,widecharmap,revcharmap)
+ dump_trie_interim_list(trie,widecharmap,revcharmap,next_alloc)
+ dump_trie_interim_table(trie,widecharmap,revcharmap,next_alloc)
These routines dump out a trie in a somewhat readable format.
The _interim_ variants are used for debugging the interim
*/
/*
- dump_trie(trie)
Dumps the final compressed table form of the trie to Perl_debug_log.
Used for debugging make_trie().
*/
STATIC void
-S_dump_trie(pTHX_ const struct _reg_trie_data *trie,U32 depth)
+S_dump_trie(pTHX_ const struct _reg_trie_data *trie, HV *widecharmap,
+ AV *revcharmap, U32 depth)
{
U32 state;
+ SV *sv=sv_newmortal();
+ int colwidth= widecharmap ? 6 : 4;
GET_RE_DEBUG_FLAGS_DECL;
+
PerlIO_printf( Perl_debug_log, "%*sChar : %-6s%-6s%-4s ",
(int)depth * 2 + 2,"",
"Match","Base","Ofs" );
for( state = 0 ; state < trie->uniquecharcount ; state++ ) {
- SV ** const tmp = av_fetch( trie->revcharmap, state, 0);
+ SV ** const tmp = av_fetch( revcharmap, state, 0);
if ( tmp ) {
- PerlIO_printf( Perl_debug_log, "%4.4s ", SvPV_nolen_const( *tmp ) );
+ PerlIO_printf( Perl_debug_log, "%*s",
+ colwidth,
+ pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
+ PL_colors[0], PL_colors[1],
+ (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
+ PERL_PV_ESCAPE_FIRSTCHAR
+ )
+ );
}
}
PerlIO_printf( Perl_debug_log, "\n%*sState|-----------------------",
(int)depth * 2 + 2,"");
for( state = 0 ; state < trie->uniquecharcount ; state++ )
- PerlIO_printf( Perl_debug_log, "-----");
+ PerlIO_printf( Perl_debug_log, "%.*s", colwidth, "--------");
PerlIO_printf( Perl_debug_log, "\n");
- for( state = 1 ; state < TRIE_LASTSTATE(trie) ; state++ ) {
+ for( state = 1 ; state < trie->statecount ; state++ ) {
const U32 base = trie->states[ state ].trans.base;
PerlIO_printf( Perl_debug_log, "%*s#%4"UVXf"|", (int)depth * 2 + 2,"", (UV)state);
( base + ofs - trie->uniquecharcount < trie->lasttrans ) &&
trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
{
- PerlIO_printf( Perl_debug_log, "%4"UVXf" ",
+ PerlIO_printf( Perl_debug_log, "%*"UVXf,
+ colwidth,
(UV)trie->trans[ base + ofs - trie->uniquecharcount ].next );
} else {
- PerlIO_printf( Perl_debug_log, "%4s "," ." );
+ PerlIO_printf( Perl_debug_log, "%*s",colwidth," ." );
}
}
}
}
/*
- dump_trie_interim_list(trie,next_alloc)
Dumps a fully constructed but uncompressed trie in list form.
List tries normally only are used for construction when the number of
possible chars (trie->uniquecharcount) is very high.
Used for debugging make_trie().
*/
STATIC void
-S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie, U32 next_alloc,U32 depth)
+S_dump_trie_interim_list(pTHX_ const struct _reg_trie_data *trie,
+ HV *widecharmap, AV *revcharmap, U32 next_alloc,
+ U32 depth)
{
U32 state;
+ SV *sv=sv_newmortal();
+ int colwidth= widecharmap ? 6 : 4;
GET_RE_DEBUG_FLAGS_DECL;
/* print out the table precompression. */
- PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s",
- (int)depth * 2 + 2,"", (int)depth * 2 + 2,"");
- PerlIO_printf( Perl_debug_log, "------:-----+-----------------" );
+ PerlIO_printf( Perl_debug_log, "%*sState :Word | Transition Data\n%*s%s",
+ (int)depth * 2 + 2,"", (int)depth * 2 + 2,"",
+ "------:-----+-----------------\n" );
for( state=1 ; state < next_alloc ; state ++ ) {
U16 charid;
- PerlIO_printf( Perl_debug_log, "\n%*s %4"UVXf" :",
+ PerlIO_printf( Perl_debug_log, "%*s %4"UVXf" :",
(int)depth * 2 + 2,"", (UV)state );
if ( ! trie->states[ state ].wordnum ) {
PerlIO_printf( Perl_debug_log, "%5s| ","");
);
}
for( charid = 1 ; charid <= TRIE_LIST_USED( state ) ; charid++ ) {
- SV ** const tmp = av_fetch( trie->revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
- PerlIO_printf( Perl_debug_log, "%s:%3X=%4"UVXf" | ",
- SvPV_nolen_const( *tmp ),
- TRIE_LIST_ITEM(state,charid).forid,
- (UV)TRIE_LIST_ITEM(state,charid).newstate
- );
+ SV ** const tmp = av_fetch( revcharmap, TRIE_LIST_ITEM(state,charid).forid, 0);
+ if ( tmp ) {
+ PerlIO_printf( Perl_debug_log, "%*s:%3X=%4"UVXf" | ",
+ colwidth,
+ pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
+ PL_colors[0], PL_colors[1],
+ (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
+ PERL_PV_ESCAPE_FIRSTCHAR
+ ) ,
+ TRIE_LIST_ITEM(state,charid).forid,
+ (UV)TRIE_LIST_ITEM(state,charid).newstate
+ );
+ if (!(charid % 10))
+ PerlIO_printf(Perl_debug_log, "\n%*s| ",
+ (int)((depth * 2) + 14), "");
+ }
}
-
+ PerlIO_printf( Perl_debug_log, "\n");
}
}
/*
- dump_trie_interim_table(trie,next_alloc)
Dumps a fully constructed but uncompressed trie in table form.
This is the normal DFA style state transition table, with a few
twists to facilitate compression later.
Used for debugging make_trie().
*/
STATIC void
-S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie, U32 next_alloc, U32 depth)
+S_dump_trie_interim_table(pTHX_ const struct _reg_trie_data *trie,
+ HV *widecharmap, AV *revcharmap, U32 next_alloc,
+ U32 depth)
{
U32 state;
U16 charid;
+ SV *sv=sv_newmortal();
+ int colwidth= widecharmap ? 6 : 4;
GET_RE_DEBUG_FLAGS_DECL;
/*
PerlIO_printf( Perl_debug_log, "%*sChar : ",(int)depth * 2 + 2,"" );
for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
- SV ** const tmp = av_fetch( trie->revcharmap, charid, 0);
+ SV ** const tmp = av_fetch( revcharmap, charid, 0);
if ( tmp ) {
- PerlIO_printf( Perl_debug_log, "%4.4s ", SvPV_nolen_const( *tmp ) );
+ PerlIO_printf( Perl_debug_log, "%*s",
+ colwidth,
+ pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), colwidth,
+ PL_colors[0], PL_colors[1],
+ (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
+ PERL_PV_ESCAPE_FIRSTCHAR
+ )
+ );
}
}
PerlIO_printf( Perl_debug_log, "\n%*sState+-",(int)depth * 2 + 2,"" );
for( charid=0 ; charid < trie->uniquecharcount ; charid++ ) {
- PerlIO_printf( Perl_debug_log, "%4s-", "----" );
+ PerlIO_printf( Perl_debug_log, "%.*s", colwidth,"--------");
}
PerlIO_printf( Perl_debug_log, "\n" );
(UV)TRIE_NODENUM( state ) );
for( charid = 0 ; charid < trie->uniquecharcount ; charid++ ) {
- PerlIO_printf( Perl_debug_log, "%4"UVXf" ",
- (UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next ) );
+ UV v=(UV)SAFE_TRIE_NODENUM( trie->trans[ state + charid ].next );
+ if (v)
+ PerlIO_printf( Perl_debug_log, "%*"UVXf, colwidth, v );
+ else
+ PerlIO_printf( Perl_debug_log, "%*s", colwidth, "." );
}
if ( ! trie->states[ TRIE_NODENUM( state ) ].wordnum ) {
PerlIO_printf( Perl_debug_log, " (%4"UVXf")\n", (UV)trie->trans[ state ].check );
#endif
-#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
- ( ( base + charid >= ucharcount \
- && base + charid < ubound \
- && state == trie->trans[ base - ucharcount + charid ].check \
- && trie->trans[ base - ucharcount + charid ].next ) \
- ? trie->trans[ base - ucharcount + charid ].next \
- : ( state==1 ? special : 0 ) \
- )
+/* make_trie(startbranch,first,last,tail,word_count,flags,depth)
+ 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
+ count : words in the sequence
+ flags : currently the OP() type we will be building one of /EXACT(|F|Fl)/
+ depth : indent depth
-STATIC void
-S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
-{
-/* The Trie is constructed and compressed now so we can build a fail array now if its needed
+Inplace optimizes a sequence of 2 or more Branch-Exact nodes into a TRIE node.
- This is apparently the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
- "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
- ISBN 0-201-10088-6
+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:
- We find the fail state for each state in the trie, this state is the longest proper
- suffix of the current states 'word' that is also a proper prefix of another word in our
- trie. State 1 represents the word '' and is the thus the default fail state. This allows
- the DFA not to have to restart after its tried and failed a word at a given point, it
- simply continues as though it had been matching the other word in the first place.
- Consider
- 'abcdgu'=~/abcdefg|cdgu/
- When we get to 'd' we are still matching the first word, we would encounter 'g' which would
- fail, which would bring use to the state representing 'd' in the second word where we would
- try 'g' and succeed, prodceding to match 'cdgu'.
- */
- /* add a fail transition */
- reg_trie_data *trie=(reg_trie_data *)RExC_rx->data->data[ARG(source)];
- U32 *q;
- const U32 ucharcount = trie->uniquecharcount;
- const U32 numstates = trie->laststate;
- const U32 ubound = trie->lasttrans + ucharcount;
- U32 q_read = 0;
- U32 q_write = 0;
- U32 charid;
- U32 base = trie->states[ 1 ].trans.base;
- U32 *fail;
- reg_ac_data *aho;
- const U32 data_slot = add_data( pRExC_state, 1, "T" );
- GET_RE_DEBUG_FLAGS_DECL;
+ /he|she|his|hers/
- ARG_SET( stclass, data_slot );
- Newxz( aho, 1, reg_ac_data );
- RExC_rx->data->data[ data_slot ] = (void*)aho;
- aho->trie=trie;
- aho->states=(reg_trie_state *)savepvn((const char*)trie->states,
- (trie->laststate+1)*sizeof(reg_trie_state));
- Newxz( q, numstates, U32);
- Newxz( aho->fail, numstates, U32 );
- fail= aho->fail;
- fail[ 0 ] = fail[ 1 ] = 1;
+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.
- for ( charid = 0; charid < ucharcount ; charid++ ) {
- const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
- if ( newstate ) {
- q[ q_write ] = newstate;
- /* set to point at the root */
- fail[ q[ q_write++ ] ]=1;
- }
- }
- while ( q_read < q_write) {
- const U32 cur = q[ q_read++ % numstates ];
- base = trie->states[ cur ].trans.base;
+ +-h->+-e->[3]-+-r->(8)-+-s->[9]
+ | |
+ | (2)
+ | |
+ (1) +-i->(6)-+-s->[7]
+ |
+ +-s->(3)-+-h->(4)-+-e->[5]
- for ( charid = 0 ; charid < ucharcount ; charid++ ) {
- const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
- if (ch_state) {
- U32 fail_state = cur;
- U32 fail_base;
- do {
- fail_state = fail[ fail_state ];
- fail_base = aho->states[ fail_state ].trans.base;
- } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
+ 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 <ac>(16)
+ 8: BRANCH(11)
+ 9: EXACT <ad>(16)
+ 11: BRANCH(14)
+ 12: EXACT <ab>(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]
+ <ac>
+ <ad>
+ <ab>
+ 16: SUCCEED(0)
+ 17: NOTHING(18)
+ 18: END(0)
+
+Cases where tail != last would be like /(?foo|bar)baz/:
+
+ 1: BRANCH(4)
+ 2: EXACT <foo>(8)
+ 4: BRANCH(7)
+ 5: EXACT <bar>(8)
+ 7: TAIL(8)
+ 8: EXACT <baz>(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]
+ <foo>
+ <bar>
+ 7: TAIL(8)
+ 8: EXACT <baz>(10)
+ 10: END(0)
+
+ d = uvuni_to_utf8_flags(d, uv, 0);
+
+is the recommended Unicode-aware way of saying
+
+ *(d++) = uv;
+*/
+
+#define TRIE_STORE_REVCHAR \
+ STMT_START { \
+ SV *tmp = newSVpvs(""); \
+ if (UTF) SvUTF8_on(tmp); \
+ Perl_sv_catpvf( aTHX_ tmp, "%c", (int)uvc ); \
+ av_push( revcharmap, tmp ); \
+ } STMT_END
+
+#define TRIE_READ_CHAR STMT_START { \
+ wordlen++; \
+ 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_PUSH(state,fid,ns) STMT_START { \
+ if ( TRIE_LIST_CUR( state ) >=TRIE_LIST_LEN( state ) ) { \
+ U32 ging = TRIE_LIST_LEN( state ) *= 2; \
+ Renew( trie->states[ state ].trans.list, ging, 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 { \
+ Newxz( trie->states[ state ].trans.list, \
+ 4, reg_trie_trans_le ); \
+ TRIE_LIST_CUR( state ) = 1; \
+ TRIE_LIST_LEN( state ) = 4; \
+} STMT_END
- fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
- fail[ ch_state ] = fail_state;
- if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
- {
- aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
- }
- q[ q_write++ % numstates] = ch_state;
- }
- }
- }
+#define TRIE_HANDLE_WORD(state) STMT_START { \
+ U16 dupe= trie->states[ state ].wordnum; \
+ regnode * const noper_next = regnext( noper ); \
+ \
+ if (trie->wordlen) \
+ trie->wordlen[ curword ] = wordlen; \
+ DEBUG_r({ \
+ /* store the word for dumping */ \
+ SV* tmp; \
+ if (OP(noper) != NOTHING) \
+ tmp = newSVpvn(STRING(noper), STR_LEN(noper)); \
+ else \
+ tmp = newSVpvn( "", 0 ); \
+ if ( UTF ) SvUTF8_on( tmp ); \
+ av_push( trie_words, tmp ); \
+ }); \
+ \
+ curword++; \
+ \
+ if ( noper_next < tail ) { \
+ if (!trie->jump) \
+ trie->jump = PerlMemShared_calloc( word_count + 1, sizeof(U16) ); \
+ trie->jump[curword] = (U16)(noper_next - convert); \
+ if (!jumper) \
+ jumper = noper_next; \
+ if (!nextbranch) \
+ nextbranch= regnext(cur); \
+ } \
+ \
+ if ( dupe ) { \
+ /* So it's a dupe. This means we need to maintain a */\
+ /* linked-list from the first to the next. */\
+ /* we only allocate the nextword buffer when there */\
+ /* a dupe, so first time we have to do the allocation */\
+ if (!trie->nextword) \
+ trie->nextword = \
+ PerlMemShared_calloc( word_count + 1, sizeof(U16)); \
+ while ( trie->nextword[dupe] ) \
+ dupe= trie->nextword[dupe]; \
+ trie->nextword[dupe]= curword; \
+ } else { \
+ /* we haven't inserted this word yet. */ \
+ trie->states[ state ].wordnum = curword; \
+ } \
+} STMT_END
- DEBUG_TRIE_COMPILE_MORE_r({
- PerlIO_printf(Perl_debug_log, "%*sFail: 1", (int)(depth * 2), "");
- for( q_read=2; q_read<numstates; q_read++ ) {
- PerlIO_printf(Perl_debug_log, ", %"UVuf, fail[q_read]);
- }
- PerlIO_printf(Perl_debug_log, "\n");
- });
- Safefree(q);
- /*RExC_seen |= REG_SEEN_TRIEDFA;*/
-}
+#define TRIE_TRANS_STATE(state,base,ucharcount,charid,special) \
+ ( ( base + charid >= ucharcount \
+ && base + charid < ubound \
+ && state == trie->trans[ base - ucharcount + charid ].check \
+ && trie->trans[ base - ucharcount + charid ].next ) \
+ ? trie->trans[ base - ucharcount + charid ].next \
+ : ( state==1 ? special : 0 ) \
+ )
+#define MADE_TRIE 1
+#define MADE_JUMP_TRIE 2
+#define MADE_EXACT_TRIE 4
STATIC I32
-S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 flags, U32 depth)
+S_make_trie(pTHX_ RExC_state_t *pRExC_state, regnode *startbranch, regnode *first, regnode *last, regnode *tail, U32 word_count, U32 flags, U32 depth)
{
dVAR;
/* first pass, loop through and scan words */
reg_trie_data *trie;
+ HV *widecharmap = NULL;
+ AV *revcharmap = newAV();
regnode *cur;
const U32 uniflags = UTF8_ALLOW_DEFAULT;
STRLEN len = 0;
UV uvc = 0;
U16 curword = 0;
U32 next_alloc = 0;
+ regnode *jumper = NULL;
+ regnode *nextbranch = NULL;
+ regnode *convert = NULL;
/* we just use folder as a flag in utf8 */
const U8 * const folder = ( flags == EXACTF
? PL_fold
)
);
- const U32 data_slot = add_data( pRExC_state, 1, "t" );
- SV *re_trie_maxbuff;
-#ifndef DEBUGGING
- /* these are only used during construction but are useful during
- * debugging so we store them in the struct when debugging.
- * Wordcount is actually superfluous in debugging as we have
- * (AV*)trie->words to use for it, but that's not available when
- * not debugging... We could make the macro use the AV during
- * debugging though...
+#ifdef DEBUGGING
+ const U32 data_slot = add_data( pRExC_state, 4, "tuuu" );
+ AV *trie_words = NULL;
+ /* along with revcharmap, this only used during construction but both are
+ * useful during debugging so we store them in the struct when debugging.
*/
- U16 trie_wordcount=0;
+#else
+ const U32 data_slot = add_data( pRExC_state, 2, "tu" );
STRLEN trie_charcount=0;
- /*U32 trie_laststate=0;*/
- AV *trie_revcharmap;
#endif
+ SV *re_trie_maxbuff;
GET_RE_DEBUG_FLAGS_DECL;
+#ifndef DEBUGGING
+ PERL_UNUSED_ARG(depth);
+#endif
- Newxz( trie, 1, reg_trie_data );
+ trie = PerlMemShared_calloc( 1, sizeof(reg_trie_data) );
trie->refcount = 1;
trie->startstate = 1;
- RExC_rx->data->data[ data_slot ] = (void*)trie;
- Newxz( trie->charmap, 256, U16 );
+ trie->wordcount = word_count;
+ RExC_rxi->data->data[ data_slot ] = (void*)trie;
+ trie->charmap = PerlMemShared_calloc( 256, sizeof(U16) );
if (!(UTF && folder))
- Newxz( trie->bitmap, ANYOF_BITMAP_SIZE, char );
+ trie->bitmap = PerlMemShared_calloc( ANYOF_BITMAP_SIZE, 1 );
DEBUG_r({
- trie->words = newAV();
+ trie_words = newAV();
});
- TRIE_REVCHARMAP(trie) = newAV();
re_trie_maxbuff = get_sv(RE_TRIE_MAXBUF_NAME, 1);
if (!SvIOK(re_trie_maxbuff)) {
}
DEBUG_OPTIMISE_r({
PerlIO_printf( Perl_debug_log,
- "%*smake_trie start==%d, first==%d, last==%d, tail==%d\n",
+ "%*smake_trie start==%d, first==%d, last==%d, tail==%d depth=%d\n",
(int)depth * 2 + 2, "",
REG_NODE_NUM(startbranch),REG_NODE_NUM(first),
- REG_NODE_NUM(last), REG_NODE_NUM(tail));
+ REG_NODE_NUM(last), REG_NODE_NUM(tail),
+ (int)depth);
});
+
+ /* Find the node we are going to overwrite */
+ if ( first == startbranch && OP( last ) != BRANCH ) {
+ /* whole branch chain */
+ convert = first;
+ } else {
+ /* branch sub-chain */
+ convert = NEXTOPER( first );
+ }
+
/* -- First loop and Setup --
We first traverse the branches and scan each word to determine if it
U32 wordlen = 0; /* required init */
STRLEN chars=0;
- TRIE_WORDCOUNT(trie)++;
if (OP(noper) == NOTHING) {
trie->minlen= 0;
continue;
}
} else {
SV** svpp;
- if ( !trie->widecharmap )
- trie->widecharmap = newHV();
+ if ( !widecharmap )
+ widecharmap = newHV();
- svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 1 );
+ svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 1 );
if ( !svpp )
Perl_croak( aTHX_ "error creating/fetching widecharmap entry for 0x%"UVXf, uvc );
DEBUG_TRIE_COMPILE_r(
PerlIO_printf( Perl_debug_log, "%*sTRIE(%s): W:%d C:%d Uq:%d Min:%d Max:%d\n",
(int)depth * 2 + 2,"",
- ( trie->widecharmap ? "UTF8" : "NATIVE" ), TRIE_WORDCOUNT(trie),
+ ( widecharmap ? "UTF8" : "NATIVE" ), (int)word_count,
(int)TRIE_CHARCOUNT(trie), trie->uniquecharcount,
(int)trie->minlen, (int)trie->maxlen )
);
- Newxz( trie->wordlen, TRIE_WORDCOUNT(trie), U32 );
+ trie->wordlen = PerlMemShared_calloc( word_count, sizeof(U32) );
/*
We now know what we are dealing with in terms of unique chars and
STRLEN transcount = 1;
- Newxz( trie->states, TRIE_CHARCOUNT(trie) + 2, reg_trie_state );
+ DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
+ "%*sCompiling trie using list compiler\n",
+ (int)depth * 2 + 2, ""));
+
+ trie->states = PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
+ sizeof(reg_trie_state) );
TRIE_LIST_NEW(1);
next_alloc = 2;
U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
if (OP(noper) != NOTHING) {
- for ( ; uc < e ; uc += len ) {
+ for ( ; uc < e ; uc += len ) {
- TRIE_READ_CHAR;
+ TRIE_READ_CHAR;
- if ( uvc < 256 ) {
- charid = trie->charmap[ uvc ];
- } else {
- SV** const svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 0);
- if ( !svpp ) {
- charid = 0;
+ if ( uvc < 256 ) {
+ charid = trie->charmap[ uvc ];
} else {
- charid=(U16)SvIV( *svpp );
+ SV** const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
+ if ( !svpp ) {
+ charid = 0;
+ } else {
+ charid=(U16)SvIV( *svpp );
+ }
}
- }
- if ( charid ) {
+ /* charid is now 0 if we dont know the char read, or nonzero if we do */
+ if ( charid ) {
- U16 check;
- U32 newstate = 0;
+ 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;
+ 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 );
}
- 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 */
- }
}
TRIE_HANDLE_WORD(state);
} /* end second pass */
- TRIE_LASTSTATE(trie) = next_alloc;
- Renew( trie->states, next_alloc, reg_trie_state );
+ /* next alloc is the NEXT state to be allocated */
+ trie->statecount = next_alloc;
+ trie->states = PerlMemShared_realloc( trie->states, next_alloc
+ * sizeof(reg_trie_state) );
/* and now dump it out before we compress it */
- DEBUG_TRIE_COMPILE_MORE_r(
- dump_trie_interim_list(trie,next_alloc,depth+1)
- );
+ DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_list(trie, widecharmap,
+ revcharmap, next_alloc,
+ depth+1)
+ );
- Newxz( trie->trans, transcount ,reg_trie_trans );
+ trie->trans
+ = PerlMemShared_calloc( transcount, sizeof(reg_trie_trans) );
{
U32 state;
U32 tp = 0;
}
if ( transcount < tp + maxid - minid + 1) {
transcount *= 2;
- Renew( trie->trans, transcount, reg_trie_trans );
+ trie->trans
+ = PerlMemShared_realloc( trie->trans,
+ transcount
+ * sizeof(reg_trie_trans) );
Zero( trie->trans + (transcount / 2), transcount / 2 , reg_trie_trans );
}
base = trie->uniquecharcount + tp - minid;
use TRIE_NODENUM() to convert.
*/
-
-
- Newxz( trie->trans, ( TRIE_CHARCOUNT(trie) + 1 ) * trie->uniquecharcount + 1,
- reg_trie_trans );
- Newxz( trie->states, TRIE_CHARCOUNT(trie) + 2, reg_trie_state );
+ DEBUG_TRIE_COMPILE_MORE_r( PerlIO_printf( Perl_debug_log,
+ "%*sCompiling trie using table compiler\n",
+ (int)depth * 2 + 2, ""));
+
+ trie->trans = PerlMemShared_calloc( ( TRIE_CHARCOUNT(trie) + 1 )
+ * trie->uniquecharcount + 1,
+ sizeof(reg_trie_trans) );
+ trie->states = PerlMemShared_calloc( TRIE_CHARCOUNT(trie) + 2,
+ sizeof(reg_trie_state) );
next_alloc = trie->uniquecharcount + 1;
U8 foldbuf[ UTF8_MAXBYTES_CASE + 1 ];
if ( OP(noper) != NOTHING ) {
- for ( ; uc < e ; uc += len ) {
+ for ( ; uc < e ; uc += len ) {
- TRIE_READ_CHAR;
+ TRIE_READ_CHAR;
- if ( uvc < 256 ) {
- charid = trie->charmap[ uvc ];
- } else {
- SV* const * const svpp = hv_fetch( trie->widecharmap, (char*)&uvc, sizeof( UV ), 0);
- charid = svpp ? (U16)SvIV(*svpp) : 0;
- }
- if ( charid ) {
- charid--;
- if ( !trie->trans[ state + charid ].next ) {
- trie->trans[ state + charid ].next = next_alloc;
- trie->trans[ state ].check++;
- next_alloc += trie->uniquecharcount;
+ if ( uvc < 256 ) {
+ charid = trie->charmap[ uvc ];
+ } else {
+ SV* const * const svpp = hv_fetch( widecharmap, (char*)&uvc, sizeof( UV ), 0);
+ charid = svpp ? (U16)SvIV(*svpp) : 0;
}
- state = trie->trans[ state + charid ].next;
- } else {
- Perl_croak( aTHX_ "panic! In trie construction, no char mapping for %"IVdf, uvc );
+ 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 */
}
- /* charid is now 0 if we dont know the char read, or nonzero if we do */
- }
}
accept_state = TRIE_NODENUM( state );
TRIE_HANDLE_WORD(accept_state);
} /* end second pass */
/* and now dump it out before we compress it */
- DEBUG_TRIE_COMPILE_MORE_r(
- dump_trie_interim_table(trie,next_alloc,depth+1)
- );
+ DEBUG_TRIE_COMPILE_MORE_r(dump_trie_interim_table(trie, widecharmap,
+ revcharmap,
+ next_alloc, depth+1));
{
/*
even earlier), but the .check field determines if the transition is
valid.
+ XXX - wrong maybe?
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
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.
+ pointer doesnt find one the single transition is appended as normal.
- Once compressed we can Renew/realloc the structures to release the
excess space.
const U32 laststate = TRIE_NODENUM( next_alloc );
U32 state, charid;
U32 pos = 0, zp=0;
- TRIE_LASTSTATE(trie) = laststate;
+ trie->statecount = laststate;
for ( state = 1 ; state < laststate ; state++ ) {
U8 flag = 0;
}
}
trie->lasttrans = pos + 1;
- Renew( trie->states, laststate + 1, reg_trie_state);
+ trie->states = PerlMemShared_realloc( trie->states, laststate
+ * sizeof(reg_trie_state) );
DEBUG_TRIE_COMPILE_MORE_r(
PerlIO_printf( Perl_debug_log,
"%*sAlloc: %d Orig: %"IVdf" elements, Final:%"IVdf". Savings of %%%5.2f\n",
} /* end table compress */
}
+ DEBUG_TRIE_COMPILE_MORE_r(
+ PerlIO_printf(Perl_debug_log, "%*sStatecount:%"UVxf" Lasttrans:%"UVxf"\n",
+ (int)depth * 2 + 2, "",
+ (UV)trie->statecount,
+ (UV)trie->lasttrans)
+ );
/* resize the trans array to remove unused space */
- Renew( trie->trans, trie->lasttrans, reg_trie_trans);
+ trie->trans = PerlMemShared_realloc( trie->trans, trie->lasttrans
+ * sizeof(reg_trie_trans) );
/* and now dump out the compressed format */
- DEBUG_TRIE_COMPILE_r(
- dump_trie(trie,depth+1)
- );
+ DEBUG_TRIE_COMPILE_r(dump_trie(trie, widecharmap, revcharmap, depth+1));
{ /* Modify the program and insert the new TRIE node*/
- regnode *convert;
U8 nodetype =(U8)(flags & 0xFF);
char *str=NULL;
+
#ifdef DEBUGGING
- U32 mjd_offset;
- U32 mjd_nodelen;
+ regnode *optimize = NULL;
+ U32 mjd_offset = 0;
+ U32 mjd_nodelen = 0;
#endif
/*
This means we convert either the first branch or the first Exact,
the whole branch sequence, including the first.
*/
/* Find the node we are going to overwrite */
- if ( first == startbranch && OP( last ) != BRANCH ) {
- /* whole branch chain */
- convert = first;
- DEBUG_r({
- const regnode *nop = NEXTOPER( convert );
- mjd_offset= Node_Offset((nop));
- mjd_nodelen= Node_Length((nop));
- });
- } else {
+ if ( first != startbranch || OP( last ) == BRANCH ) {
/* branch sub-chain */
- convert = NEXTOPER( first );
NEXT_OFF( first ) = (U16)(last - first);
DEBUG_r({
mjd_offset= Node_Offset((convert));
mjd_nodelen= Node_Length((convert));
});
+ /* whole branch chain */
+ } else {
+ DEBUG_r({
+ const regnode *nop = NEXTOPER( convert );
+ mjd_offset= Node_Offset((nop));
+ mjd_nodelen= Node_Length((nop));
+ });
}
+
DEBUG_OPTIMISE_r(
PerlIO_printf(Perl_debug_log, "%*sMJD offset:%"UVuf" MJD length:%"UVuf"\n",
(int)depth * 2 + 2, "",
- mjd_offset,mjd_nodelen)
+ (UV)mjd_offset, (UV)mjd_nodelen)
);
/* But first we check to see if there is a common prefix we can
split out as an EXACT and put in front of the TRIE node. */
trie->startstate= 1;
- if ( trie->bitmap && !trie->widecharmap ) {
+ if ( trie->bitmap && !widecharmap && !trie->jump ) {
U32 state;
- DEBUG_OPTIMISE_r(
- PerlIO_printf(Perl_debug_log, "%*sLaststate:%"UVuf"\n",
- (int)depth * 2 + 2, "",
- TRIE_LASTSTATE(trie))
- );
- for ( state = 1 ; state < TRIE_LASTSTATE(trie)-1 ; state++ ) {
+ for ( state = 1 ; state < trie->statecount-1 ; state++ ) {
U32 ofs = 0;
I32 idx = -1;
U32 count = 0;
trie->trans[ base + ofs - trie->uniquecharcount ].check == state )
{
if ( ++count > 1 ) {
- SV **tmp = av_fetch( TRIE_REVCHARMAP(trie), ofs, 0);
+ SV **tmp = av_fetch( revcharmap, ofs, 0);
const U8 *ch = (U8*)SvPV_nolen_const( *tmp );
if ( state == 1 ) break;
if ( count == 2 ) {
PerlIO_printf(Perl_debug_log,
"%*sNew Start State=%"UVuf" Class: [",
(int)depth * 2 + 2, "",
- state));
+ (UV)state));
if (idx >= 0) {
- SV ** const tmp = av_fetch( TRIE_REVCHARMAP(trie), idx, 0);
+ SV ** const tmp = av_fetch( revcharmap, idx, 0);
const U8 * const ch = (U8*)SvPV_nolen_const( *tmp );
TRIE_BITMAP_SET(trie,*ch);
}
}
if ( count == 1 ) {
- SV **tmp = av_fetch( TRIE_REVCHARMAP(trie), idx, 0);
- const char *ch = SvPV_nolen_const( *tmp );
- DEBUG_OPTIMISE_r(
+ SV **tmp = av_fetch( revcharmap, idx, 0);
+ char *ch = SvPV_nolen( *tmp );
+ DEBUG_OPTIMISE_r({
+ SV *sv=sv_newmortal();
PerlIO_printf( Perl_debug_log,
"%*sPrefix State: %"UVuf" Idx:%"UVuf" Char='%s'\n",
(int)depth * 2 + 2, "",
- state, idx, ch)
- );
+ (UV)state, (UV)idx,
+ pv_pretty(sv, SvPV_nolen_const(*tmp), SvCUR(*tmp), 6,
+ PL_colors[0], PL_colors[1],
+ (SvUTF8(*tmp) ? PERL_PV_ESCAPE_UNI : 0) |
+ PERL_PV_ESCAPE_FIRSTCHAR
+ )
+ );
+ });
if ( state==1 ) {
OP( convert ) = nodetype;
str=STRING(convert);
STR_LEN(convert)=0;
}
- *str++=*ch;
- STR_LEN(convert)++;
-
+ while (*ch) {
+ *str++ = *ch++;
+ STR_LEN(convert)++;
+ }
+
} else {
+#ifdef DEBUGGING
if (state>1)
DEBUG_OPTIMISE_r(PerlIO_printf( Perl_debug_log,"]\n"));
+#endif
break;
}
}
trie->maxlen -= (state - 1);
DEBUG_r({
regnode *fix = convert;
+ U32 word = trie->wordcount;
mjd_nodelen++;
Set_Node_Offset_Length(convert, mjd_offset, state - 1);
while( ++fix < n ) {
Set_Node_Offset_Length(fix, 0, 0);
}
+ while (word--) {
+ SV ** const tmp = av_fetch( trie_words, word, 0 );
+ if (tmp) {
+ if ( STR_LEN(convert) <= SvCUR(*tmp) )
+ sv_chop(*tmp, SvPV_nolen(*tmp) + STR_LEN(convert));
+ else
+ sv_chop(*tmp, SvPV_nolen(*tmp) + SvCUR(*tmp));
+ }
+ }
});
if (trie->maxlen) {
convert = n;
} else {
NEXT_OFF(convert) = (U16)(tail - convert);
+ DEBUG_r(optimize= n);
}
}
}
+ if (!jumper)
+ jumper = last;
if ( trie->maxlen ) {
- OP( convert ) = TRIE;
NEXT_OFF( convert ) = (U16)(tail - convert);
ARG_SET( convert, data_slot );
+ /* Store the offset to the first unabsorbed branch in
+ jump[0], which is otherwise unused by the jump logic.
+ We use this when dumping a trie and during optimisation. */
+ if (trie->jump)
+ trie->jump[0] = (U16)(nextbranch - convert);
+
+ /* XXXX */
+ if ( !trie->states[trie->startstate].wordnum && trie->bitmap &&
+ ( (char *)jumper - (char *)convert) >= (int)sizeof(struct regnode_charclass) )
+ {
+ OP( convert ) = TRIEC;
+ Copy(trie->bitmap, ((struct regnode_charclass *)convert)->bitmap, ANYOF_BITMAP_SIZE, char);
+ PerlMemShared_free(trie->bitmap);
+ trie->bitmap= NULL;
+ } else
+ OP( convert ) = TRIE;
/* store the type in the flags */
convert->flags = nodetype;
- /* XXX We really should free up the resource in trie now, as we wont use them */
+ DEBUG_r({
+ optimize = convert
+ + NODE_STEP_REGNODE
+ + regarglen[ OP( convert ) ];
+ });
+ /* XXX We really should free up the resource in trie now,
+ as we won't use them - (which resources?) dmq */
}
/* needed for dumping*/
- DEBUG_r({
- regnode *optimize = convert + NODE_STEP_REGNODE + regarglen[ TRIE ];
+ DEBUG_r(if (optimize) {
regnode *opt = convert;
- while (++opt<optimize) {
+ while ( ++opt < optimize) {
Set_Node_Offset_Length(opt,0,0);
}
- /* 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.
- We also "fix" the offsets
+ /*
+ Try to clean up some of the debris left after the
+ optimisation.
*/
- while( optimize < last ) {
+ while( optimize < jumper ) {
mjd_nodelen += Node_Length((optimize));
OP( optimize ) = OPTIMIZED;
Set_Node_Offset_Length(optimize,0,0);
Set_Node_Offset_Length(convert,mjd_offset,mjd_nodelen);
});
} /* end node insert */
+ RExC_rxi->data->data[ data_slot + 1 ] = (void*)widecharmap;
+#ifdef DEBUGGING
+ RExC_rxi->data->data[ data_slot + TRIE_WORDS_OFFSET ] = (void*)trie_words;
+ RExC_rxi->data->data[ data_slot + 3 ] = (void*)revcharmap;
+#else
+ SvREFCNT_dec(revcharmap);
+#endif
+ return trie->jump
+ ? MADE_JUMP_TRIE
+ : trie->startstate>1
+ ? MADE_EXACT_TRIE
+ : MADE_TRIE;
+}
+
+STATIC void
+S_make_trie_failtable(pTHX_ RExC_state_t *pRExC_state, regnode *source, regnode *stclass, U32 depth)
+{
+/* The Trie is constructed and compressed now so we can build a fail array now if its needed
+
+ This is basically the Aho-Corasick algorithm. Its from exercise 3.31 and 3.32 in the
+ "Red Dragon" -- Compilers, principles, techniques, and tools. Aho, Sethi, Ullman 1985/88
+ ISBN 0-201-10088-6
+
+ We find the fail state for each state in the trie, this state is the longest proper
+ suffix of the current states 'word' that is also a proper prefix of another word in our
+ trie. State 1 represents the word '' and is the thus the default fail state. This allows
+ the DFA not to have to restart after its tried and failed a word at a given point, it
+ simply continues as though it had been matching the other word in the first place.
+ Consider
+ 'abcdgu'=~/abcdefg|cdgu/
+ When we get to 'd' we are still matching the first word, we would encounter 'g' which would
+ fail, which would bring use to the state representing 'd' in the second word where we would
+ try 'g' and succeed, prodceding to match 'cdgu'.
+ */
+ /* add a fail transition */
+ const U32 trie_offset = ARG(source);
+ reg_trie_data *trie=(reg_trie_data *)RExC_rxi->data->data[trie_offset];
+ U32 *q;
+ const U32 ucharcount = trie->uniquecharcount;
+ const U32 numstates = trie->statecount;
+ const U32 ubound = trie->lasttrans + ucharcount;
+ U32 q_read = 0;
+ U32 q_write = 0;
+ U32 charid;
+ U32 base = trie->states[ 1 ].trans.base;
+ U32 *fail;
+ reg_ac_data *aho;
+ const U32 data_slot = add_data( pRExC_state, 1, "T" );
+ GET_RE_DEBUG_FLAGS_DECL;
#ifndef DEBUGGING
- SvREFCNT_dec(TRIE_REVCHARMAP(trie));
+ PERL_UNUSED_ARG(depth);
#endif
- return 1;
+
+
+ ARG_SET( stclass, data_slot );
+ aho = PerlMemShared_calloc( 1, sizeof(reg_ac_data) );
+ RExC_rxi->data->data[ data_slot ] = (void*)aho;
+ aho->trie=trie_offset;
+ aho->states=(reg_trie_state *)PerlMemShared_malloc( numstates * sizeof(reg_trie_state) );
+ Copy( trie->states, aho->states, numstates, reg_trie_state );
+ Newxz( q, numstates, U32);
+ aho->fail = PerlMemShared_calloc( numstates, sizeof(U32) );
+ aho->refcount = 1;
+ fail = aho->fail;
+ /* initialize fail[0..1] to be 1 so that we always have
+ a valid final fail state */
+ fail[ 0 ] = fail[ 1 ] = 1;
+
+ for ( charid = 0; charid < ucharcount ; charid++ ) {
+ const U32 newstate = TRIE_TRANS_STATE( 1, base, ucharcount, charid, 0 );
+ if ( newstate ) {
+ q[ q_write ] = newstate;
+ /* set to point at the root */
+ fail[ q[ q_write++ ] ]=1;
+ }
+ }
+ while ( q_read < q_write) {
+ const U32 cur = q[ q_read++ % numstates ];
+ base = trie->states[ cur ].trans.base;
+
+ for ( charid = 0 ; charid < ucharcount ; charid++ ) {
+ const U32 ch_state = TRIE_TRANS_STATE( cur, base, ucharcount, charid, 1 );
+ if (ch_state) {
+ U32 fail_state = cur;
+ U32 fail_base;
+ do {
+ fail_state = fail[ fail_state ];
+ fail_base = aho->states[ fail_state ].trans.base;
+ } while ( !TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 ) );
+
+ fail_state = TRIE_TRANS_STATE( fail_state, fail_base, ucharcount, charid, 1 );
+ fail[ ch_state ] = fail_state;
+ if ( !aho->states[ ch_state ].wordnum && aho->states[ fail_state ].wordnum )
+ {
+ aho->states[ ch_state ].wordnum = aho->states[ fail_state ].wordnum;
+ }
+ q[ q_write++ % numstates] = ch_state;
+ }
+ }
+ }
+ /* restore fail[0..1] to 0 so that we "fall out" of the AC loop
+ when we fail in state 1, this allows us to use the
+ charclass scan to find a valid start char. This is based on the principle
+ that theres a good chance the string being searched contains lots of stuff
+ that cant be a start char.
+ */
+ fail[ 0 ] = fail[ 1 ] = 0;
+ DEBUG_TRIE_COMPILE_r({
+ PerlIO_printf(Perl_debug_log,
+ "%*sStclass Failtable (%"UVuf" states): 0",
+ (int)(depth * 2), "", (UV)numstates
+ );
+ for( q_read=1; q_read<numstates; q_read++ ) {
+ PerlIO_printf(Perl_debug_log, ", %"UVuf, (UV)fail[q_read]);
+ }
+ PerlIO_printf(Perl_debug_log, "\n");
+ });
+ Safefree(q);
+ /*RExC_seen |= REG_SEEN_TRIEDFA;*/
}
+
/*
* 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.
#endif
#define DEBUG_PEEP(str,scan,depth) \
- DEBUG_OPTIMISE_r({ \
+ DEBUG_OPTIMISE_r({if (scan){ \
SV * const mysv=sv_newmortal(); \
regnode *Next = regnext(scan); \
regprop(RExC_rx, mysv, scan); \
- PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s [%d]\n", \
+ PerlIO_printf(Perl_debug_log, "%*s" str ">%3d: %s (%d)\n", \
(int)depth*2, "", REG_NODE_NUM(scan), SvPV_nolen_const(mysv),\
Next ? (REG_NODE_NUM(Next)) : 0 ); \
- });
+ }});
+
+
+
+
#define JOIN_EXACT(scan,min,flags) \
if (PL_regkind[OP(scan)] == EXACT) \
U32 stopnow = 0;
#ifdef DEBUGGING
regnode *stop = scan;
-#endif
GET_RE_DEBUG_FLAGS_DECL;
+#else
+ PERL_UNUSED_ARG(depth);
+#endif
+#ifndef EXPERIMENTAL_INPLACESCAN
+ PERL_UNUSED_ARG(flags);
+ PERL_UNUSED_ARG(val);
+#endif
DEBUG_PEEP("join",scan,depth);
/* Skip NOTHING, merge EXACT*. */
if (OP(n) == TAIL || n > next)
stringok = 0;
if (PL_regkind[OP(n)] == NOTHING) {
-
DEBUG_PEEP("skip:",n,depth);
NEXT_OFF(scan) += NEXT_OFF(n);
next = n + NODE_STEP_REGNODE;
n = regnext(n);
}
else if (stringok) {
- const int oldl = STR_LEN(scan);
+ const unsigned int oldl = STR_LEN(scan);
regnode * const nnext = regnext(n);
DEBUG_PEEP("merg",n,depth);
if (stopnow) break;
}
-#ifdef EXPERIMENTAL_INPLACESCAN
- if (flags && !NEXT_OFF(n)) {
- DEBUG_PEEP("atch",val,depth);
- if (reg_off_by_arg[OP(n)]) {
- ARG_SET(n, val - n);
- }
- else {
- NEXT_OFF(n) = val - n;
- }
- stopnow=1;
- }
+#ifdef EXPERIMENTAL_INPLACESCAN
+ if (flags && !NEXT_OFF(n)) {
+ DEBUG_PEEP("atch", val, depth);
+ if (reg_off_by_arg[OP(n)]) {
+ ARG_SET(n, val - n);
+ }
+ else {
+ NEXT_OFF(n) = val - n;
+ }
+ stopnow = 1;
+ }
#endif
}
char * const s0 = STRING(scan), *s, *t;
char * const s1 = s0 + STR_LEN(scan) - 1;
char * const s2 = s1 - 4;
+#ifdef EBCDIC /* RD tunifold greek 0390 and 03B0 */
+ const char t0[] = "\xaf\x49\xaf\x42";
+#else
const char t0[] = "\xcc\x88\xcc\x81";
+#endif
const char * const t1 = t0 + 3;
for (s = s0 + 2;
s < s2 && (t = ninstr(s, s1, t0, t1));
s = t + 4) {
+#ifdef EBCDIC
+ if (((U8)t[-1] == 0x68 && (U8)t[-2] == 0xB4) ||
+ ((U8)t[-1] == 0x46 && (U8)t[-2] == 0xB5))
+#else
if (((U8)t[-1] == 0xB9 && (U8)t[-2] == 0xCE) ||
((U8)t[-1] == 0x85 && (U8)t[-2] == 0xCF))
+#endif
*min -= 4;
}
}
/* Stops at toplevel WHILEM as well as at "last". At end *scanp is set
to the position after last scanned or to NULL. */
-
+#define INIT_AND_WITHP \
+ assert(!and_withp); \
+ Newx(and_withp,1,struct regnode_charclass_class); \
+ SAVEFREEPV(and_withp)
+
+/* this is a chain of data about sub patterns we are processing that
+ need to be handled seperately/specially in study_chunk. Its so
+ we can simulate recursion without losing state. */
+struct scan_frame;
+typedef struct scan_frame {
+ regnode *last; /* last node to process in this frame */
+ regnode *next; /* next node to process when last is reached */
+ struct scan_frame *prev; /*previous frame*/
+ I32 stop; /* what stopparen do we use */
+} scan_frame;
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)
+S_study_chunk(pTHX_ RExC_state_t *pRExC_state, regnode **scanp,
+ I32 *minlenp, I32 *deltap,
+ regnode *last,
+ scan_data_t *data,
+ I32 stopparen,
+ U8* recursed,
+ struct regnode_charclass_class *and_withp,
+ U32 flags, U32 depth)
/* scanp: Start here (read-write). */
/* deltap: Write maxlen-minlen here. */
/* last: Stop before this one. */
+ /* data: string data about the pattern */
+ /* stopparen: treat close N as END */
+ /* recursed: which subroutines have we recursed into */
+ /* and_withp: Valid if flags & SCF_DO_STCLASS_OR */
{
dVAR;
I32 min = 0, pars = 0, code;
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;
+ regnode *first_non_open = scan;
+ I32 stopmin = I32_MAX;
+ scan_frame *frame = NULL;
GET_RE_DEBUG_FLAGS_DECL;
- while (scan && OP(scan) != END && scan < last) {
+#ifdef DEBUGGING
+ StructCopy(&zero_scan_data, &data_fake, scan_data_t);
+#endif
+
+ if ( depth == 0 ) {
+ while (first_non_open && OP(first_non_open) == OPEN)
+ first_non_open=regnext(first_non_open);
+ }
+
+
+ fake_study_recurse:
+ while ( scan && OP(scan) != END && scan < last ){
/* Peephole optimizer: */
+ DEBUG_STUDYDATA(data,depth);
DEBUG_PEEP("Peep",scan,depth);
-
JOIN_EXACT(scan,&min,0);
/* Follow the next-chain of the current node and optimize
/* 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) {
+ || OP(scan) == IFTHEN) {
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) {
+ if (OP(next) == code || code == IFTHEN) {
+ /* NOTE - There is similar code to this block below for handling
+ TRIE nodes on a re-study. If you change stuff here check there
+ too. */
I32 max1 = 0, min1 = I32_MAX, num = 0;
struct regnode_charclass_class accum;
regnode * const startbranch=scan;
- if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
- scan_commit(pRExC_state, data); /* Cannot merge strings after this. */
+ if (flags & SCF_DO_SUBSTR)
+ scan_commit(pRExC_state, data, minlenp); /* Cannot merge strings after this. */
if (flags & SCF_DO_STCLASS)
cl_init_zero(pRExC_state, &accum);
num++;
data_fake.flags = 0;
- if (data) {
+ if (data) {
data_fake.whilem_c = data->whilem_c;
data_fake.last_closep = data->last_closep;
}
else
data_fake.last_closep = &fake;
+
+ data_fake.pos_delta = delta;
next = regnext(scan);
scan = NEXTOPER(scan);
if (code != BRANCH)
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);
+ minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
+ next, &data_fake,
+ stopparen, recursed, NULL, f,depth+1);
if (min1 > minnext)
min1 = minnext;
if (max1 < minnext + deltanext)
scan = next;
if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
pars++;
+ if (data_fake.flags & SCF_SEEN_ACCEPT) {
+ if ( stopmin > minnext)
+ stopmin = min + min1;
+ flags &= ~SCF_DO_SUBSTR;
+ if (data)
+ data->flags |= SCF_SEEN_ACCEPT;
+ }
if (data) {
if (data_fake.flags & SF_HAS_EVAL)
data->flags |= SF_HAS_EVAL;
}
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_STCLASS_OR) {
cl_or(pRExC_state, data->start_class, &accum);
if (min1) {
- cl_and(data->start_class, &and_with);
+ cl_and(data->start_class, and_withp);
flags &= ~SCF_DO_STCLASS;
}
}
else {
/* Switch to OR mode: cache the old value of
* data->start_class */
- StructCopy(data->start_class, &and_with,
+ INIT_AND_WITHP;
+ StructCopy(data->start_class, and_withp,
struct regnode_charclass_class);
flags &= ~SCF_DO_STCLASS_AND;
StructCopy(&accum, data->start_class,
}
}
+ if (PERL_ENABLE_TRIE_OPTIMISATION && OP( startbranch ) == BRANCH ) {
/* demq.
Assuming this was/is a branch we are dealing with: 'scan' now
it is. We now start at the beginning of the sequence and look
for subsequences of
- BRANCH->EXACT=>X
- BRANCH->EXACT=>X
+ BRANCH->EXACT=>x1
+ BRANCH->EXACT=>x2
+ tail
which would be constructed from a pattern like /A|LIST|OF|WORDS/
We have two cases
- 1. patterns where the whole set of branch can be converted to a trie,
+ 1. patterns where the whole set of branch can be converted.
- 2. patterns where only a subset of the alternations can be
- converted to a trie.
+ 2. patterns where only a subset can be converted.
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
'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.
+ There is an additional case, that being where there is a
+ common prefix, which gets split out into an EXACT like node
+ preceding the TRIE node.
+
+ If x(1..n)==tail then we can do a simple trie, if not we make
+ a "jump" trie, such that when we match the appropriate word
+ we "jump" to the appopriate tail node. Essentailly we turn
+ a nested if into a case structure of sorts.
*/
- if (PERL_ENABLE_TRIE_OPTIMISATION) {
+
int made=0;
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 ) {
+ if ( SvIV(re_trie_maxbuff)>=0 ) {
regnode *cur;
regnode *first = (regnode *)NULL;
regnode *last = (regnode *)NULL;
/* dont use tail as the end marker for this traverse */
for ( cur = startbranch ; cur != scan ; cur = regnext( cur ) ) {
regnode * const noper = NEXTOPER( cur );
+#if defined(DEBUGGING) || defined(NOJUMPTRIE)
regnode * const noper_next = regnext( noper );
+#endif
DEBUG_OPTIMISE_r({
regprop(RExC_rx, mysv, cur);
if ( (((first && optype!=NOTHING) ? OP( noper ) == optype
: PL_regkind[ OP( noper ) ] == EXACT )
|| OP(noper) == NOTHING )
- && noper_next == tail && count<U16_MAX)
+#ifdef NOJUMPTRIE
+ && noper_next == tail
+#endif
+ && count < U16_MAX)
{
count++;
if ( !first || optype == NOTHING ) {
}
} else {
if ( last ) {
- made+=make_trie( pRExC_state, startbranch, first, cur, tail, optype, depth+1 );
+ make_trie( pRExC_state,
+ startbranch, first, cur, tail, count,
+ optype, depth+1 );
}
if ( PL_regkind[ OP( noper ) ] == EXACT
- && noper_next == tail )
- {
+#ifdef NOJUMPTRIE
+ && noper_next == tail
+#endif
+ ){
count = 1;
first = cur;
optype = OP( noper );
});
if ( last ) {
- made+= make_trie( pRExC_state, startbranch, first, scan, tail, optype, depth+1 );
+ made= make_trie( pRExC_state, startbranch, first, scan, tail, count, optype, depth+1 );
#ifdef TRIE_STUDY_OPT
- if ( made && startbranch == first ) {
- if ( OP(first)!=TRIE )
- flags |= SCF_EXACT_TRIE;
- else {
- regnode *chk=*scanp;
- while ( OP( chk ) == OPEN )
- chk = regnext( chk );
- if (chk==first)
- flags |= SCF_EXACT_TRIE;
+ if ( ((made == MADE_EXACT_TRIE &&
+ startbranch == first)
+ || ( first_non_open == first )) &&
+ depth==0 ) {
+ flags |= SCF_TRIE_RESTUDY;
+ if ( startbranch == first
+ && scan == tail )
+ {
+ RExC_seen &=~REG_TOP_LEVEL_BRANCHES;
}
- }
+ }
#endif
}
}
} /* do trie */
+
}
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) == SUSPEND || OP(scan) == GOSUB || OP(scan) == GOSTART) {
+ scan_frame *newframe = NULL;
+ I32 paren;
+ regnode *start;
+ regnode *end;
+
+ if (OP(scan) != SUSPEND) {
+ /* set the pointer */
+ if (OP(scan) == GOSUB) {
+ paren = ARG(scan);
+ RExC_recurse[ARG2L(scan)] = scan;
+ start = RExC_open_parens[paren-1];
+ end = RExC_close_parens[paren-1];
+ } else {
+ paren = 0;
+ start = RExC_rxi->program + 1;
+ end = RExC_opend;
+ }
+ if (!recursed) {
+ Newxz(recursed, (((RExC_npar)>>3) +1), U8);
+ SAVEFREEPV(recursed);
+ }
+ if (!PAREN_TEST(recursed,paren+1)) {
+ PAREN_SET(recursed,paren+1);
+ Newx(newframe,1,scan_frame);
+ } else {
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(pRExC_state,data,minlenp);
+ 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 {
+ Newx(newframe,1,scan_frame);
+ paren = stopparen;
+ start = scan+2;
+ end = regnext(scan);
+ }
+ if (newframe) {
+ assert(start);
+ assert(end);
+ SAVEFREEPV(newframe);
+ newframe->next = regnext(scan);
+ newframe->last = last;
+ newframe->stop = stopparen;
+ newframe->prev = frame;
+
+ frame = newframe;
+ scan = start;
+ stopparen = paren;
+ last = end;
+
+ continue;
+ }
}
else if (OP(scan) == EXACT) {
I32 l = STR_LEN(scan);
else
data->start_class->flags |= ANYOF_UNICODE_ALL;
data->start_class->flags &= ~ANYOF_EOS;
- cl_and(data->start_class, &and_with);
+ cl_and(data->start_class, and_withp);
}
flags &= ~SCF_DO_STCLASS;
}
/* Search for fixed substrings supports EXACT only. */
if (flags & SCF_DO_SUBSTR) {
assert(data);
- scan_commit(pRExC_state, data);
+ scan_commit(pRExC_state, data, minlenp);
}
if (UTF) {
const U8 * const s = (U8 *)STRING(scan);
ANYOF_BITMAP_SET(data->start_class, uc);
data->start_class->flags &= ~ANYOF_EOS;
}
- cl_and(data->start_class, &and_with);
+ cl_and(data->start_class, and_withp);
}
flags &= ~SCF_DO_STCLASS;
}
-#ifdef TRIE_STUDY_OPT
- else if (OP(scan) == TRIE) {
- reg_trie_data *trie=RExC_rx->data->data[ ARG(scan) ];
- min += trie->minlen;
- delta += (trie->maxlen - trie->minlen);
- flags &= ~SCF_DO_STCLASS; /* xxx */
- if (flags & SCF_DO_SUBSTR) {
- scan_commit(pRExC_state,data); /* Cannot expect anything... */
- data->pos_min += trie->minlen;
- data->pos_delta += (trie->maxlen - trie->minlen);
- if (trie->maxlen != trie->minlen)
- data->longest = &(data->longest_float);
- }
- }
-#endif
else if (strchr((const char*)PL_varies,OP(scan))) {
I32 mincount, maxcount, minnext, deltanext, fl = 0;
I32 f = flags, pos_before = 0;
is_inf = is_inf_internal = 1;
scan = regnext(scan);
if (flags & SCF_DO_SUBSTR) {
- scan_commit(pRExC_state, data); /* Cannot extend fixed substrings */
+ scan_commit(pRExC_state, data, minlenp); /* Cannot extend fixed substrings */
data->longest = &(data->longest_float);
}
goto optimize_curly_tail;
case CURLY:
- mincount = ARG1(scan);
- maxcount = ARG2(scan);
+ if (stopparen>0 && (OP(scan)==CURLYN || OP(scan)==CURLYM)
+ && (scan->flags == stopparen))
+ {
+ mincount = 1;
+ maxcount = 1;
+ } else {
+ 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->flags = ((lp <= (I32)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 */
+ if (mincount == 0) scan_commit(pRExC_state,data,minlenp); /* Cannot extend fixed substrings */
pos_before = data->pos_min;
}
if (data) {
f &= ~SCF_WHILEM_VISITED_POS;
/* This will finish on WHILEM, setting scan, or on NULL: */
- minnext = study_chunk(pRExC_state, &scan, &deltanext, last, data,
+ minnext = study_chunk(pRExC_state, &scan, minlenp, &deltanext,
+ last, data, stopparen, recursed, NULL,
(mincount == 0
? (f & ~SCF_DO_SUBSTR) : f),depth+1);
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,
+ INIT_AND_WITHP;
+ StructCopy(data->start_class, and_withp,
struct regnode_charclass_class);
flags &= ~SCF_DO_STCLASS_AND;
StructCopy(&this_class, data->start_class,
} 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);
+ cl_and(data->start_class, and_withp);
}
else if (flags & SCF_DO_STCLASS_AND)
cl_and(data->start_class, &this_class);
nxt = regnext(nxt);
if (OP(nxt) != CLOSE)
goto nogo;
+ if (RExC_open_parens) {
+ RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
+ RExC_close_parens[ARG(nxt1)-1]=nxt+2; /*close->while*/
+ }
/* 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. */
if (OP(nxt) != CLOSE)
FAIL("Panic opt close");
oscan->flags = (U8)ARG(nxt);
+ if (RExC_open_parens) {
+ RExC_open_parens[ARG(nxt1)-1]=oscan; /*open->CURLYM*/
+ RExC_close_parens[ARG(nxt1)-1]=nxt2+1; /*close->NOTHING*/
+ }
OP(nxt1) = OPTIMIZED; /* was OPEN. */
OP(nxt) = OPTIMIZED; /* was CLOSE. */
+
#ifdef DEBUGGING
OP(nxt1 + 1) = OPTIMIZED; /* was count. */
OP(nxt + 1) = OPTIMIZED; /* was count. */
}
#endif
/* Optimize again: */
- study_chunk(pRExC_state, &nxt1, &deltanext, nxt,
- NULL, 0,depth+1);
+ study_chunk(pRExC_state, &nxt1, minlenp, &deltanext, nxt,
+ NULL, stopparen, recursed, NULL, 0,depth+1);
}
else
oscan->flags = 0;
if (mincount != maxcount) {
/* Cannot extend fixed substrings found inside
the group. */
- scan_commit(pRExC_state,data);
+ scan_commit(pRExC_state,data,minlenp);
if (mincount && last_str) {
SV * const sv = data->last_found;
MAGIC * const mg = SvUTF8(sv) && SvMAGICAL(sv) ?
continue;
default: /* REF and CLUMP only? */
if (flags & SCF_DO_SUBSTR) {
- scan_commit(pRExC_state,data); /* Cannot expect anything... */
+ scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
data->longest = &(data->longest_float);
}
is_inf = is_inf_internal = 1;
int value = 0;
if (flags & SCF_DO_SUBSTR) {
- scan_commit(pRExC_state,data);
+ scan_commit(pRExC_state,data,minlenp);
data->pos_min++;
}
min++;
break;
}
if (flags & SCF_DO_STCLASS_OR)
- cl_and(data->start_class, &and_with);
+ cl_and(data->start_class, and_withp);
flags &= ~SCF_DO_STCLASS;
}
}
/* 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");
+ if ( !PERL_ENABLE_POSITIVE_ASSERTION_STUDY
+ || OP(scan) == UNLESSM )
+ {
+ /* Negative Lookahead/lookbehind
+ In this case we can't do fixed string optimisation.
+ */
+
+ 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 if (minnext > U8_MAX) {
- vFAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
+ else
+ data_fake.last_closep = &fake;
+ data_fake.pos_delta = delta;
+ 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;
}
- scan->flags = (U8)minnext;
- }
- if (data) {
- if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
- pars++;
- if (data_fake.flags & SF_HAS_EVAL)
- data->flags |= SF_HAS_EVAL;
- data->whilem_c = data_fake.whilem_c;
+ if (flags & SCF_WHILEM_VISITED_POS)
+ f |= SCF_WHILEM_VISITED_POS;
+ next = regnext(scan);
+ nscan = NEXTOPER(NEXTOPER(scan));
+ minnext = study_chunk(pRExC_state, &nscan, minlenp, &deltanext,
+ last, &data_fake, stopparen, recursed, NULL, f, depth+1);
+ if (scan->flags) {
+ if (deltanext) {
+ FAIL("Variable length lookbehind not implemented");
+ }
+ else if (minnext > (I32)U8_MAX) {
+ FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
+ }
+ scan->flags = (U8)minnext;
+ }
+ if (data) {
+ if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (data_fake.flags & SF_HAS_EVAL)
+ data->flags |= SF_HAS_EVAL;
+ 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;
+ }
}
- if (f & SCF_DO_STCLASS_AND) {
- const int was = (data->start_class->flags & ANYOF_EOS);
+#if PERL_ENABLE_POSITIVE_ASSERTION_STUDY
+ else {
+ /* Positive Lookahead/lookbehind
+ In this case we can do fixed string optimisation,
+ but we must be careful about it. Note in the case of
+ lookbehind the positions will be offset by the minimum
+ length of the pattern, something we won't know about
+ until after the recurse.
+ */
+ I32 deltanext, fake = 0;
+ regnode *nscan;
+ struct regnode_charclass_class intrnl;
+ int f = 0;
+ /* We use SAVEFREEPV so that when the full compile
+ is finished perl will clean up the allocated
+ minlens when its all done. This was we don't
+ have to worry about freeing them when we know
+ they wont be used, which would be a pain.
+ */
+ I32 *minnextp;
+ Newx( minnextp, 1, I32 );
+ SAVEFREEPV(minnextp);
+
+ if (data) {
+ StructCopy(data, &data_fake, scan_data_t);
+ if ((flags & SCF_DO_SUBSTR) && data->last_found) {
+ f |= SCF_DO_SUBSTR;
+ if (scan->flags)
+ scan_commit(pRExC_state, &data_fake,minlenp);
+ data_fake.last_found=newSVsv(data->last_found);
+ }
+ }
+ else
+ data_fake.last_closep = &fake;
+ data_fake.flags = 0;
+ data_fake.pos_delta = delta;
+ if (is_inf)
+ data_fake.flags |= SF_IS_INF;
+ 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));
+
+ *minnextp = study_chunk(pRExC_state, &nscan, minnextp, &deltanext,
+ last, &data_fake, stopparen, recursed, NULL, f,depth+1);
+ if (scan->flags) {
+ if (deltanext) {
+ FAIL("Variable length lookbehind not implemented");
+ }
+ else if (*minnextp > (I32)U8_MAX) {
+ FAIL2("Lookbehind longer than %"UVuf" not implemented", (UV)U8_MAX);
+ }
+ scan->flags = (U8)*minnextp;
+ }
+
+ *minnextp += min;
+
+ 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;
+ }
+ if (data) {
+ if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (data_fake.flags & SF_HAS_EVAL)
+ data->flags |= SF_HAS_EVAL;
+ data->whilem_c = data_fake.whilem_c;
+ if ((flags & SCF_DO_SUBSTR) && data_fake.last_found) {
+ if (RExC_rx->minlen<*minnextp)
+ RExC_rx->minlen=*minnextp;
+ scan_commit(pRExC_state, &data_fake, minnextp);
+ SvREFCNT_dec(data_fake.last_found);
+
+ if ( data_fake.minlen_fixed != minlenp )
+ {
+ data->offset_fixed= data_fake.offset_fixed;
+ data->minlen_fixed= data_fake.minlen_fixed;
+ data->lookbehind_fixed+= scan->flags;
+ }
+ if ( data_fake.minlen_float != minlenp )
+ {
+ data->minlen_float= data_fake.minlen_float;
+ data->offset_float_min=data_fake.offset_float_min;
+ data->offset_float_max=data_fake.offset_float_max;
+ data->lookbehind_float+= scan->flags;
+ }
+ }
+ }
+
- cl_and(data->start_class, &intrnl);
- if (was)
- data->start_class->flags |= ANYOF_EOS;
}
+#endif
}
else if (OP(scan) == OPEN) {
- pars++;
+ if (stopparen != (I32)ARG(scan))
+ pars++;
}
else if (OP(scan) == CLOSE) {
+ if (stopparen == (I32)ARG(scan)) {
+ break;
+ }
if ((I32)ARG(scan) == is_par) {
next = regnext(scan);
if (data)
data->flags |= SF_HAS_EVAL;
}
- else if (OP(scan) == LOGICAL && scan->flags == 2) { /* Embedded follows */
+ else if ( PL_regkind[OP(scan)] == ENDLIKE ) {
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(pRExC_state,data,minlenp);
+ flags &= ~SCF_DO_SUBSTR;
+ }
+ if (data && OP(scan)==ACCEPT) {
+ data->flags |= SCF_SEEN_ACCEPT;
+ if (stopmin > min)
+ stopmin = min;
+ }
+ }
+ else if (OP(scan) == LOGICAL && scan->flags == 2) /* Embedded follows */
+ {
if (flags & SCF_DO_SUBSTR) {
- scan_commit(pRExC_state,data);
+ scan_commit(pRExC_state,data,minlenp);
data->longest = &(data->longest_float);
}
is_inf = is_inf_internal = 1;
cl_anything(pRExC_state, data->start_class);
flags &= ~SCF_DO_STCLASS;
}
+ else if (OP(scan) == GPOS) {
+ if (!(RExC_rx->extflags & RXf_GPOS_FLOAT) &&
+ !(delta || is_inf || (data && data->pos_delta)))
+ {
+ if (!(RExC_rx->extflags & RXf_ANCH) && (flags & SCF_DO_SUBSTR))
+ RExC_rx->extflags |= RXf_ANCH_GPOS;
+ if (RExC_rx->gofs < (U32)min)
+ RExC_rx->gofs = min;
+ } else {
+ RExC_rx->extflags |= RXf_GPOS_FLOAT;
+ RExC_rx->gofs = 0;
+ }
+ }
+#ifdef TRIE_STUDY_OPT
+#ifdef FULL_TRIE_STUDY
+ else if (PL_regkind[OP(scan)] == TRIE) {
+ /* NOTE - There is similar code to this block above for handling
+ BRANCH nodes on the initial study. If you change stuff here
+ check there too. */
+ regnode *trie_node= scan;
+ regnode *tail= regnext(scan);
+ reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
+ I32 max1 = 0, min1 = I32_MAX;
+ struct regnode_charclass_class accum;
+
+ if (flags & SCF_DO_SUBSTR) /* XXXX Add !SUSPEND? */
+ scan_commit(pRExC_state, data,minlenp); /* Cannot merge strings after this. */
+ if (flags & SCF_DO_STCLASS)
+ cl_init_zero(pRExC_state, &accum);
+
+ if (!trie->jump) {
+ min1= trie->minlen;
+ max1= trie->maxlen;
+ } else {
+ const regnode *nextbranch= NULL;
+ U32 word;
+
+ for ( word=1 ; word <= trie->wordcount ; word++)
+ {
+ I32 deltanext=0, minnext=0, f = 0, fake;
+ struct regnode_charclass_class this_class;
+
+ 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;
+ data_fake.pos_delta = delta;
+ 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;
+
+ if (trie->jump[word]) {
+ if (!nextbranch)
+ nextbranch = trie_node + trie->jump[0];
+ scan= trie_node + trie->jump[word];
+ /* We go from the jump point to the branch that follows
+ it. Note this means we need the vestigal unused branches
+ even though they arent otherwise used.
+ */
+ minnext = study_chunk(pRExC_state, &scan, minlenp,
+ &deltanext, (regnode *)nextbranch, &data_fake,
+ stopparen, recursed, NULL, f,depth+1);
+ }
+ if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
+ nextbranch= regnext((regnode*)nextbranch);
+
+ if (min1 > (I32)(minnext + trie->minlen))
+ min1 = minnext + trie->minlen;
+ if (max1 < (I32)(minnext + deltanext + trie->maxlen))
+ max1 = minnext + deltanext + trie->maxlen;
+ if (deltanext == I32_MAX)
+ is_inf = is_inf_internal = 1;
+
+ if (data_fake.flags & (SF_HAS_PAR|SF_IN_PAR))
+ pars++;
+ if (data_fake.flags & SCF_SEEN_ACCEPT) {
+ if ( stopmin > min + min1)
+ stopmin = min + min1;
+ flags &= ~SCF_DO_SUBSTR;
+ if (data)
+ data->flags |= SCF_SEEN_ACCEPT;
+ }
+ if (data) {
+ if (data_fake.flags & SF_HAS_EVAL)
+ data->flags |= SF_HAS_EVAL;
+ data->whilem_c = data_fake.whilem_c;
+ }
+ if (flags & SCF_DO_STCLASS)
+ cl_or(pRExC_state, &accum, &this_class);
+ }
+ }
+ 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_withp);
+ 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 */
+ INIT_AND_WITHP;
+ StructCopy(data->start_class, and_withp,
+ 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;
+ }
+ }
+ scan= tail;
+ continue;
+ }
+#else
+ else if (PL_regkind[OP(scan)] == TRIE) {
+ reg_trie_data *trie = (reg_trie_data*)RExC_rxi->data->data[ ARG(scan) ];
+ U8*bang=NULL;
+
+ min += trie->minlen;
+ delta += (trie->maxlen - trie->minlen);
+ flags &= ~SCF_DO_STCLASS; /* xxx */
+ if (flags & SCF_DO_SUBSTR) {
+ scan_commit(pRExC_state,data,minlenp); /* Cannot expect anything... */
+ data->pos_min += trie->minlen;
+ data->pos_delta += (trie->maxlen - trie->minlen);
+ if (trie->maxlen != trie->minlen)
+ data->longest = &(data->longest_float);
+ }
+ if (trie->jump) /* no more substrings -- for now /grr*/
+ flags &= ~SCF_DO_SUBSTR;
+ }
+#endif /* old or new */
+#endif /* TRIE_STUDY_OPT */
/* Else: zero-length, ignore. */
scan = regnext(scan);
}
+ if (frame) {
+ last = frame->last;
+ scan = frame->next;
+ stopparen = frame->stop;
+ frame = frame->prev;
+ goto fake_study_recurse;
+ }
finish:
+ assert(!frame);
+
*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)
+ if (is_par > (I32)U8_MAX)
is_par = 0;
if (is_par && pars==1 && data) {
data->flags |= SF_IN_PAR;
data->flags &= ~SF_IN_PAR;
}
if (flags & SCF_DO_STCLASS_OR)
- cl_and(data->start_class, &and_with);
- if (flags & SCF_EXACT_TRIE)
- data->flags |= SCF_EXACT_TRIE;
- return min;
+ cl_and(data->start_class, and_withp);
+ if (flags & SCF_TRIE_RESTUDY)
+ data->flags |= SCF_TRIE_RESTUDY;
+
+ DEBUG_STUDYDATA(data,depth);
+
+ return min < stopmin ? min : stopmin;
}
-STATIC I32
-S_add_data(RExC_state_t *pRExC_state, I32 n, const char *s)
+STATIC U32
+S_add_data(RExC_state_t *pRExC_state, U32 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 {
- Newxc(RExC_rx->data, sizeof(*RExC_rx->data) + sizeof(void*) * (n - 1),
- char, struct reg_data);
- Newx(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;
+ U32 count = RExC_rxi->data ? RExC_rxi->data->count : 0;
+
+ Renewc(RExC_rxi->data,
+ sizeof(*RExC_rxi->data) + sizeof(void*) * (count + n - 1),
+ char, struct reg_data);
+ if(count)
+ Renew(RExC_rxi->data->what, count + n, U8);
+ else
+ Newx(RExC_rxi->data->what, n, U8);
+ RExC_rxi->data->count = count + n;
+ Copy(s, RExC_rxi->data->what + count, n, U8);
+ return count;
}
#ifndef PERL_IN_XSUB_RE
#endif
+#ifdef TRIE_STUDY_OPT
+#define CHECK_RESTUDY_GOTO \
+ if ( \
+ (data.flags & SCF_TRIE_RESTUDY) \
+ && ! restudied++ \
+ ) goto reStudy
+#else
+#define CHECK_RESTUDY_GOTO
+#endif
+
/*
- pregcomp - compile a regular expression into internal code
*
* Beware that the optimization-preparation code in here knows about some
* of the structure of the compiled regexp. [I'll say.]
*/
+
+
+
+#ifndef PERL_IN_XSUB_RE
+#define RE_ENGINE_PTR &PL_core_reg_engine
+#else
+extern const struct regexp_engine my_reg_engine;
+#define RE_ENGINE_PTR &my_reg_engine
+#endif
+/* these make a few things look better, to avoid indentation */
+#define BEGIN_BLOCK {
+#define END_BLOCK }
+
regexp *
Perl_pregcomp(pTHX_ char *exp, char *xend, PMOP *pm)
{
dVAR;
+ GET_RE_DEBUG_FLAGS_DECL;
+ DEBUG_r(if (!PL_colorset) reginitcolors());
+#ifndef PERL_IN_XSUB_RE
+ BEGIN_BLOCK
+ /* Dispatch a request to compile a regexp to correct
+ regexp engine. */
+ HV * const table = GvHV(PL_hintgv);
+ if (table) {
+ SV **ptr= hv_fetchs(table, "regcomp", FALSE);
+ if (ptr && SvIOK(*ptr) && SvIV(*ptr)) {
+ const regexp_engine *eng=INT2PTR(regexp_engine*,SvIV(*ptr));
+ DEBUG_COMPILE_r({
+ PerlIO_printf(Perl_debug_log, "Using engine %"UVxf"\n",
+ SvIV(*ptr));
+ });
+ return CALLREGCOMP_ENG(eng, exp, xend, pm);
+ }
+ }
+ END_BLOCK
+#endif
+ BEGIN_BLOCK
register regexp *r;
+ register regexp_internal *ri;
regnode *scan;
regnode *first;
I32 flags;
int restudied= 0;
RExC_state_t copyRExC_state;
#endif
-
- 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]);
+ SV *dsv= sv_newmortal();
+ RE_PV_QUOTED_DECL(s, RExC_utf8,
+ dsv, RExC_precomp, (xend - exp), 60);
+ PerlIO_printf(Perl_debug_log, "%sCompiling REx%s %s\n",
+ PL_colors[4],PL_colors[5],s);
});
RExC_flags = pm->op_pmflags;
RExC_sawback = 0;
RExC_end = xend;
RExC_naughty = 0;
RExC_npar = 1;
+ RExC_cpar = 1;
+ RExC_nestroot = 0;
RExC_size = 0L;
RExC_emit = &PL_regdummy;
RExC_whilem_seen = 0;
+ RExC_charnames = NULL;
+ RExC_open_parens = NULL;
+ RExC_close_parens = NULL;
+ RExC_opend = NULL;
+ RExC_paren_names = NULL;
+ RExC_recurse = NULL;
+ RExC_recurse_count = 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);
RExC_precomp = NULL;
return(NULL);
}
- DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "Required "));
- DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "size %"IVdf" nodes ", (IV)RExC_size));
- DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nStarting second pass (creation)\n"));
DEBUG_PARSE_r({
+ PerlIO_printf(Perl_debug_log,
+ "Required size %"IVdf" nodes\n"
+ "Starting second pass (creation)\n",
+ (IV)RExC_size);
RExC_lastnum=0;
RExC_lastparse=NULL;
});
-
-
/* Small enough for pointer-storage convention?
If extralen==0, this means that we will not need long jumps. */
if (RExC_size >= 0x10000L && RExC_extralen)
if (RExC_whilem_seen > 15)
RExC_whilem_seen = 15;
- /* Allocate space and initialize. */
- Newxc(r, sizeof(regexp) + (unsigned)RExC_size * sizeof(regnode),
- char, regexp);
- if (r == NULL)
- FAIL("Regexp out of space");
+#ifdef DEBUGGING
+ /* Make room for a sentinel value at the end of the program */
+ RExC_size++;
+#endif
+ /* Allocate space and zero-initialize. Note, the two step process
+ of zeroing when in debug mode, thus anything assigned has to
+ happen after that */
+ Newxz(r, 1, regexp);
+ Newxc(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode),
+ char, regexp_internal);
+ if ( r == NULL || ri == 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);
+ Zero(ri, sizeof(regexp_internal) + (unsigned)RExC_size * sizeof(regnode), char);
+#else
+ /* bulk initialize base fields with 0. */
+ Zero(ri, sizeof(regexp_internal), char);
#endif
+
+ /* non-zero initialization begins here */
+ RXi_SET( r, ri );
+ r->engine= RE_ENGINE_PTR;
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 = NULL;
-#endif
- r->reganch = pm->op_pmflags & PMf_COMPILETIME;
+ r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME;
+ r->intflags = 0;
r->nparens = RExC_npar - 1; /* set early to validate backrefs */
- r->lastparen = 0; /* mg.c reads this. */
-
- r->substrs = 0; /* Useful during FAIL. */
- r->startp = 0; /* Useful during FAIL. */
- r->endp = 0; /* Useful during FAIL. */
+
+ if (RExC_seen & REG_SEEN_RECURSE) {
+ Newxz(RExC_open_parens, RExC_npar,regnode *);
+ SAVEFREEPV(RExC_open_parens);
+ Newxz(RExC_close_parens,RExC_npar,regnode *);
+ SAVEFREEPV(RExC_close_parens);
+ }
- Newxz(r->offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
- if (r->offsets) {
- r->offsets[0] = RExC_size;
+ /* Useful during FAIL. */
+ Newxz(ri->offsets, 2*RExC_size+1, U32); /* MJD 20001228 */
+ if (ri->offsets) {
+ ri->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",
+ ri->offsets ? "Got" : "Couldn't get",
(UV)((2*RExC_size+1) * sizeof(U32))));
RExC_rx = r;
+ RExC_rxi = ri;
/* Second pass: emit code. */
RExC_flags = pm->op_pmflags; /* don't let top level (?i) bleed */
RExC_end = xend;
RExC_naughty = 0;
RExC_npar = 1;
- RExC_emit_start = r->program;
- RExC_emit = r->program;
+ RExC_cpar = 1;
+ RExC_emit_start = ri->program;
+ RExC_emit = ri->program;
+#ifdef DEBUGGING
+ /* put a sentinal on the end of the program so we can check for
+ overwrites */
+ ri->program[RExC_size].type = 255;
+#endif
/* Store the count of eval-groups for security checks: */
- RExC_emit->next_off = (U16)((RExC_seen_evals > U16_MAX) ? U16_MAX : RExC_seen_evals);
+ RExC_emit->next_off = (RExC_seen_evals > (I32)U16_MAX) ? U16_MAX : (U16)RExC_seen_evals;
REGC((U8)REG_MAGIC, (char*) RExC_emit++);
- r->data = 0;
if (reg(pRExC_state, 0, &flags,1) == NULL)
return(NULL);
+
/* XXXX To minimize changes to RE engine we always allocate
3-units-long substrs field. */
Newx(r->substrs, 1, struct reg_substr_data);
+ if (RExC_recurse_count) {
+ Newxz(RExC_recurse,RExC_recurse_count,regnode *);
+ SAVEFREEPV(RExC_recurse);
+ }
reStudy:
+ r->minlen = minlen = sawplus = sawopen = 0;
Zero(r->substrs, 1, struct reg_substr_data);
- StructCopy(&zero_scan_data, &data, scan_data_t);
#ifdef TRIE_STUDY_OPT
if ( restudied ) {
+ U32 seen=RExC_seen;
DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log,"Restudying\n"));
- RExC_state=copyRExC_state;
- if (data.longest_fixed)
+
+ RExC_state = copyRExC_state;
+ if (seen & REG_TOP_LEVEL_BRANCHES)
+ RExC_seen |= REG_TOP_LEVEL_BRANCHES;
+ else
+ RExC_seen &= ~REG_TOP_LEVEL_BRANCHES;
+ if (data.last_found) {
SvREFCNT_dec(data.longest_fixed);
- if (data.longest_float)
SvREFCNT_dec(data.longest_float);
- if (data.last_found)
SvREFCNT_dec(data.last_found);
+ }
+ StructCopy(&zero_scan_data, &data, scan_data_t);
} else {
- copyRExC_state=RExC_state;
+ StructCopy(&zero_scan_data, &data, scan_data_t);
+ copyRExC_state = RExC_state;
}
+#else
+ StructCopy(&zero_scan_data, &data, scan_data_t);
#endif
+
/* Dig out information for optimizations. */
- r->reganch = pm->op_pmflags & PMf_COMPILETIME; /* Again? */
+ r->extflags = pm->op_pmflags & RXf_PMf_COMPILETIME; /* Again? */
pm->op_pmflags = RExC_flags;
if (UTF)
- r->reganch |= ROPT_UTF8; /* Unicode in it? */
- r->regstclass = NULL;
+ r->extflags |= RXf_UTF8; /* Unicode in it? */
+ ri->regstclass = NULL;
if (RExC_naughty >= 10) /* Probably an expensive pattern. */
- r->reganch |= ROPT_NAUGHTY;
- scan = r->program + 1; /* First BRANCH. */
+ r->intflags |= PREGf_NAUGHTY;
+ scan = ri->program + 1; /* First BRANCH. */
- /* XXXX Should not we check for something else? Usually it is OPEN1... */
- if (OP(scan) != BRANCH) { /* Only one top-level choice. */
+ /* testing for BRANCH here tells us whether there is "must appear"
+ data in the pattern. If there is then we can use it for optimisations */
+ if (!(RExC_seen & REG_TOP_LEVEL_BRANCHES)) { /* Only one top-level choice. */
I32 fake;
STRLEN longest_float_length, longest_fixed_length;
struct regnode_charclass_class ch_class; /* pointed to by data */
/* An {n,m} with n>0 */
(PL_regkind[OP(first)] == CURLY && ARG1(first) > 0) )
{
- DEBUG_PEEP("first:",first,0);
+
if (OP(first) == PLUS)
sawplus = 1;
else
if (OP(first) == IFMATCH) {
first = NEXTOPER(first);
first += EXTRA_STEP_2ARGS;
- } else /*xxx possible optimisation for /(?=)/*/
+ } else /* XXX possible optimisation for /(?=)/ */
first = NEXTOPER(first);
}
/* Starting-point info. */
again:
+ DEBUG_PEEP("first:",first,0);
/* Ignore EXACT as we deal with it later. */
if (PL_regkind[OP(first)] == EXACT) {
if (OP(first) == EXACT)
NOOP; /* Empty, get anchored substr later. */
else if ((OP(first) == EXACTF || OP(first) == EXACTFL))
- r->regstclass = first;
+ ri->regstclass = first;
}
#ifdef TRIE_STCLASS
- else if (OP(first) == TRIE &&
- ((reg_trie_data *)r->data->data[ ARG(first) ])->minlen>0)
+ else if (PL_regkind[OP(first)] == TRIE &&
+ ((reg_trie_data *)ri->data->data[ ARG(first) ])->minlen>0)
{
+ regnode *trie_op;
/* this can happen only on restudy */
- struct regnode_1 *trie_op;
- Newxz(trie_op,1,struct regnode_1);
- StructCopy(first,trie_op,struct regnode_1);
- make_trie_failtable(pRExC_state, (regnode *)first, (regnode *)trie_op, 0);
- r->regstclass = (regnode *)trie_op;
+ if ( OP(first) == TRIE ) {
+ struct regnode_1 *trieop =
+ PerlMemShared_calloc(1, sizeof(struct regnode_1));
+ StructCopy(first,trieop,struct regnode_1);
+ trie_op=(regnode *)trieop;
+ } else {
+ struct regnode_charclass *trieop =
+ PerlMemShared_calloc(1, sizeof(struct regnode_charclass));
+ StructCopy(first,trieop,struct regnode_charclass);
+ trie_op=(regnode *)trieop;
+ }
+ OP(trie_op)+=2;
+ make_trie_failtable(pRExC_state, (regnode *)first, trie_op, 0);
+ ri->regstclass = trie_op;
}
#endif
else if (strchr((const char*)PL_simple,OP(first)))
- r->regstclass = first;
+ ri->regstclass = first;
else if (PL_regkind[OP(first)] == BOUND ||
PL_regkind[OP(first)] == NBOUND)
- r->regstclass = first;
+ ri->regstclass = first;
else if (PL_regkind[OP(first)] == BOL) {
- r->reganch |= (OP(first) == MBOL
- ? ROPT_ANCH_MBOL
+ r->extflags |= (OP(first) == MBOL
+ ? RXf_ANCH_MBOL
: (OP(first) == SBOL
- ? ROPT_ANCH_SBOL
- : ROPT_ANCH_BOL));
+ ? RXf_ANCH_SBOL
+ : RXf_ANCH_BOL));
first = NEXTOPER(first);
goto again;
}
else if (OP(first) == GPOS) {
- r->reganch |= ROPT_ANCH_GPOS;
+ r->extflags |= RXf_ANCH_GPOS;
first = NEXTOPER(first);
goto again;
}
- else if (!sawopen && (OP(first) == STAR &&
+ else if ((!sawopen || !RExC_sawback) &&
+ (OP(first) == STAR &&
PL_regkind[OP(NEXTOPER(first))] == REG_ANY) &&
- !(r->reganch & ROPT_ANCH) )
+ !(r->extflags & RXf_ANCH) && !(RExC_seen & REG_SEEN_EVAL))
{
/* 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;
+ ? RXf_ANCH_MBOL
+ : RXf_ANCH_SBOL;
+ r->extflags |= type;
+ r->intflags |= PREGf_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;
+ r->intflags |= PREGf_SKIP;
/* Scan is after the zeroth branch, first is atomic matcher. */
#ifdef TRIE_STUDY_OPT
- DEBUG_COMPILE_r(
+ DEBUG_PARSE_r(
if (!restudied)
PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
(IV)(first - scan + 1))
);
#else
- DEBUG_COMPILE_r(
+ DEBUG_PARSE_r(
PerlIO_printf(Perl_debug_log, "first at %"IVdf"\n",
(IV)(first - scan + 1))
);
* 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 = newSVpvs("");
data.longest_float = newSVpvs("");
data.last_found = newSVpvs("");
data.longest = &(data.longest_fixed);
first = scan;
- if (!r->regstclass) {
+ if (!ri->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, &minlen, &fake, scan + RExC_size, /* Up to end */
+ &data, -1, NULL, NULL,
+ SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
- minlen = study_chunk(pRExC_state, &first, &fake, scan + RExC_size, /* Up to end */
- &data, SCF_DO_SUBSTR | SCF_WHILEM_VISITED_POS | stclass_flag,0);
-
-#ifdef TRIE_STUDY_OPT
- if ( (data.flags & SCF_EXACT_TRIE) && ! restudied++ ) {
- goto reStudy;
- }
-#endif
+ CHECK_RESTUDY_GOTO;
+
+
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);
+ && (!(RExC_seen & REG_SEEN_GPOS) || (r->extflags & RXf_ANCH_GPOS)))
+ r->extflags |= RXf_CHECK_ALL;
+ scan_commit(pRExC_state, &data,&minlen);
SvREFCNT_dec(data.last_found);
+ /* Note that code very similar to this but for anchored string
+ follows immediately below, changes may need to be made to both.
+ Be careful.
+ */
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;
+ || (RExC_flags & RXf_PMf_MULTILINE))))
+ {
+ I32 t,ml;
- if (SvCUR(data.longest_fixed) /* ok to leave SvCUR */
+ 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)+. */
+ /* copy the information about the longest float from the reg_scan_data
+ over to the program. */
if (SvUTF8(data.longest_float)) {
r->float_utf8 = data.longest_float;
r->float_substr = NULL;
r->float_substr = data.longest_float;
r->float_utf8 = NULL;
}
- r->float_min_offset = data.offset_float_min;
+ /* float_end_shift is how many chars that must be matched that
+ follow this item. We calculate it ahead of time as once the
+ lookbehind offset is added in we lose the ability to correctly
+ calculate it.*/
+ ml = data.minlen_float ? *(data.minlen_float)
+ : (I32)longest_float_length;
+ r->float_end_shift = ml - data.offset_float_min
+ - longest_float_length + (SvTAIL(data.longest_float) != 0)
+ + data.lookbehind_float;
+ r->float_min_offset = data.offset_float_min - data.lookbehind_float;
r->float_max_offset = data.offset_float_max;
+ if (data.offset_float_max < I32_MAX) /* Don't offset infinity */
+ r->float_max_offset -= data.lookbehind_float;
+
t = (data.flags & SF_FL_BEFORE_EOL /* Can't have SEOL and MULTI */
&& (!(data.flags & SF_FL_BEFORE_MEOL)
- || (RExC_flags & PMf_MULTILINE)));
+ || (RExC_flags & RXf_PMf_MULTILINE)));
fbm_compile(data.longest_float, t ? FBMcf_TAIL : 0);
}
else {
longest_float_length = 0;
}
+ /* Note that code very similar to this but for floating string
+ is immediately above, changes may need to be made to both.
+ Be careful.
+ */
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;
+ || (RExC_flags & RXf_PMf_MULTILINE))))
+ {
+ I32 t,ml;
+ /* copy the information about the longest fixed
+ from the reg_scan_data over to the program. */
if (SvUTF8(data.longest_fixed)) {
r->anchored_utf8 = data.longest_fixed;
r->anchored_substr = NULL;
r->anchored_substr = data.longest_fixed;
r->anchored_utf8 = NULL;
}
- r->anchored_offset = data.offset_fixed;
+ /* fixed_end_shift is how many chars that must be matched that
+ follow this item. We calculate it ahead of time as once the
+ lookbehind offset is added in we lose the ability to correctly
+ calculate it.*/
+ ml = data.minlen_fixed ? *(data.minlen_fixed)
+ : (I32)longest_fixed_length;
+ r->anchored_end_shift = ml - data.offset_fixed
+ - longest_fixed_length + (SvTAIL(data.longest_fixed) != 0)
+ + data.lookbehind_fixed;
+ r->anchored_offset = data.offset_fixed - data.lookbehind_fixed;
+
t = (data.flags & SF_FIX_BEFORE_EOL /* Can't have SEOL and MULTI */
&& (!(data.flags & SF_FIX_BEFORE_MEOL)
- || (RExC_flags & PMf_MULTILINE)));
+ || (RExC_flags & RXf_PMf_MULTILINE)));
fbm_compile(data.longest_fixed, t ? FBMcf_TAIL : 0);
}
else {
SvREFCNT_dec(data.longest_fixed);
longest_fixed_length = 0;
}
- if (r->regstclass
- && (OP(r->regstclass) == REG_ANY || OP(r->regstclass) == SANY))
- r->regstclass = NULL;
+ if (ri->regstclass
+ && (OP(ri->regstclass) == REG_ANY || OP(ri->regstclass) == SANY))
+ ri->regstclass = NULL;
if ((!(r->anchored_substr || r->anchored_utf8) || r->anchored_offset)
&& stclass_flag
&& !(data.start_class->flags & ANYOF_EOS)
&& !cl_is_anything(data.start_class))
{
- const I32 n = add_data(pRExC_state, 1, "f");
+ const U32 n = add_data(pRExC_state, 1, "f");
- Newx(RExC_rx->data->data[n], 1,
+ Newx(RExC_rxi->data->data[n], 1,
struct regnode_charclass_class);
StructCopy(data.start_class,
- (struct regnode_charclass_class*)RExC_rx->data->data[n],
+ (struct regnode_charclass_class*)RExC_rxi->data->data[n],
struct regnode_charclass_class);
- r->regstclass = (regnode*)RExC_rx->data->data[n];
- r->reganch &= ~ROPT_SKIP; /* Used in find_byclass(). */
+ ri->regstclass = (regnode*)RExC_rxi->data->data[n];
+ r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
DEBUG_COMPILE_r({ SV *sv = sv_newmortal();
regprop(r, sv, (regnode*)data.start_class);
PerlIO_printf(Perl_debug_log,
/* A temporary algorithm prefers floated substr to fixed one to dig more info. */
if (longest_fixed_length > longest_float_length) {
+ r->check_end_shift = r->anchored_end_shift;
r->check_substr = r->anchored_substr;
r->check_utf8 = r->anchored_utf8;
r->check_offset_min = r->check_offset_max = r->anchored_offset;
- if (r->reganch & ROPT_ANCH_SINGLE)
- r->reganch |= ROPT_NOSCAN;
+ if (r->extflags & RXf_ANCH_SINGLE)
+ r->extflags |= RXf_NOSCAN;
}
else {
+ r->check_end_shift = r->float_end_shift;
r->check_substr = r->float_substr;
r->check_utf8 = r->float_utf8;
- r->check_offset_min = data.offset_float_min;
- r->check_offset_max = data.offset_float_max;
+ r->check_offset_min = r->float_min_offset;
+ r->check_offset_max = r->float_max_offset;
}
/* XXXX Currently intuiting is not compatible with ANCH_GPOS.
This should be changed ASAP! */
- if ((r->check_substr || r->check_utf8) && !(r->reganch & ROPT_ANCH_GPOS)) {
- r->reganch |= RE_USE_INTUIT;
+ if ((r->check_substr || r->check_utf8) && !(r->extflags & RXf_ANCH_GPOS)) {
+ r->extflags |= RXf_USE_INTUIT;
if (SvTAIL(r->check_substr ? r->check_substr : r->check_utf8))
- r->reganch |= RE_INTUIT_TAIL;
+ r->extflags |= RXf_INTUIT_TAIL;
}
+ /* XXX Unneeded? dmq (shouldn't as this is handled elsewhere)
+ if ( (STRLEN)minlen < longest_float_length )
+ minlen= longest_float_length;
+ if ( (STRLEN)minlen < longest_fixed_length )
+ minlen= longest_fixed_length;
+ */
}
else {
/* Several toplevels. Best we can is to set minlen. */
struct regnode_charclass_class ch_class;
I32 last_close = 0;
- DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "\n"));
+ DEBUG_PARSE_r(PerlIO_printf(Perl_debug_log, "\nMulti Top Level\n"));
- scan = r->program + 1;
+ scan = ri->program + 1;
cl_init(pRExC_state, &ch_class);
data.start_class = &ch_class;
data.last_closep = &last_close;
- minlen = study_chunk(pRExC_state, &scan, &fake, scan + RExC_size,
- &data, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
-
-#ifdef TRIE_STUDY_OPT
- if ( (data.flags & SCF_EXACT_TRIE) && ! restudied++ ) {
- goto reStudy;
- }
-#endif
+
+ minlen = study_chunk(pRExC_state, &scan, &minlen, &fake, scan + RExC_size,
+ &data, -1, NULL, NULL, SCF_DO_STCLASS_AND|SCF_WHILEM_VISITED_POS,0);
+
+ CHECK_RESTUDY_GOTO;
r->check_substr = r->check_utf8 = r->anchored_substr = r->anchored_utf8
= r->float_substr = r->float_utf8 = NULL;
if (!(data.start_class->flags & ANYOF_EOS)
&& !cl_is_anything(data.start_class))
{
- const I32 n = add_data(pRExC_state, 1, "f");
+ const U32 n = add_data(pRExC_state, 1, "f");
- Newx(RExC_rx->data->data[n], 1,
+ Newx(RExC_rxi->data->data[n], 1,
struct regnode_charclass_class);
StructCopy(data.start_class,
- (struct regnode_charclass_class*)RExC_rx->data->data[n],
+ (struct regnode_charclass_class*)RExC_rxi->data->data[n],
struct regnode_charclass_class);
- r->regstclass = (regnode*)RExC_rx->data->data[n];
- r->reganch &= ~ROPT_SKIP; /* Used in find_byclass(). */
+ ri->regstclass = (regnode*)RExC_rxi->data->data[n];
+ r->intflags &= ~PREGf_SKIP; /* Used in find_byclass(). */
DEBUG_COMPILE_r({ SV* sv = sv_newmortal();
regprop(r, sv, (regnode*)data.start_class);
PerlIO_printf(Perl_debug_log,
}
}
- r->minlen = minlen;
+ /* Guard against an embedded (?=) or (?<=) with a longer minlen than
+ the "real" pattern. */
+ DEBUG_OPTIMISE_r({
+ PerlIO_printf(Perl_debug_log,"minlen: %"IVdf" r->minlen:%"IVdf"\n",
+ (IV)minlen, (IV)r->minlen);
+ });
+ r->minlenret = minlen;
+ if (r->minlen < minlen)
+ r->minlen = minlen;
+
if (RExC_seen & REG_SEEN_GPOS)
- r->reganch |= ROPT_GPOS_SEEN;
+ r->extflags |= RXf_GPOS_SEEN;
if (RExC_seen & REG_SEEN_LOOKBEHIND)
- r->reganch |= ROPT_LOOKBEHIND_SEEN;
+ r->extflags |= RXf_LOOKBEHIND_SEEN;
if (RExC_seen & REG_SEEN_EVAL)
- r->reganch |= ROPT_EVAL_SEEN;
+ r->extflags |= RXf_EVAL_SEEN;
if (RExC_seen & REG_SEEN_CANY)
- r->reganch |= ROPT_CANY_SEEN;
+ r->extflags |= RXf_CANY_SEEN;
+ if (RExC_seen & REG_SEEN_VERBARG)
+ r->intflags |= PREGf_VERBARG_SEEN;
+ if (RExC_seen & REG_SEEN_CUTGROUP)
+ r->intflags |= PREGf_CUTGROUP_SEEN;
+ if (RExC_paren_names)
+ r->paren_names = (HV*)SvREFCNT_inc(RExC_paren_names);
+ else
+ r->paren_names = NULL;
+
+ if (RExC_recurse_count) {
+ for ( ; RExC_recurse_count ; RExC_recurse_count-- ) {
+ const regnode *scan = RExC_recurse[RExC_recurse_count-1];
+ ARG2L_SET( scan, RExC_open_parens[ARG(scan)-1] - scan );
+ }
+ }
Newxz(r->startp, RExC_npar, I32);
Newxz(r->endp, RExC_npar, I32);
+ /* assume we don't need to swap parens around before we match */
- DEBUG_r( RX_DEBUG_on(r) );
DEBUG_DUMP_r({
PerlIO_printf(Perl_debug_log,"Final program:\n");
regdump(r);
});
+ DEBUG_OFFSETS_r(if (ri->offsets) {
+ const U32 len = ri->offsets[0];
+ U32 i;
+ GET_RE_DEBUG_FLAGS_DECL;
+ PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)ri->offsets[0]);
+ for (i = 1; i <= len; i++) {
+ if (ri->offsets[i*2-1] || ri->offsets[i*2])
+ PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
+ (UV)i, (UV)ri->offsets[i*2-1], (UV)ri->offsets[i*2]);
+ }
+ PerlIO_printf(Perl_debug_log, "\n");
+ });
return(r);
+ END_BLOCK
+}
+
+#undef CORE_ONLY_BLOCK
+#undef END_BLOCK
+#undef RE_ENGINE_PTR
+
+#ifndef PERL_IN_XSUB_RE
+SV*
+Perl_reg_named_buff_sv(pTHX_ SV* namesv)
+{
+ I32 parno = 0; /* no match */
+ if (PL_curpm) {
+ const REGEXP * const rx = PM_GETRE(PL_curpm);
+ if (rx && rx->paren_names) {
+ HE *he_str = hv_fetch_ent( rx->paren_names, namesv, 0, 0 );
+ if (he_str) {
+ IV i;
+ SV* sv_dat=HeVAL(he_str);
+ I32 *nums=(I32*)SvPVX(sv_dat);
+ for ( i=0; i<SvIVX(sv_dat); i++ ) {
+ if ((I32)(rx->lastparen) >= nums[i] &&
+ rx->endp[nums[i]] != -1)
+ {
+ parno = nums[i];
+ break;
+ }
+ }
+ }
+ }
+ }
+ if ( !parno ) {
+ return 0;
+ } else {
+ GV *gv_paren;
+ SV *sv= sv_newmortal();
+ Perl_sv_setpvf(aTHX_ sv, "%"IVdf,(IV)parno);
+ gv_paren= Perl_gv_fetchsv(aTHX_ sv, GV_ADD, SVt_PVGV);
+ return GvSVn(gv_paren);
+ }
}
+#endif
+
+/* Scans the name of a named buffer from the pattern.
+ * If flags is REG_RSN_RETURN_NULL returns null.
+ * If flags is REG_RSN_RETURN_NAME returns an SV* containing the name
+ * If flags is REG_RSN_RETURN_DATA returns the data SV* corresponding
+ * to the parsed name as looked up in the RExC_paren_names hash.
+ * If there is an error throws a vFAIL().. type exception.
+ */
+#define REG_RSN_RETURN_NULL 0
+#define REG_RSN_RETURN_NAME 1
+#define REG_RSN_RETURN_DATA 2
+
+STATIC SV*
+S_reg_scan_name(pTHX_ RExC_state_t *pRExC_state, U32 flags) {
+ char *name_start = RExC_parse;
+ if ( UTF ) {
+ STRLEN numlen;
+ while( isIDFIRST_uni(utf8n_to_uvchr((U8*)RExC_parse,
+ RExC_end - RExC_parse, &numlen, UTF8_ALLOW_DEFAULT)))
+ {
+ RExC_parse += numlen;
+ }
+ } else {
+ while( isIDFIRST(*RExC_parse) )
+ RExC_parse++;
+ }
+ if ( flags ) {
+ SV* sv_name = sv_2mortal(Perl_newSVpvn(aTHX_ name_start,
+ (int)(RExC_parse - name_start)));
+ if (UTF)
+ SvUTF8_on(sv_name);
+ if ( flags == REG_RSN_RETURN_NAME)
+ return sv_name;
+ else if (flags==REG_RSN_RETURN_DATA) {
+ HE *he_str = NULL;
+ SV *sv_dat = NULL;
+ if ( ! sv_name ) /* should not happen*/
+ Perl_croak(aTHX_ "panic: no svname in reg_scan_name");
+ if (RExC_paren_names)
+ he_str = hv_fetch_ent( RExC_paren_names, sv_name, 0, 0 );
+ if ( he_str )
+ sv_dat = HeVAL(he_str);
+ if ( ! sv_dat )
+ vFAIL("Reference to nonexistent named group");
+ return sv_dat;
+ }
+ else {
+ Perl_croak(aTHX_ "panic: bad flag in reg_scan_name");
+ }
+ /* NOT REACHED */
+ }
+ return NULL;
+}
#define DEBUG_PARSE_MSG(funcname) DEBUG_PARSE_r({ \
int rem=(int)(RExC_end - RExC_parse); \
else \
num=REG_NODE_NUM(RExC_emit); \
if (RExC_lastnum!=num) \
- PerlIO_printf(Perl_debug_log,"|%4d",num); \
+ PerlIO_printf(Perl_debug_log,"|%4d",num); \
else \
- PerlIO_printf(Perl_debug_log,"|%4s",""); \
+ PerlIO_printf(Perl_debug_log,"|%4s",""); \
PerlIO_printf(Perl_debug_log,"|%*s%-4s", \
(int)((depth*2)), "", \
(funcname) \
DEBUG_PARSE_MSG((funcname)); \
PerlIO_printf(Perl_debug_log,"%4s","\n"); \
})
+#define DEBUG_PARSE_FMT(funcname,fmt,args) DEBUG_PARSE_r({ \
+ DEBUG_PARSE_MSG((funcname)); \
+ PerlIO_printf(Perl_debug_log,fmt "\n",args); \
+})
/*
- reg - regular expression, i.e. main body or parenthesized thing
*
#define REGTAIL_STUDY(x,y,z) regtail((x),(y),(z),depth+1)
#endif
+/* this idea is borrowed from STR_WITH_LEN in handy.h */
+#define CHECK_WORD(s,v,l) \
+ (((sizeof(s)-1)==(l)) && (strnEQ(start_verb, (s ""), (sizeof(s)-1))))
+
STATIC regnode *
S_reg(pTHX_ RExC_state_t *pRExC_state, I32 paren, I32 *flagp,U32 depth)
/* paren: Parenthesized? 0=top, 1=(, inside: changed to letter. */
/* Make an OPEN node, if parenthesized. */
if (paren) {
+ if ( *RExC_parse == '*') { /* (*VERB:ARG) */
+ char *start_verb = RExC_parse;
+ STRLEN verb_len = 0;
+ char *start_arg = NULL;
+ unsigned char op = 0;
+ int argok = 1;
+ int internal_argval = 0; /* internal_argval is only useful if !argok */
+ while ( *RExC_parse && *RExC_parse != ')' ) {
+ if ( *RExC_parse == ':' ) {
+ start_arg = RExC_parse + 1;
+ break;
+ }
+ RExC_parse++;
+ }
+ ++start_verb;
+ verb_len = RExC_parse - start_verb;
+ if ( start_arg ) {
+ RExC_parse++;
+ while ( *RExC_parse && *RExC_parse != ')' )
+ RExC_parse++;
+ if ( *RExC_parse != ')' )
+ vFAIL("Unterminated verb pattern argument");
+ if ( RExC_parse == start_arg )
+ start_arg = NULL;
+ } else {
+ if ( *RExC_parse != ')' )
+ vFAIL("Unterminated verb pattern");
+ }
+
+ switch ( *start_verb ) {
+ case 'A': /* (*ACCEPT) */
+ if ( CHECK_WORD("ACCEPT",start_verb,verb_len) ) {
+ op = ACCEPT;
+ internal_argval = RExC_nestroot;
+ }
+ break;
+ case 'C': /* (*COMMIT) */
+ if ( CHECK_WORD("COMMIT",start_verb,verb_len) )
+ op = COMMIT;
+ break;
+ case 'F': /* (*FAIL) */
+ if ( verb_len==1 || CHECK_WORD("FAIL",start_verb,verb_len) ) {
+ op = OPFAIL;
+ argok = 0;
+ }
+ break;
+ case ':': /* (*:NAME) */
+ case 'M': /* (*MARK:NAME) */
+ if ( verb_len==0 || CHECK_WORD("MARK",start_verb,verb_len) ) {
+ op = MARKPOINT;
+ argok = -1;
+ }
+ break;
+ case 'P': /* (*PRUNE) */
+ if ( CHECK_WORD("PRUNE",start_verb,verb_len) )
+ op = PRUNE;
+ break;
+ case 'S': /* (*SKIP) */
+ if ( CHECK_WORD("SKIP",start_verb,verb_len) )
+ op = SKIP;
+ break;
+ case 'T': /* (*THEN) */
+ /* [19:06] <TimToady> :: is then */
+ if ( CHECK_WORD("THEN",start_verb,verb_len) ) {
+ op = CUTGROUP;
+ RExC_seen |= REG_SEEN_CUTGROUP;
+ }
+ break;
+ }
+ if ( ! op ) {
+ RExC_parse++;
+ vFAIL3("Unknown verb pattern '%.*s'",
+ verb_len, start_verb);
+ }
+ if ( argok ) {
+ if ( start_arg && internal_argval ) {
+ vFAIL3("Verb pattern '%.*s' may not have an argument",
+ verb_len, start_verb);
+ } else if ( argok < 0 && !start_arg ) {
+ vFAIL3("Verb pattern '%.*s' has a mandatory argument",
+ verb_len, start_verb);
+ } else {
+ ret = reganode(pRExC_state, op, internal_argval);
+ if ( ! internal_argval && ! SIZE_ONLY ) {
+ if (start_arg) {
+ SV *sv = newSVpvn( start_arg, RExC_parse - start_arg);
+ ARG(ret) = add_data( pRExC_state, 1, "S" );
+ RExC_rxi->data->data[ARG(ret)]=(void*)sv;
+ ret->flags = 0;
+ } else {
+ ret->flags = 1;
+ }
+ }
+ }
+ if (!internal_argval)
+ RExC_seen |= REG_SEEN_VERBARG;
+ } else if ( start_arg ) {
+ vFAIL3("Verb pattern '%.*s' may not have an argument",
+ verb_len, start_verb);
+ } else {
+ ret = reg_node(pRExC_state, op);
+ }
+ nextchar(pRExC_state);
+ return ret;
+ } else
if (*RExC_parse == '?') { /* (?...) */
U32 posflags = 0, negflags = 0;
U32 *flagsp = &posflags;
paren = *RExC_parse++;
ret = NULL; /* For look-ahead/behind. */
switch (paren) {
+
case '<': /* (?<...) */
- RExC_seen |= REG_SEEN_LOOKBEHIND;
if (*RExC_parse == '!')
paren = ',';
- if (*RExC_parse != '=' && *RExC_parse != '!')
- goto unknown;
+ else if (*RExC_parse != '=')
+ { /* (?<...>) */
+ char *name_start;
+ SV *svname;
+ paren= '>';
+ case '\'': /* (?'...') */
+ name_start= RExC_parse;
+ svname = reg_scan_name(pRExC_state,
+ SIZE_ONLY ? /* reverse test from the others */
+ REG_RSN_RETURN_NAME :
+ REG_RSN_RETURN_NULL);
+ if (RExC_parse == name_start)
+ goto unknown;
+ if (*RExC_parse != paren)
+ vFAIL2("Sequence (?%c... not terminated",
+ paren=='>' ? '<' : paren);
+ if (SIZE_ONLY) {
+ HE *he_str;
+ SV *sv_dat = NULL;
+ if (!svname) /* shouldnt happen */
+ Perl_croak(aTHX_
+ "panic: reg_scan_name returned NULL");
+ if (!RExC_paren_names) {
+ RExC_paren_names= newHV();
+ sv_2mortal((SV*)RExC_paren_names);
+ }
+ he_str = hv_fetch_ent( RExC_paren_names, svname, 1, 0 );
+ if ( he_str )
+ sv_dat = HeVAL(he_str);
+ if ( ! sv_dat ) {
+ /* croak baby croak */
+ Perl_croak(aTHX_
+ "panic: paren_name hash element allocation failed");
+ } else if ( SvPOK(sv_dat) ) {
+ IV count=SvIV(sv_dat);
+ I32 *pv=(I32*)SvGROW(sv_dat,SvCUR(sv_dat)+sizeof(I32)+1);
+ SvCUR_set(sv_dat,SvCUR(sv_dat)+sizeof(I32));
+ pv[count]=RExC_npar;
+ SvIVX(sv_dat)++;
+ } else {
+ (void)SvUPGRADE(sv_dat,SVt_PVNV);
+ sv_setpvn(sv_dat, (char *)&(RExC_npar), sizeof(I32));
+ SvIOK_on(sv_dat);
+ SvIVX(sv_dat)= 1;
+ }
+
+ /*sv_dump(sv_dat);*/
+ }
+ nextchar(pRExC_state);
+ paren = 1;
+ goto capturing_parens;
+ }
+ RExC_seen |= REG_SEEN_LOOKBEHIND;
RExC_parse++;
case '=': /* (?=...) */
case '!': /* (?!...) */
RExC_seen_zerolen++;
+ if (*RExC_parse == ')') {
+ ret=reg_node(pRExC_state, OPFAIL);
+ nextchar(pRExC_state);
+ return ret;
+ }
case ':': /* (?:...) */
case '>': /* (?>...) */
break;
nextchar(pRExC_state);
*flagp = TRYAGAIN;
return NULL;
+ case '0' : /* (?0) */
+ case 'R' : /* (?R) */
+ if (*RExC_parse != ')')
+ FAIL("Sequence (?R) not terminated");
+ ret = reg_node(pRExC_state, GOSTART);
+ nextchar(pRExC_state);
+ return ret;
+ /*notreached*/
+ { /* named and numeric backreferences */
+ I32 num;
+ char * parse_start;
+ case '&': /* (?&NAME) */
+ parse_start = RExC_parse - 1;
+ {
+ SV *sv_dat = reg_scan_name(pRExC_state,
+ SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
+ num = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
+ }
+ goto gen_recurse_regop;
+ /* NOT REACHED */
+ case '+':
+ if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
+ RExC_parse++;
+ vFAIL("Illegal pattern");
+ }
+ goto parse_recursion;
+ /* NOT REACHED*/
+ case '-': /* (?-1) */
+ if (!(RExC_parse[0] >= '1' && RExC_parse[0] <= '9')) {
+ RExC_parse--; /* rewind to let it be handled later */
+ goto parse_flags;
+ }
+ /*FALLTHROUGH */
+ case '1': case '2': case '3': case '4': /* (?1) */
+ case '5': case '6': case '7': case '8': case '9':
+ RExC_parse--;
+ parse_recursion:
+ num = atoi(RExC_parse);
+ parse_start = RExC_parse - 1; /* MJD */
+ if (*RExC_parse == '-')
+ RExC_parse++;
+ while (isDIGIT(*RExC_parse))
+ RExC_parse++;
+ if (*RExC_parse!=')')
+ vFAIL("Expecting close bracket");
+
+ gen_recurse_regop:
+ if ( paren == '-' ) {
+ /*
+ Diagram of capture buffer numbering.
+ Top line is the normal capture buffer numbers
+ Botton line is the negative indexing as from
+ the X (the (?-2))
+
+ + 1 2 3 4 5 X 6 7
+ /(a(x)y)(a(b(c(?-2)d)e)f)(g(h))/
+ - 5 4 3 2 1 X x x
+
+ */
+ num = RExC_npar + num;
+ if (num < 1) {
+ RExC_parse++;
+ vFAIL("Reference to nonexistent group");
+ }
+ } else if ( paren == '+' ) {
+ num = RExC_npar + num - 1;
+ }
+
+ ret = reganode(pRExC_state, GOSUB, num);
+ if (!SIZE_ONLY) {
+ if (num > (I32)RExC_rx->nparens) {
+ RExC_parse++;
+ vFAIL("Reference to nonexistent group");
+ }
+ ARG2L_SET( ret, RExC_recurse_count++);
+ RExC_emit++;
+ DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
+ "Recurse #%"UVuf" to %"IVdf"\n", (UV)ARG(ret), (IV)ARG2L(ret)));
+ } else {
+ RExC_size++;
+ }
+ RExC_seen |= REG_SEEN_RECURSE;
+ Set_Node_Length(ret, 1 + regarglen[OP(ret)]); /* MJD */
+ Set_Node_Offset(ret, parse_start); /* MJD */
+
+ nextchar(pRExC_state);
+ return ret;
+ } /* named and numeric backreferences */
+ /* NOT REACHED */
+
case 'p': /* (?p...) */
if (SIZE_ONLY && ckWARN2(WARN_DEPRECATED, WARN_REGEXP))
vWARNdep(RExC_parse, "(?p{}) is deprecated - use (??{})");
/* FALL THROUGH */
case '{': /* (?{...}) */
{
- I32 count = 1, n = 0;
+ I32 count = 1;
+ U32 n = 0;
char c;
char *s = RExC_parse;
LEAVE;
n = add_data(pRExC_state, 3, "nop");
- RExC_rx->data->data[n] = (void*)rop;
- RExC_rx->data->data[n+1] = (void*)sop;
- RExC_rx->data->data[n+2] = (void*)pad;
+ RExC_rxi->data->data[n] = (void*)rop;
+ RExC_rxi->data->data[n+1] = (void*)sop;
+ RExC_rxi->data->data[n+2] = (void*)pad;
SvREFCNT_dec(sv);
}
else { /* First pass */
}
case '(': /* (?(?{...})...) and (?(?=...)...) */
{
+ int is_define= 0;
if (RExC_parse[0] == '?') { /* (?(?...)) */
if (RExC_parse[1] == '=' || RExC_parse[1] == '!'
|| RExC_parse[1] == '<'
goto insert_if;
}
}
+ else if ( RExC_parse[0] == '<' /* (?(<NAME>)...) */
+ || RExC_parse[0] == '\'' ) /* (?('NAME')...) */
+ {
+ char ch = RExC_parse[0] == '<' ? '>' : '\'';
+ char *name_start= RExC_parse++;
+ U32 num = 0;
+ SV *sv_dat=reg_scan_name(pRExC_state,
+ SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
+ if (RExC_parse == name_start || *RExC_parse != ch)
+ vFAIL2("Sequence (?(%c... not terminated",
+ (ch == '>' ? '<' : ch));
+ RExC_parse++;
+ if (!SIZE_ONLY) {
+ num = add_data( pRExC_state, 1, "S" );
+ RExC_rxi->data->data[num]=(void*)sv_dat;
+ SvREFCNT_inc(sv_dat);
+ }
+ ret = reganode(pRExC_state,NGROUPP,num);
+ goto insert_if_check_paren;
+ }
+ else if (RExC_parse[0] == 'D' &&
+ RExC_parse[1] == 'E' &&
+ RExC_parse[2] == 'F' &&
+ RExC_parse[3] == 'I' &&
+ RExC_parse[4] == 'N' &&
+ RExC_parse[5] == 'E')
+ {
+ ret = reganode(pRExC_state,DEFINEP,0);
+ RExC_parse +=6 ;
+ is_define = 1;
+ goto insert_if_check_paren;
+ }
+ else if (RExC_parse[0] == 'R') {
+ RExC_parse++;
+ parno = 0;
+ if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
+ parno = atoi(RExC_parse++);
+ while (isDIGIT(*RExC_parse))
+ RExC_parse++;
+ } else if (RExC_parse[0] == '&') {
+ SV *sv_dat;
+ RExC_parse++;
+ sv_dat = reg_scan_name(pRExC_state,
+ SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
+ parno = sv_dat ? *((I32 *)SvPVX(sv_dat)) : 0;
+ }
+ ret = reganode(pRExC_state,INSUBP,parno);
+ goto insert_if_check_paren;
+ }
else if (RExC_parse[0] >= '1' && RExC_parse[0] <= '9' ) {
/* (?(1)...) */
char c;
RExC_parse++;
ret = reganode(pRExC_state, GROUPP, parno);
+ insert_if_check_paren:
if ((c = *nextchar(pRExC_state)) != ')')
vFAIL("Switch condition not recognized");
insert_if:
if (flags&HASWIDTH)
*flagp |= HASWIDTH;
if (c == '|') {
+ if (is_define)
+ vFAIL("(?(DEFINE)....) does not allow branches");
lastbr = reganode(pRExC_state, IFTHEN, 0); /* Fake one for optimizer. */
regbranch(pRExC_state, &flags, 1,depth+1);
REGTAIL(pRExC_state, ret, lastbr);
}
}
else { /* (...) */
+ capturing_parens:
parno = RExC_npar;
RExC_npar++;
+
ret = reganode(pRExC_state, OPEN, parno);
+ if (!SIZE_ONLY ){
+ if (!RExC_nestroot)
+ RExC_nestroot = parno;
+ if (RExC_seen & REG_SEEN_RECURSE) {
+ DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
+ "Setting open paren #%"IVdf" to %d\n",
+ (IV)parno, REG_NODE_NUM(ret)));
+ RExC_open_parens[parno-1]= ret;
+ }
+ }
Set_Node_Length(ret, 1); /* MJD */
Set_Node_Offset(ret, RExC_parse); /* MJD */
is_open = 1;
return(NULL);
if (*RExC_parse == '|') {
if (!SIZE_ONLY && RExC_extralen) {
- reginsert(pRExC_state, BRANCHJ, br);
+ reginsert(pRExC_state, BRANCHJ, br, depth+1);
}
else { /* MJD */
- reginsert(pRExC_state, BRANCH, br);
+ reginsert(pRExC_state, BRANCH, br, depth+1);
Set_Node_Length(br, paren != 0);
Set_Node_Offset_To_R(br-RExC_emit_start, parse_start-RExC_start);
}
ender = reg_node(pRExC_state, TAIL);
break;
case 1:
+ RExC_cpar++;
ender = reganode(pRExC_state, CLOSE, parno);
+ if (!SIZE_ONLY && RExC_seen & REG_SEEN_RECURSE) {
+ DEBUG_OPTIMISE_MORE_r(PerlIO_printf(Perl_debug_log,
+ "Setting close paren #%"IVdf" to %d\n",
+ (IV)parno, REG_NODE_NUM(ender)));
+ RExC_close_parens[parno-1]= ender;
+ if (RExC_nestroot == parno)
+ RExC_nestroot = 0;
+ }
Set_Node_Offset(ender,RExC_parse+1); /* MJD */
Set_Node_Length(ender,1); /* MJD */
break;
break;
case 0:
ender = reg_node(pRExC_state, END);
+ if (!SIZE_ONLY) {
+ assert(!RExC_opend); /* there can only be one! */
+ RExC_opend = ender;
+ }
break;
}
- REGTAIL_STUDY(pRExC_state, lastbr, ender);
+ REGTAIL(pRExC_state, lastbr, ender);
if (have_branch && !SIZE_ONLY) {
+ if (depth==1)
+ RExC_seen |= REG_TOP_LEVEL_BRANCHES;
+
/* Hook the tails of the branches to the closing node. */
for (br = ret; br; br = regnext(br)) {
const U8 op = PL_regkind[OP(br)];
if (paren == '>')
node = SUSPEND, flag = 0;
- reginsert(pRExC_state, node,ret);
+ reginsert(pRExC_state, node,ret, depth+1);
Set_Node_Cur_Length(ret);
Set_Node_Offset(ret, parse_start + 1);
ret->flags = flag;
I32 min;
I32 max = REG_INFTY;
char *parse_start;
+ const char *maxpos = NULL;
GET_RE_DEBUG_FLAGS_DECL;
DEBUG_PARSE("piec");
op = *RExC_parse;
if (op == '{' && regcurly(RExC_parse)) {
- const char *maxpos = NULL;
+ maxpos = NULL;
parse_start = RExC_parse; /* MJD */
next = RExC_parse + 1;
while (isDIGIT(*next) || *next == ',') {
do_curly:
if ((flags&SIMPLE)) {
RExC_naughty += 2 + RExC_naughty / 2;
- reginsert(pRExC_state, CURLY, ret);
+ reginsert(pRExC_state, CURLY, ret, depth+1);
Set_Node_Offset(ret, parse_start+1); /* MJD */
Set_Node_Cur_Length(ret);
}
w->flags = 0;
REGTAIL(pRExC_state, ret, w);
if (!SIZE_ONLY && RExC_extralen) {
- reginsert(pRExC_state, LONGJMP,ret);
- reginsert(pRExC_state, NOTHING,ret);
+ reginsert(pRExC_state, LONGJMP,ret, depth+1);
+ reginsert(pRExC_state, NOTHING,ret, depth+1);
NEXT_OFF(ret) = 3; /* Go over LONGJMP. */
}
- reginsert(pRExC_state, CURLYX,ret);
+ reginsert(pRExC_state, CURLYX,ret, depth+1);
/* MJD hk */
Set_Node_Offset(ret, parse_start+1);
Set_Node_Length(ret,
*flagp = (op != '+') ? (WORST|SPSTART|HASWIDTH) : (WORST|HASWIDTH);
if (op == '*' && (flags&SIMPLE)) {
- reginsert(pRExC_state, STAR, ret);
+ reginsert(pRExC_state, STAR, ret, depth+1);
ret->flags = 0;
RExC_naughty += 4;
}
goto do_curly;
}
else if (op == '+' && (flags&SIMPLE)) {
- reginsert(pRExC_state, PLUS, ret);
+ reginsert(pRExC_state, PLUS, ret, depth+1);
ret->flags = 0;
RExC_naughty += 3;
}
origparse);
}
- if (*RExC_parse == '?') {
+ if (RExC_parse < RExC_end && *RExC_parse == '?') {
nextchar(pRExC_state);
- reginsert(pRExC_state, MINMOD, ret);
+ reginsert(pRExC_state, MINMOD, ret, depth+1);
REGTAIL(pRExC_state, ret, ret + NODE_STEP_REGNODE);
}
- if (ISMULT2(RExC_parse)) {
+#ifndef REG_ALLOW_MINMOD_SUSPEND
+ else
+#endif
+ if (RExC_parse < RExC_end && *RExC_parse == '+') {
+ regnode *ender;
+ nextchar(pRExC_state);
+ ender = reg_node(pRExC_state, SUCCEED);
+ REGTAIL(pRExC_state, ret, ender);
+ reginsert(pRExC_state, SUSPEND, ret, depth+1);
+ ret->flags = 0;
+ ender = reg_node(pRExC_state, TAIL);
+ REGTAIL(pRExC_state, ret, ender);
+ /*ret= ender;*/
+ }
+
+ if (RExC_parse < RExC_end && ISMULT2(RExC_parse)) {
RExC_parse++;
vFAIL("Nested quantifiers");
}
return(ret);
}
+
+/* reg_namedseq(pRExC_state,UVp)
+
+ This is expected to be called by a parser routine that has
+ recognized'\N' and needs to handle the rest. RExC_parse is
+ expected to point at the first char following the N at the time
+ of the call.
+
+ If valuep is non-null then it is assumed that we are parsing inside
+ of a charclass definition and the first codepoint in the resolved
+ string is returned via *valuep and the routine will return NULL.
+ In this mode if a multichar string is returned from the charnames
+ handler a warning will be issued, and only the first char in the
+ sequence will be examined. If the string returned is zero length
+ then the value of *valuep is undefined and NON-NULL will
+ be returned to indicate failure. (This will NOT be a valid pointer
+ to a regnode.)
+
+ If value is null then it is assumed that we are parsing normal text
+ and inserts a new EXACT node into the program containing the resolved
+ string and returns a pointer to the new node. If the string is
+ zerolength a NOTHING node is emitted.
+
+ On success RExC_parse is set to the char following the endbrace.
+ Parsing failures will generate a fatal errorvia vFAIL(...)
+
+ NOTE: We cache all results from the charnames handler locally in
+ the RExC_charnames hash (created on first use) to prevent a charnames
+ handler from playing silly-buggers and returning a short string and
+ then a long string for a given pattern. Since the regexp program
+ size is calculated during an initial parse this would result
+ in a buffer overrun so we cache to prevent the charname result from
+ changing during the course of the parse.
+
+ */
+STATIC regnode *
+S_reg_namedseq(pTHX_ RExC_state_t *pRExC_state, UV *valuep)
+{
+ char * name; /* start of the content of the name */
+ char * endbrace; /* endbrace following the name */
+ SV *sv_str = NULL;
+ SV *sv_name = NULL;
+ STRLEN len; /* this has various purposes throughout the code */
+ bool cached = 0; /* if this is true then we shouldn't refcount dev sv_str */
+ regnode *ret = NULL;
+
+ if (*RExC_parse != '{') {
+ vFAIL("Missing braces on \\N{}");
+ }
+ name = RExC_parse+1;
+ endbrace = strchr(RExC_parse, '}');
+ if ( ! endbrace ) {
+ RExC_parse++;
+ vFAIL("Missing right brace on \\N{}");
+ }
+ RExC_parse = endbrace + 1;
+
+
+ /* RExC_parse points at the beginning brace,
+ endbrace points at the last */
+ if ( name[0]=='U' && name[1]=='+' ) {
+ /* its a "unicode hex" notation {U+89AB} */
+ I32 fl = PERL_SCAN_ALLOW_UNDERSCORES
+ | PERL_SCAN_DISALLOW_PREFIX
+ | (SIZE_ONLY ? PERL_SCAN_SILENT_ILLDIGIT : 0);
+ UV cp;
+ len = (STRLEN)(endbrace - name - 2);
+ cp = grok_hex(name + 2, &len, &fl, NULL);
+ if ( len != (STRLEN)(endbrace - name - 2) ) {
+ cp = 0xFFFD;
+ }
+ if (cp > 0xff)
+ RExC_utf8 = 1;
+ if ( valuep ) {
+ *valuep = cp;
+ return NULL;
+ }
+ sv_str= Perl_newSVpvf_nocontext("%c",(int)cp);
+ } else {
+ /* fetch the charnames handler for this scope */
+ HV * const table = GvHV(PL_hintgv);
+ SV **cvp= table ?
+ hv_fetchs(table, "charnames", FALSE) :
+ NULL;
+ SV *cv= cvp ? *cvp : NULL;
+ HE *he_str;
+ int count;
+ /* create an SV with the name as argument */
+ sv_name = newSVpvn(name, endbrace - name);
+
+ if (!table || !(PL_hints & HINT_LOCALIZE_HH)) {
+ vFAIL2("Constant(\\N{%s}) unknown: "
+ "(possibly a missing \"use charnames ...\")",
+ SvPVX(sv_name));
+ }
+ if (!cvp || !SvOK(*cvp)) { /* when $^H{charnames} = undef; */
+ vFAIL2("Constant(\\N{%s}): "
+ "$^H{charnames} is not defined",SvPVX(sv_name));
+ }
+
+
+
+ if (!RExC_charnames) {
+ /* make sure our cache is allocated */
+ RExC_charnames = newHV();
+ sv_2mortal((SV*)RExC_charnames);
+ }
+ /* see if we have looked this one up before */
+ he_str = hv_fetch_ent( RExC_charnames, sv_name, 0, 0 );
+ if ( he_str ) {
+ sv_str = HeVAL(he_str);
+ cached = 1;
+ } else {
+ dSP ;
+
+ ENTER ;
+ SAVETMPS ;
+ PUSHMARK(SP) ;
+
+ XPUSHs(sv_name);
+
+ PUTBACK ;
+
+ count= call_sv(cv, G_SCALAR);
+
+ if (count == 1) { /* XXXX is this right? dmq */
+ sv_str = POPs;
+ SvREFCNT_inc_simple_void(sv_str);
+ }
+
+ SPAGAIN ;
+ PUTBACK ;
+ FREETMPS ;
+ LEAVE ;
+
+ if ( !sv_str || !SvOK(sv_str) ) {
+ vFAIL2("Constant(\\N{%s}): Call to &{$^H{charnames}} "
+ "did not return a defined value",SvPVX(sv_name));
+ }
+ if (hv_store_ent( RExC_charnames, sv_name, sv_str, 0))
+ cached = 1;
+ }
+ }
+ if (valuep) {
+ char *p = SvPV(sv_str, len);
+ if (len) {
+ STRLEN numlen = 1;
+ if ( SvUTF8(sv_str) ) {
+ *valuep = utf8_to_uvchr((U8*)p, &numlen);
+ if (*valuep > 0x7F)
+ RExC_utf8 = 1;
+ /* XXXX
+ We have to turn on utf8 for high bit chars otherwise
+ we get failures with
+
+ "ss" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
+ "SS" =~ /[\N{LATIN SMALL LETTER SHARP S}]/i
+
+ This is different from what \x{} would do with the same
+ codepoint, where the condition is > 0xFF.
+ - dmq
+ */
+
+
+ } else {
+ *valuep = (UV)*p;
+ /* warn if we havent used the whole string? */
+ }
+ if (numlen<len && SIZE_ONLY && ckWARN(WARN_REGEXP)) {
+ vWARN2(RExC_parse,
+ "Ignoring excess chars from \\N{%s} in character class",
+ SvPVX(sv_name)
+ );
+ }
+ } else if (SIZE_ONLY && ckWARN(WARN_REGEXP)) {
+ vWARN2(RExC_parse,
+ "Ignoring zero length \\N{%s} in character class",
+ SvPVX(sv_name)
+ );
+ }
+ if (sv_name)
+ SvREFCNT_dec(sv_name);
+ if (!cached)
+ SvREFCNT_dec(sv_str);
+ return len ? NULL : (regnode *)&len;
+ } else if(SvCUR(sv_str)) {
+
+ char *s;
+ char *p, *pend;
+ STRLEN charlen = 1;
+ char * parse_start = name-3; /* needed for the offsets */
+ GET_RE_DEBUG_FLAGS_DECL; /* needed for the offsets */
+
+ ret = reg_node(pRExC_state,
+ (U8)(FOLD ? (LOC ? EXACTFL : EXACTF) : EXACT));
+ s= STRING(ret);
+
+ if ( RExC_utf8 && !SvUTF8(sv_str) ) {
+ sv_utf8_upgrade(sv_str);
+ } else if ( !RExC_utf8 && SvUTF8(sv_str) ) {
+ RExC_utf8= 1;
+ }
+
+ p = SvPV(sv_str, len);
+ pend = p + len;
+ /* len is the length written, charlen is the size the char read */
+ for ( len = 0; p < pend; p += charlen ) {
+ if (UTF) {
+ UV uvc = utf8_to_uvchr((U8*)p, &charlen);
+ if (FOLD) {
+ STRLEN foldlen,numlen;
+ U8 tmpbuf[UTF8_MAXBYTES_CASE+1], *foldbuf;
+ uvc = toFOLD_uni(uvc, tmpbuf, &foldlen);
+ /* Emit all the Unicode characters. */
+
+ for (foldbuf = tmpbuf;
+ foldlen;
+ foldlen -= numlen)
+ {
+ uvc = utf8_to_uvchr(foldbuf, &numlen);
+ if (numlen > 0) {
+ const STRLEN unilen = reguni(pRExC_state, uvc, s);
+ s += unilen;
+ len += unilen;
+ /* In EBCDIC the numlen
+ * and unilen can differ. */
+ foldbuf += numlen;
+ if (numlen >= foldlen)
+ break;
+ }
+ else
+ break; /* "Can't happen." */
+ }
+ } else {
+ const STRLEN unilen = reguni(pRExC_state, uvc, s);
+ if (unilen > 0) {
+ s += unilen;
+ len += unilen;
+ }
+ }
+ } else {
+ len++;
+ REGC(*p, s++);
+ }
+ }
+ if (SIZE_ONLY) {
+ RExC_size += STR_SZ(len);
+ } else {
+ STR_LEN(ret) = len;
+ RExC_emit += STR_SZ(len);
+ }
+ Set_Node_Cur_Length(ret); /* MJD */
+ RExC_parse--;
+ nextchar(pRExC_state);
+ } else {
+ ret = reg_node(pRExC_state,NOTHING);
+ }
+ if (!cached) {
+ SvREFCNT_dec(sv_str);
+ }
+ if (sv_name) {
+ SvREFCNT_dec(sv_name);
+ }
+ return ret;
+
+}
+
+
+/*
+ * reg_recode
+ *
+ * It returns the code point in utf8 for the value in *encp.
+ * value: a code value in the source encoding
+ * encp: a pointer to an Encode object
+ *
+ * If the result from Encode is not a single character,
+ * it returns U+FFFD (Replacement character) and sets *encp to NULL.
+ */
+STATIC UV
+S_reg_recode(pTHX_ const char value, SV **encp)
+{
+ STRLEN numlen = 1;
+ SV * const sv = sv_2mortal(newSVpvn(&value, numlen));
+ const char * const s = encp && *encp ? sv_recode_to_utf8(sv, *encp)
+ : SvPVX(sv);
+ const STRLEN newlen = SvCUR(sv);
+ UV uv = UNICODE_REPLACEMENT;
+
+ if (newlen)
+ uv = SvUTF8(sv)
+ ? utf8n_to_uvchr((U8*)s, newlen, &numlen, UTF8_ALLOW_DEFAULT)
+ : *(U8*)s;
+
+ if (!newlen || numlen != newlen) {
+ uv = UNICODE_REPLACEMENT;
+ if (encp)
+ *encp = NULL;
+ }
+ return uv;
+}
+
+
/*
- regatom - the lowest level
*
case '^':
RExC_seen_zerolen++;
nextchar(pRExC_state);
- if (RExC_flags & PMf_MULTILINE)
+ if (RExC_flags & RXf_PMf_MULTILINE)
ret = reg_node(pRExC_state, MBOL);
- else if (RExC_flags & PMf_SINGLELINE)
+ else if (RExC_flags & RXf_PMf_SINGLELINE)
ret = reg_node(pRExC_state, SBOL);
else
ret = reg_node(pRExC_state, BOL);
nextchar(pRExC_state);
if (*RExC_parse)
RExC_seen_zerolen++;
- if (RExC_flags & PMf_MULTILINE)
+ if (RExC_flags & RXf_PMf_MULTILINE)
ret = reg_node(pRExC_state, MEOL);
- else if (RExC_flags & PMf_SINGLELINE)
+ else if (RExC_flags & RXf_PMf_SINGLELINE)
ret = reg_node(pRExC_state, SEOL);
else
ret = reg_node(pRExC_state, EOL);
break;
case '.':
nextchar(pRExC_state);
- if (RExC_flags & PMf_SINGLELINE)
+ if (RExC_flags & RXf_PMf_SINGLELINE)
ret = reg_node(pRExC_state, SANY);
else
ret = reg_node(pRExC_state, REG_ANY);
*flagp |= HASWIDTH|SIMPLE;
}
break;
+ case 'N':
+ /* Handle \N{NAME} here and not below because it can be
+ multicharacter. join_exact() will join them up later on.
+ Also this makes sure that things like /\N{BLAH}+/ and
+ \N{BLAH} being multi char Just Happen. dmq*/
+ ++RExC_parse;
+ ret= reg_namedseq(pRExC_state, NULL);
+ break;
+ case 'k': /* Handle \k<NAME> and \k'NAME' */
+ {
+ char ch= RExC_parse[1];
+ if (ch != '<' && ch != '\'') {
+ if (SIZE_ONLY)
+ vWARN( RExC_parse + 1,
+ "Possible broken named back reference treated as literal k");
+ parse_start--;
+ goto defchar;
+ } else {
+ char* name_start = (RExC_parse += 2);
+ U32 num = 0;
+ SV *sv_dat = reg_scan_name(pRExC_state,
+ SIZE_ONLY ? REG_RSN_RETURN_NULL : REG_RSN_RETURN_DATA);
+ ch= (ch == '<') ? '>' : '\'';
+
+ if (RExC_parse == name_start || *RExC_parse != ch)
+ vFAIL2("Sequence \\k%c... not terminated",
+ (ch == '>' ? '<' : ch));
+
+ RExC_sawback = 1;
+ ret = reganode(pRExC_state,
+ (U8)(FOLD ? (LOC ? NREFFL : NREFF) : NREF),
+ num);
+ *flagp |= HASWIDTH;
+
+
+ if (!SIZE_ONLY) {
+ num = add_data( pRExC_state, 1, "S" );
+ ARG_SET(ret,num);
+ RExC_rxi->data->data[num]=(void*)sv_dat;
+ SvREFCNT_inc(sv_dat);
+ }
+ /* override incorrect value set in reganode MJD */
+ Set_Node_Offset(ret, parse_start+1);
+ Set_Node_Cur_Length(ret); /* MJD */
+ nextchar(pRExC_state);
+
+ }
+ break;
+ }
case 'n':
case 'r':
case 't':
case 'c':
case '0':
goto defchar;
+ case 'R':
case '1': case '2': case '3': case '4':
case '5': case '6': case '7': case '8': case '9':
{
- const I32 num = atoi(RExC_parse);
-
+ I32 num;
+ bool isrel=(*RExC_parse=='R');
+ if (isrel)
+ RExC_parse++;
+ num = atoi(RExC_parse);
+ if (isrel) {
+ num = RExC_cpar - num;
+ if (num < 1)
+ vFAIL("Reference to nonexistent or unclosed group");
+ }
if (num > 9 && num >= RExC_npar)
goto defchar;
else {
while (isDIGIT(*RExC_parse))
RExC_parse++;
- if (!SIZE_ONLY && num > (I32)RExC_rx->nparens)
- vFAIL("Reference to nonexistent group");
+ if (!SIZE_ONLY) {
+ if (num > (I32)RExC_rx->nparens)
+ vFAIL("Reference to nonexistent group");
+ /* People make this error all the time apparently.
+ So we cant fail on it, even though we should
+
+ else if (num >= RExC_cpar)
+ vFAIL("Reference to unclosed group will always match");
+ */
+ }
RExC_sawback = 1;
ret = reganode(pRExC_state,
(U8)(FOLD ? (LOC ? REFFL : REFF) : REF),
break;
case '#':
- if (RExC_flags & PMf_EXTENDED) {
+ if (RExC_flags & RXf_PMf_EXTENDED) {
while (RExC_parse < RExC_end && *RExC_parse != '\n')
RExC_parse++;
if (RExC_parse < RExC_end)
{
char * const oldp = p;
- if (RExC_flags & PMf_EXTENDED)
+ if (RExC_flags & RXf_PMf_EXTENDED)
p = regwhite(p, RExC_end);
switch (*p) {
case '^':
case 'D':
case 'p':
case 'P':
+ case 'N':
+ case 'R':
--p;
goto loopdone;
case 'n':
ender = grok_hex(p, &numlen, &flags, NULL);
p += numlen;
}
+ if (PL_encoding && ender < 0x100)
+ goto recode_encoding;
break;
case 'c':
p++;
--p;
goto loopdone;
}
+ if (PL_encoding && ender < 0x100)
+ goto recode_encoding;
+ break;
+ recode_encoding:
+ {
+ SV* enc = PL_encoding;
+ ender = reg_recode((const char)(U8)ender, &enc);
+ if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
+ vWARN(p, "Invalid escape in the specified encoding");
+ RExC_utf8 = 1;
+ }
break;
case '\0':
if (p >= RExC_end)
ender = *p++;
break;
}
- if (RExC_flags & PMf_EXTENDED)
+ if (RExC_flags & RXf_PMf_EXTENDED)
p = regwhite(p, RExC_end);
if (UTF && FOLD) {
/* Prime the casefolded buffer. */
break;
}
- /* If the encoding pragma is in effect recode the text of
- * any EXACT-kind nodes. */
- if (PL_encoding && PL_regkind[OP(ret)] == EXACT) {
- const STRLEN oldlen = STR_LEN(ret);
- SV * const sv = sv_2mortal(newSVpvn(STRING(ret), oldlen));
-
- if (RExC_utf8)
- SvUTF8_on(sv);
- if (sv_utf8_downgrade(sv, TRUE)) {
- const char * const s = sv_recode_to_utf8(sv, PL_encoding);
- const STRLEN newlen = SvCUR(sv);
-
- if (SvUTF8(sv))
- RExC_utf8 = 1;
- if (!SIZE_ONLY) {
- GET_RE_DEBUG_FLAGS_DECL;
- DEBUG_COMPILE_r(PerlIO_printf(Perl_debug_log, "recode %*s to %*s\n",
- (int)oldlen, STRING(ret),
- (int)newlen, s));
- Copy(s, STRING(ret), newlen, char);
- STR_LEN(ret) += newlen - oldlen;
- RExC_emit += STR_SZ(newlen) - STR_SZ(oldlen);
- } else
- RExC_size += STR_SZ(newlen) - STR_SZ(oldlen);
- }
- }
-
return(ret);
}
S_regclass(pTHX_ RExC_state_t *pRExC_state, U32 depth)
{
dVAR;
- register UV value;
+ register UV value = 0;
register UV nextvalue;
register IV prevvalue = OOB_UNICODE;
register IV range = 0;
case we need to change the emitted regop to an EXACT. */
const char * orig_parse = RExC_parse;
GET_RE_DEBUG_FLAGS_DECL;
+#ifndef DEBUGGING
+ PERL_UNUSED_ARG(depth);
+#endif
+
DEBUG_PARSE("clas");
/* Assume we are going to generate an ANYOF node. */
if (UCHARAT(RExC_parse) == ']')
goto charclassloop;
+parseit:
while (RExC_parse < RExC_end && UCHARAT(RExC_parse) != ']') {
charclassloop:
case 'S': namedclass = ANYOF_NSPACE; break;
case 'd': namedclass = ANYOF_DIGIT; break;
case 'D': namedclass = ANYOF_NDIGIT; break;
+ case 'N': /* Handle \N{NAME} in class */
+ {
+ /* We only pay attention to the first char of
+ multichar strings being returned. I kinda wonder
+ if this makes sense as it does change the behaviour
+ from earlier versions, OTOH that behaviour was broken
+ as well. */
+ UV v; /* value is register so we cant & it /grrr */
+ if (reg_namedseq(pRExC_state, &v)) {
+ goto parseit;
+ }
+ value= v;
+ }
+ break;
case 'p':
case 'P':
{
value = grok_hex(RExC_parse, &numlen, &flags, NULL);
RExC_parse += numlen;
}
+ if (PL_encoding && value < 0x100)
+ goto recode_encoding;
break;
case 'c':
value = UCHARAT(RExC_parse++);
break;
case '0': case '1': case '2': case '3': case '4':
case '5': case '6': case '7': case '8': case '9':
- {
- I32 flags = 0;
- numlen = 3;
- value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
- RExC_parse += numlen;
- break;
- }
+ {
+ I32 flags = 0;
+ numlen = 3;
+ value = grok_oct(--RExC_parse, &numlen, &flags, NULL);
+ RExC_parse += numlen;
+ if (PL_encoding && value < 0x100)
+ goto recode_encoding;
+ break;
+ }
+ recode_encoding:
+ {
+ SV* enc = PL_encoding;
+ value = reg_recode((const char)(U8)value, &enc);
+ if (!enc && SIZE_ONLY && ckWARN(WARN_REGEXP))
+ vWARN(RExC_parse,
+ "Invalid escape in the specified encoding");
+ break;
+ }
default:
if (!SIZE_ONLY && isALPHA(value) && ckWARN(WARN_REGEXP))
vWARN2(RExC_parse,
STRLEN foldlen;
const UV f = to_uni_fold(natvalue, foldbuf, &foldlen);
+#ifdef EBCDIC /* RD t/uni/fold ff and 6b */
+ if (RExC_precomp[0] == ':' &&
+ RExC_precomp[1] == '[' &&
+ (f == 0xDF || f == 0x92)) {
+ f = NATIVE_TO_UNI(f);
+ }
+#endif
/* If folding and foldable and a single
* character, insert also the folded version
* to the charclass. */
if (f != value) {
+#ifdef EBCDIC /* RD tunifold ligatures s,t fb05, fb06 */
+ if ((RExC_precomp[0] == ':' &&
+ RExC_precomp[1] == '[' &&
+ (f == 0xA2 &&
+ (value == 0xFB05 || value == 0xFB06))) ?
+ foldlen == ((STRLEN)UNISKIP(f) - 1) :
+ foldlen == (STRLEN)UNISKIP(f) )
+#else
if (foldlen == (STRLEN)UNISKIP(f))
+#endif
Perl_sv_catpvf(aTHX_ listsv,
"%04"UVxf"\n", f);
else {
av_store(av, 2, (SV*)unicode_alternate);
rv = newRV_noinc((SV*)av);
n = add_data(pRExC_state, 1, "s");
- RExC_rx->data->data[n] = (void*)rv;
+ RExC_rxi->data->data[n] = (void*)rv;
ARG_SET(ret, n);
}
return ret;
RExC_parse++;
continue;
}
- if (RExC_flags & PMf_EXTENDED) {
+ if (RExC_flags & RXf_PMf_EXTENDED) {
if (isSPACE(*RExC_parse)) {
RExC_parse++;
continue;
RExC_size += 1;
return(ret);
}
+#ifdef DEBUGGING
+ if (OP(RExC_emit) == 255)
+ Perl_croak(aTHX_ "panic: reg_node overrun trying to emit %s: %d ",
+ reg_name[op], OP(RExC_emit));
+#endif
NODE_ALIGN_FILL(ret);
ptr = ret;
FILL_ADVANCE_NODE(ptr, op);
}
RExC_emit = ptr;
-
return(ret);
}
if (SIZE_ONLY) {
SIZE_ALIGN(RExC_size);
RExC_size += 2;
+ /*
+ We can't do this:
+
+ assert(2==regarglen[op]+1);
+
+ Anything larger than this has to allocate the extra amount.
+ If we changed this to be:
+
+ RExC_size += (1 + regarglen[op]);
+
+ then it wouldn't matter. Its not clear what side effect
+ might come from that so its not done so far.
+ -- dmq
+ */
return(ret);
}
-
+#ifdef DEBUGGING
+ if (OP(RExC_emit) == 255)
+ Perl_croak(aTHX_ "panic: reganode overwriting end of allocated program space");
+#endif
NODE_ALIGN_FILL(ret);
ptr = ret;
FILL_ADVANCE_NODE_ARG(ptr, op, arg);
}
RExC_emit = ptr;
-
return(ret);
}
* Means relocating the operand.
*/
STATIC void
-S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd)
+S_reginsert(pTHX_ RExC_state_t *pRExC_state, U8 op, regnode *opnd, U32 depth)
{
dVAR;
register regnode *src;
register regnode *dst;
register regnode *place;
const int offset = regarglen[(U8)op];
+ const int size = NODE_STEP_REGNODE + offset;
GET_RE_DEBUG_FLAGS_DECL;
/* (PL_regkind[(U8)op] == CURLY ? EXTRA_STEP_2ARGS : 0); */
-
+ DEBUG_PARSE_FMT("inst"," - %s",reg_name[op]);
if (SIZE_ONLY) {
- RExC_size += NODE_STEP_REGNODE + offset;
+ RExC_size += size;
return;
}
src = RExC_emit;
- RExC_emit += NODE_STEP_REGNODE + offset;
+ RExC_emit += size;
dst = RExC_emit;
+ if (RExC_open_parens) {
+ int paren;
+ DEBUG_PARSE_FMT("inst"," - %"IVdf, (IV)RExC_npar);
+ for ( paren=0 ; paren < RExC_npar ; paren++ ) {
+ if ( RExC_open_parens[paren] >= opnd ) {
+ DEBUG_PARSE_FMT("open"," - %d",size);
+ RExC_open_parens[paren] += size;
+ } else {
+ DEBUG_PARSE_FMT("open"," - %s","ok");
+ }
+ if ( RExC_close_parens[paren] >= opnd ) {
+ DEBUG_PARSE_FMT("close"," - %d",size);
+ RExC_close_parens[paren] += size;
+ } else {
+ DEBUG_PARSE_FMT("close"," - %s","ok");
+ }
+ }
+ }
+
while (src > opnd) {
StructCopy(--src, --dst, regnode);
if (RExC_offsets) { /* MJD 20010112 */
? "Overwriting end of array!\n" : "OK",
(UV)(place - RExC_emit_start),
(UV)(RExC_parse - RExC_start),
- RExC_offsets[0]));
+ (UV)RExC_offsets[0]));
Set_Node_Offset(place, RExC_parse);
Set_Node_Length(place, 1);
}
dVAR;
register regnode *scan;
GET_RE_DEBUG_FLAGS_DECL;
+#ifndef DEBUGGING
+ PERL_UNUSED_ARG(depth);
+#endif
if (SIZE_ONLY)
return;
SV * const mysv=sv_newmortal();
DEBUG_PARSE_MSG((scan==p ? "tail" : ""));
regprop(RExC_rx, mysv, scan);
- PerlIO_printf(Perl_debug_log, "~ %s (%d)\n",
- SvPV_nolen_const(mysv), REG_NODE_NUM(scan));
+ PerlIO_printf(Perl_debug_log, "~ %s (%d) %s %s\n",
+ SvPV_nolen_const(mysv), REG_NODE_NUM(scan),
+ (temp == NULL ? "->" : ""),
+ (temp == NULL ? reg_name[OP(val)] : "")
+ );
});
if (temp == NULL)
break;
SV * const mysv=sv_newmortal();
DEBUG_PARSE_MSG((scan==p ? "tsdy" : ""));
regprop(RExC_rx, mysv, scan);
- PerlIO_printf(Perl_debug_log, "~ %s (%s) (%d)\n",
+ PerlIO_printf(Perl_debug_log, "~ %s (%d) -> %s\n",
SvPV_nolen_const(mysv),
- reg_name[exact],
- REG_NODE_NUM(scan));
+ REG_NODE_NUM(scan),
+ reg_name[exact]);
});
if (temp == NULL)
break;
SV * const mysv_val=sv_newmortal();
DEBUG_PARSE_MSG("");
regprop(RExC_rx, mysv_val, val);
- PerlIO_printf(Perl_debug_log, "~ attach to %s (%d) offset to %d\n",
- SvPV_nolen_const(mysv_val),
- REG_NODE_NUM(val),
- val - scan
+ PerlIO_printf(Perl_debug_log, "~ attach to %s (%"IVdf") offset to %"IVdf"\n",
+ SvPV_nolen_const(mysv_val),
+ (IV)REG_NODE_NUM(val),
+ (IV)(val - scan)
);
});
if (reg_off_by_arg[OP(scan)]) {
#ifdef DEBUGGING
dVAR;
SV * const sv = sv_newmortal();
+ SV *dsv= sv_newmortal();
+ RXi_GET_DECL(r,ri);
- (void)dumpuntil(r, r->program, r->program + 1, NULL, sv, 0);
+ (void)dumpuntil(r, ri->program, ri->program + 1, NULL, NULL, sv, 0, 0);
/* Header fields of interest. */
- if (r->anchored_substr)
+ if (r->anchored_substr) {
+ RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->anchored_substr),
+ RE_SV_DUMPLEN(r->anchored_substr), 30);
PerlIO_printf(Perl_debug_log,
- "anchored \"%s%.*s%s\"%s at %"IVdf" ",
- PL_colors[0],
- (int)(SvCUR(r->anchored_substr) - (SvTAIL(r->anchored_substr)!=0)),
- SvPVX_const(r->anchored_substr),
- PL_colors[1],
- SvTAIL(r->anchored_substr) ? "$" : "",
+ "anchored %s%s at %"IVdf" ",
+ s, RE_SV_TAIL(r->anchored_substr),
(IV)r->anchored_offset);
- else if (r->anchored_utf8)
+ } else if (r->anchored_utf8) {
+ RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->anchored_utf8),
+ RE_SV_DUMPLEN(r->anchored_utf8), 30);
PerlIO_printf(Perl_debug_log,
- "anchored utf8 \"%s%.*s%s\"%s at %"IVdf" ",
- PL_colors[0],
- (int)(SvCUR(r->anchored_utf8) - (SvTAIL(r->anchored_utf8)!=0)),
- SvPVX_const(r->anchored_utf8),
- PL_colors[1],
- SvTAIL(r->anchored_utf8) ? "$" : "",
+ "anchored utf8 %s%s at %"IVdf" ",
+ s, RE_SV_TAIL(r->anchored_utf8),
(IV)r->anchored_offset);
- if (r->float_substr)
+ }
+ if (r->float_substr) {
+ RE_PV_QUOTED_DECL(s, 0, dsv, SvPVX_const(r->float_substr),
+ RE_SV_DUMPLEN(r->float_substr), 30);
PerlIO_printf(Perl_debug_log,
- "floating \"%s%.*s%s\"%s at %"IVdf"..%"UVuf" ",
- PL_colors[0],
- (int)(SvCUR(r->float_substr) - (SvTAIL(r->float_substr)!=0)),
- SvPVX_const(r->float_substr),
- PL_colors[1],
- SvTAIL(r->float_substr) ? "$" : "",
+ "floating %s%s at %"IVdf"..%"UVuf" ",
+ s, RE_SV_TAIL(r->float_substr),
(IV)r->float_min_offset, (UV)r->float_max_offset);
- else if (r->float_utf8)
+ } else if (r->float_utf8) {
+ RE_PV_QUOTED_DECL(s, 1, dsv, SvPVX_const(r->float_utf8),
+ RE_SV_DUMPLEN(r->float_utf8), 30);
PerlIO_printf(Perl_debug_log,
- "floating utf8 \"%s%.*s%s\"%s at %"IVdf"..%"UVuf" ",
- PL_colors[0],
- (int)(SvCUR(r->float_utf8) - (SvTAIL(r->float_utf8)!=0)),
- SvPVX_const(r->float_utf8),
- PL_colors[1],
- SvTAIL(r->float_utf8) ? "$" : "",
+ "floating utf8 %s%s at %"IVdf"..%"UVuf" ",
+ s, RE_SV_TAIL(r->float_utf8),
(IV)r->float_min_offset, (UV)r->float_max_offset);
+ }
if (r->check_substr || r->check_utf8)
PerlIO_printf(Perl_debug_log,
- r->check_substr == r->float_substr
- && r->check_utf8 == r->float_utf8
- ? "(checking floating" : "(checking anchored");
- if (r->reganch & ROPT_NOSCAN)
+ (const char *)
+ (r->check_substr == r->float_substr
+ && r->check_utf8 == r->float_utf8
+ ? "(checking floating" : "(checking anchored"));
+ if (r->extflags & RXf_NOSCAN)
PerlIO_printf(Perl_debug_log, " noscan");
- if (r->reganch & ROPT_CHECK_ALL)
+ if (r->extflags & RXf_CHECK_ALL)
PerlIO_printf(Perl_debug_log, " isall");
if (r->check_substr || r->check_utf8)
PerlIO_printf(Perl_debug_log, ") ");
- if (r->regstclass) {
- regprop(r, sv, r->regstclass);
- PerlIO_printf(Perl_debug_log, "stclass \"%s\" ", SvPVX_const(sv));
+ if (ri->regstclass) {
+ regprop(r, sv, ri->regstclass);
+ PerlIO_printf(Perl_debug_log, "stclass %s ", SvPVX_const(sv));
}
- if (r->reganch & ROPT_ANCH) {
+ if (r->extflags & RXf_ANCH) {
PerlIO_printf(Perl_debug_log, "anchored");
- if (r->reganch & ROPT_ANCH_BOL)
+ if (r->extflags & RXf_ANCH_BOL)
PerlIO_printf(Perl_debug_log, "(BOL)");
- if (r->reganch & ROPT_ANCH_MBOL)
+ if (r->extflags & RXf_ANCH_MBOL)
PerlIO_printf(Perl_debug_log, "(MBOL)");
- if (r->reganch & ROPT_ANCH_SBOL)
+ if (r->extflags & RXf_ANCH_SBOL)
PerlIO_printf(Perl_debug_log, "(SBOL)");
- if (r->reganch & ROPT_ANCH_GPOS)
+ if (r->extflags & RXf_ANCH_GPOS)
PerlIO_printf(Perl_debug_log, "(GPOS)");
PerlIO_putc(Perl_debug_log, ' ');
}
- if (r->reganch & ROPT_GPOS_SEEN)
- PerlIO_printf(Perl_debug_log, "GPOS ");
- if (r->reganch & ROPT_SKIP)
+ if (r->extflags & RXf_GPOS_SEEN)
+ PerlIO_printf(Perl_debug_log, "GPOS:%"UVuf" ", (UV)r->gofs);
+ if (r->intflags & PREGf_SKIP)
PerlIO_printf(Perl_debug_log, "plus ");
- if (r->reganch & ROPT_IMPLICIT)
+ if (r->intflags & PREGf_IMPLICIT)
PerlIO_printf(Perl_debug_log, "implicit ");
- PerlIO_printf(Perl_debug_log, "minlen %ld ", (long) r->minlen);
- if (r->reganch & ROPT_EVAL_SEEN)
+ PerlIO_printf(Perl_debug_log, "minlen %"IVdf" ", (IV)r->minlen);
+ if (r->extflags & RXf_EVAL_SEEN)
PerlIO_printf(Perl_debug_log, "with eval ");
PerlIO_printf(Perl_debug_log, "\n");
- if (r->offsets) {
- const U32 len = r->offsets[0];
- GET_RE_DEBUG_FLAGS_DECL;
- DEBUG_OFFSETS_r({
- U32 i;
- PerlIO_printf(Perl_debug_log, "Offsets: [%"UVuf"]\n\t", (UV)r->offsets[0]);
- for (i = 1; i <= len; i++) {
- PerlIO_printf(Perl_debug_log, "%"UVuf":%"UVuf"[%"UVuf"] ",
- i, (UV)r->offsets[i*2-1], (UV)r->offsets[i*2]);
- }
- PerlIO_printf(Perl_debug_log, "\n");
- });
- }
#else
PERL_UNUSED_CONTEXT;
PERL_UNUSED_ARG(r);
#ifdef DEBUGGING
dVAR;
register int k;
+ RXi_GET_DECL(prog,progi);
+ GET_RE_DEBUG_FLAGS_DECL;
+
sv_setpvn(sv, "", 0);
- if (OP(o) >= reg_num) /* regnode.type is unsigned */
+
+ if (OP(o) > REGNODE_MAX) /* regnode.type is unsigned */
/* It would be nice to FAIL() here, but this may be called from
regexec.c, and it would be hard to supply pRExC_state. */
- Perl_croak(aTHX_ "Corrupted regexp opcode");
+ Perl_croak(aTHX_ "Corrupted regexp opcode %d > %d", (int)OP(o), (int)REGNODE_MAX);
sv_catpv(sv, reg_name[OP(o)]); /* Take off const! */
k = PL_regkind[OP(o)];
if (k == EXACT) {
SV * const dsv = sv_2mortal(newSVpvs(""));
- /* Using is_utf8_string() is a crude hack but it may
- * be the best for now since we have no flag "this EXACTish
- * node was UTF-8" --jhi */
- const bool do_utf8 = is_utf8_string((U8*)STRING(o), STR_LEN(o));
- const char * const s = do_utf8 ?
- pv_uni_display(dsv, (U8*)STRING(o), STR_LEN(o), 60,
- UNI_DISPLAY_REGEX) :
- STRING(o);
- const int len = do_utf8 ?
- strlen(s) :
- STR_LEN(o);
- Perl_sv_catpvf(aTHX_ sv, " <%s%.*s%s>",
- PL_colors[0],
- len, s,
- PL_colors[1]);
+ /* Using is_utf8_string() (via PERL_PV_UNI_DETECT)
+ * is a crude hack but it may be the best for now since
+ * we have no flag "this EXACTish node was UTF-8"
+ * --jhi */
+ const char * const s =
+ pv_pretty(dsv, STRING(o), STR_LEN(o), 60,
+ PL_colors[0], PL_colors[1],
+ PERL_PV_ESCAPE_UNI_DETECT |
+ PERL_PV_PRETTY_ELIPSES |
+ PERL_PV_PRETTY_LTGT
+ );
+ Perl_sv_catpvf(aTHX_ sv, " %s", s );
} else if (k == TRIE) {
- Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
/* print the details of the trie in dumpuntil instead, as
- * prog->data isn't available here */
+ * progi->data isn't available here */
+ const char op = OP(o);
+ const I32 n = ARG(o);
+ const reg_ac_data * const ac = IS_TRIE_AC(op) ?
+ (reg_ac_data *)progi->data->data[n] :
+ NULL;
+ const reg_trie_data * const trie
+ = (reg_trie_data*)progi->data->data[!IS_TRIE_AC(op) ? n : ac->trie];
+
+ Perl_sv_catpvf(aTHX_ sv, "-%s",reg_name[o->flags]);
+ DEBUG_TRIE_COMPILE_r(
+ Perl_sv_catpvf(aTHX_ sv,
+ "<S:%"UVuf"/%"IVdf" W:%"UVuf" L:%"UVuf"/%"UVuf" C:%"UVuf"/%"UVuf">",
+ (UV)trie->startstate,
+ (IV)trie->statecount-1, /* -1 because of the unused 0 element */
+ (UV)trie->wordcount,
+ (UV)trie->minlen,
+ (UV)trie->maxlen,
+ (UV)TRIE_CHARCOUNT(trie),
+ (UV)trie->uniquecharcount
+ )
+ );
+ if ( IS_ANYOF_TRIE(op) || trie->bitmap ) {
+ int i;
+ int rangestart = -1;
+ U8* bitmap = IS_ANYOF_TRIE(op) ? (U8*)ANYOF_BITMAP(o) : (U8*)TRIE_BITMAP(trie);
+ Perl_sv_catpvf(aTHX_ sv, "[");
+ for (i = 0; i <= 256; i++) {
+ if (i < 256 && BITMAP_TEST(bitmap,i)) {
+ if (rangestart == -1)
+ rangestart = i;
+ } else if (rangestart != -1) {
+ if (i <= rangestart + 3)
+ for (; rangestart < i; rangestart++)
+ put_byte(sv, rangestart);
+ else {
+ put_byte(sv, rangestart);
+ sv_catpvs(sv, "-");
+ put_byte(sv, i - 1);
+ }
+ rangestart = -1;
+ }
+ }
+ Perl_sv_catpvf(aTHX_ sv, "]");
+ }
+
} else if (k == CURLY) {
if (OP(o) == CURLYM || OP(o) == CURLYN || OP(o) == CURLYX)
Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* Parenth number */
}
else if (k == WHILEM && o->flags) /* Ordinal/of */
Perl_sv_catpvf(aTHX_ sv, "[%d/%d]", o->flags & 0xf, o->flags>>4);
- else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP )
+ else if (k == REF || k == OPEN || k == CLOSE || k == GROUPP || OP(o)==ACCEPT)
Perl_sv_catpvf(aTHX_ sv, "%d", (int)ARG(o)); /* Parenth number */
- else if (k == LOGICAL)
+ else if (k == GOSUB)
+ Perl_sv_catpvf(aTHX_ sv, "%d[%+d]", (int)ARG(o),(int)ARG2L(o)); /* Paren and offset */
+ else if (k == VERB) {
+ if (!o->flags)
+ Perl_sv_catpvf(aTHX_ sv, ":%"SVf,
+ (SV*)progi->data->data[ ARG( o ) ]);
+ } else if (k == LOGICAL)
Perl_sv_catpvf(aTHX_ sv, "[%d]", o->flags); /* 2: embedded, otherwise 1 */
else if (k == ANYOF) {
int i, rangestart = -1;
PERL_UNUSED_CONTEXT;
PERL_UNUSED_ARG(sv);
PERL_UNUSED_ARG(o);
+ PERL_UNUSED_ARG(prog);
#endif /* DEBUGGING */
}
return prog->check_substr ? prog->check_substr : prog->check_utf8;
}
+/*
+ pregfree - free a regexp
+
+ See regdupe below if you change anything here.
+*/
+
void
Perl_pregfree(pTHX_ struct regexp *r)
{
dVAR;
-#ifdef DEBUGGING
- SV * const dsv = PERL_DEBUG_PAD_ZERO(0);
-#endif
+ RXi_GET_DECL(r,ri);
GET_RE_DEBUG_FLAGS_DECL;
if (!r || (--r->refcnt > 0))
return;
- DEBUG_COMPILE_r(if (RX_DEBUG(r)){
- const char * const s = (r->reganch & ROPT_UTF8)
- ? pv_uni_display(dsv, (U8*)r->precomp, r->prelen, 60, UNI_DISPLAY_REGEX)
- : pv_display(dsv, r->precomp, r->prelen, 0, 60);
- const int len = SvCUR(dsv);
- if (!PL_colorset)
- reginitcolors();
- PerlIO_printf(Perl_debug_log,
- "%sFreeing REx:%s %s%*.*s%s%s\n",
- PL_colors[4],PL_colors[5],PL_colors[0],
- len, len, s,
- PL_colors[1],
- len > 60 ? "..." : "");
+ DEBUG_COMPILE_r({
+ if (!PL_colorset)
+ reginitcolors();
+ {
+ SV *dsv= sv_newmortal();
+ RE_PV_QUOTED_DECL(s, (r->extflags & RXf_UTF8),
+ dsv, r->precomp, r->prelen, 60);
+ PerlIO_printf(Perl_debug_log,"%sFreeing REx:%s %s\n",
+ PL_colors[4],PL_colors[5],s);
+ }
});
/* gcov results gave these as non-null 100% of the time, so there's no
optimisation in checking them before calling Safefree */
Safefree(r->precomp);
- Safefree(r->offsets); /* 20010421 MJD */
+ Safefree(ri->offsets); /* 20010421 MJD */
RX_MATCH_COPY_FREE(r);
#ifdef PERL_OLD_COPY_ON_WRITE
if (r->saved_copy)
SvREFCNT_dec(r->float_utf8);
Safefree(r->substrs);
}
- if (r->data) {
- int n = r->data->count;
+ if (r->paren_names)
+ SvREFCNT_dec(r->paren_names);
+ if (ri->data) {
+ int n = ri->data->count;
PAD* new_comppad = NULL;
PAD* old_comppad;
PADOFFSET refcnt;
while (--n >= 0) {
/* If you add a ->what type here, update the comment in regcomp.h */
- switch (r->data->what[n]) {
+ switch (ri->data->what[n]) {
case 's':
- SvREFCNT_dec((SV*)r->data->data[n]);
+ case 'S':
+ case 'u':
+ SvREFCNT_dec((SV*)ri->data->data[n]);
break;
case 'f':
- Safefree(r->data->data[n]);
+ Safefree(ri->data->data[n]);
break;
case 'p':
- new_comppad = (AV*)r->data->data[n];
+ new_comppad = (AV*)ri->data->data[n];
break;
case 'o':
if (new_comppad == NULL)
(SvTYPE(new_comppad) == SVt_PVAV) ? new_comppad : NULL
);
OP_REFCNT_LOCK;
- refcnt = OpREFCNT_dec((OP_4tree*)r->data->data[n]);
+ refcnt = OpREFCNT_dec((OP_4tree*)ri->data->data[n]);
OP_REFCNT_UNLOCK;
if (!refcnt)
- op_free((OP_4tree*)r->data->data[n]);
+ op_free((OP_4tree*)ri->data->data[n]);
PAD_RESTORE_LOCAL(old_comppad);
SvREFCNT_dec((SV*)new_comppad);
{ /* Aho Corasick add-on structure for a trie node.
Used in stclass optimization only */
U32 refcount;
- reg_ac_data *aho=(reg_ac_data*)r->data->data[n];
+ reg_ac_data *aho=(reg_ac_data*)ri->data->data[n];
OP_REFCNT_LOCK;
refcount = --aho->refcount;
OP_REFCNT_UNLOCK;
if ( !refcount ) {
- Safefree(aho->states);
- Safefree(aho->fail);
- aho->trie=NULL; /* not necessary to free this as it is
- handled by the 't' case */
- Safefree(r->data->data[n]); /* do this last!!!! */
- Safefree(r->regstclass);
+ PerlMemShared_free(aho->states);
+ PerlMemShared_free(aho->fail);
+ /* do this last!!!! */
+ PerlMemShared_free(ri->data->data[n]);
+ PerlMemShared_free(ri->regstclass);
}
}
break;
{
/* trie structure. */
U32 refcount;
- reg_trie_data *trie=(reg_trie_data*)r->data->data[n];
+ reg_trie_data *trie=(reg_trie_data*)ri->data->data[n];
OP_REFCNT_LOCK;
refcount = --trie->refcount;
OP_REFCNT_UNLOCK;
if ( !refcount ) {
- Safefree(trie->charmap);
- if (trie->widecharmap)
- SvREFCNT_dec((SV*)trie->widecharmap);
- Safefree(trie->states);
- Safefree(trie->trans);
+ PerlMemShared_free(trie->charmap);
+ PerlMemShared_free(trie->states);
+ PerlMemShared_free(trie->trans);
if (trie->bitmap)
- Safefree(trie->bitmap);
+ PerlMemShared_free(trie->bitmap);
if (trie->wordlen)
- Safefree(trie->wordlen);
-#ifdef DEBUGGING
- if (RX_DEBUG(r)) {
- if (trie->words)
- SvREFCNT_dec((SV*)trie->words);
- if (trie->revcharmap)
- SvREFCNT_dec((SV*)trie->revcharmap);
- }
-#endif
- Safefree(r->data->data[n]); /* do this last!!!! */
+ PerlMemShared_free(trie->wordlen);
+ if (trie->jump)
+ PerlMemShared_free(trie->jump);
+ if (trie->nextword)
+ PerlMemShared_free(trie->nextword);
+ /* do this last!!!! */
+ PerlMemShared_free(ri->data->data[n]);
}
}
break;
default:
- Perl_croak(aTHX_ "panic: regfree data code '%c'", r->data->what[n]);
+ Perl_croak(aTHX_ "panic: regfree data code '%c'", ri->data->what[n]);
}
}
- Safefree(r->data->what);
- Safefree(r->data);
+ Safefree(ri->data->what);
+ Safefree(ri->data);
}
Safefree(r->startp);
Safefree(r->endp);
+ if (ri->swap) {
+ Safefree(ri->swap->startp);
+ Safefree(ri->swap->endp);
+ Safefree(ri->swap);
+ }
+ Safefree(ri);
Safefree(r);
}
+#define sv_dup_inc(s,t) SvREFCNT_inc(sv_dup(s,t))
+#define av_dup_inc(s,t) (AV*)SvREFCNT_inc(sv_dup((SV*)s,t))
+#define hv_dup_inc(s,t) (HV*)SvREFCNT_inc(sv_dup((SV*)s,t))
+#define SAVEPVN(p,n) ((p) ? savepvn(p,n) : NULL)
+
+/*
+ regdupe - duplicate a regexp.
+
+ This routine is called by sv.c's re_dup and is expected to clone a
+ given regexp structure. It is a no-op when not under USE_ITHREADS.
+ (Originally this *was* re_dup() for change history see sv.c)
+
+ See pregfree() above if you change anything here.
+*/
+#if defined(USE_ITHREADS)
+regexp *
+Perl_regdupe(pTHX_ const regexp *r, CLONE_PARAMS *param)
+{
+ dVAR;
+ regexp *ret;
+ regexp_internal *reti;
+ int i, len, npar;
+ struct reg_substr_datum *s;
+ RXi_GET_DECL(r,ri);
+
+ if (!r)
+ return (REGEXP *)NULL;
+
+ if ((ret = (REGEXP *)ptr_table_fetch(PL_ptr_table, r)))
+ return ret;
+
+ len = ri->offsets[0];
+ npar = r->nparens+1;
+
+ Newxz(ret, 1, regexp);
+ Newxc(reti, sizeof(regexp_internal) + (len+1)*sizeof(regnode), char, regexp_internal);
+ RXi_SET(ret,reti);
+ Copy(ri->program, reti->program, len+1, regnode);
+
+ Newx(ret->startp, npar, I32);
+ Copy(r->startp, ret->startp, npar, I32);
+ Newx(ret->endp, npar, I32);
+ Copy(r->startp, ret->startp, npar, I32);
+ if(ri->swap) {
+ Newx(reti->swap, 1, regexp_paren_ofs);
+ /* no need to copy these */
+ Newx(reti->swap->startp, npar, I32);
+ Newx(reti->swap->endp, npar, I32);
+ } else {
+ reti->swap = NULL;
+ }
+
+ Newx(ret->substrs, 1, struct reg_substr_data);
+ for (s = ret->substrs->data, i = 0; i < 3; i++, s++) {
+ s->min_offset = r->substrs->data[i].min_offset;
+ s->max_offset = r->substrs->data[i].max_offset;
+ s->end_shift = r->substrs->data[i].end_shift;
+ s->substr = sv_dup_inc(r->substrs->data[i].substr, param);
+ s->utf8_substr = sv_dup_inc(r->substrs->data[i].utf8_substr, param);
+ }
+
+ reti->regstclass = NULL;
+ if (ri->data) {
+ struct reg_data *d;
+ const int count = ri->data->count;
+ int i;
+
+ Newxc(d, sizeof(struct reg_data) + count*sizeof(void *),
+ char, struct reg_data);
+ Newx(d->what, count, U8);
+
+ d->count = count;
+ for (i = 0; i < count; i++) {
+ d->what[i] = ri->data->what[i];
+ switch (d->what[i]) {
+ /* legal options are one of: sSfpontTu
+ see also regcomp.h and pregfree() */
+ case 's':
+ case 'S':
+ case 'p': /* actually an AV, but the dup function is identical. */
+ case 'u': /* actually an HV, but the dup function is identical. */
+ d->data[i] = sv_dup_inc((SV *)ri->data->data[i], param);
+ break;
+ case 'f':
+ /* This is cheating. */
+ Newx(d->data[i], 1, struct regnode_charclass_class);
+ StructCopy(ri->data->data[i], d->data[i],
+ struct regnode_charclass_class);
+ reti->regstclass = (regnode*)d->data[i];
+ break;
+ case 'o':
+ /* Compiled op trees are readonly and in shared memory,
+ and can thus be shared without duplication. */
+ OP_REFCNT_LOCK;
+ d->data[i] = (void*)OpREFCNT_inc((OP*)ri->data->data[i]);
+ OP_REFCNT_UNLOCK;
+ break;
+ case 'T':
+ /* Trie stclasses are readonly and can thus be shared
+ * without duplication. We free the stclass in pregfree
+ * when the corresponding reg_ac_data struct is freed.
+ */
+ reti->regstclass= ri->regstclass;
+ /* Fall through */
+ case 't':
+ OP_REFCNT_LOCK;
+ ((reg_trie_data*)ri->data->data[i])->refcount++;
+ OP_REFCNT_UNLOCK;
+ /* Fall through */
+ case 'n':
+ d->data[i] = ri->data->data[i];
+ break;
+ default:
+ Perl_croak(aTHX_ "panic: re_dup unknown data code '%c'", ri->data->what[i]);
+ }
+ }
+
+ reti->data = d;
+ }
+ else
+ reti->data = NULL;
+
+ Newx(reti->offsets, 2*len+1, U32);
+ Copy(ri->offsets, reti->offsets, 2*len+1, U32);
+
+ ret->precomp = SAVEPVN(r->precomp, r->prelen);
+ ret->refcnt = r->refcnt;
+ ret->minlen = r->minlen;
+ ret->minlenret = r->minlenret;
+ ret->prelen = r->prelen;
+ ret->nparens = r->nparens;
+ ret->lastparen = r->lastparen;
+ ret->lastcloseparen = r->lastcloseparen;
+ ret->intflags = r->intflags;
+ ret->extflags = r->extflags;
+
+ ret->sublen = r->sublen;
+
+ ret->engine = r->engine;
+
+ ret->paren_names = hv_dup_inc(r->paren_names, param);
+
+ if (RX_MATCH_COPIED(ret))
+ ret->subbeg = SAVEPVN(r->subbeg, r->sublen);
+ else
+ ret->subbeg = NULL;
+#ifdef PERL_OLD_COPY_ON_WRITE
+ ret->saved_copy = NULL;
+#endif
+
+ ptr_table_store(PL_ptr_table, r, ret);
+ return ret;
+}
+#endif
+
+/*
+ reg_stringify()
+
+ converts a regexp embedded in a MAGIC struct to its stringified form,
+ caching the converted form in the struct and returns the cached
+ string.
+
+ If lp is nonnull then it is used to return the length of the
+ resulting string
+
+ If flags is nonnull and the returned string contains UTF8 then
+ (flags & 1) will be true.
+
+ If haseval is nonnull then it is used to return whether the pattern
+ contains evals.
+
+ Normally called via macro:
+
+ CALLREG_STRINGIFY(mg,0,0);
+
+ And internally with
+
+ CALLREG_AS_STR(mg,lp,flags,haseval)
+
+ See sv_2pv_flags() in sv.c for an example of internal usage.
+
+ */
+
+char *
+Perl_reg_stringify(pTHX_ MAGIC *mg, STRLEN *lp, U32 *flags, I32 *haseval ) {
+ dVAR;
+ const regexp * const re = (regexp *)mg->mg_obj;
+ RXi_GET_DECL(re,ri);
+
+ if (!mg->mg_ptr) {
+ const char *fptr = "msix";
+ char reflags[6];
+ char ch;
+ int left = 0;
+ int right = 4;
+ bool need_newline = 0;
+ U16 reganch = (U16)((re->extflags & RXf_PMf_COMPILETIME) >> 12);
+
+ while((ch = *fptr++)) {
+ if(reganch & 1) {
+ reflags[left++] = ch;
+ }
+ else {
+ reflags[right--] = ch;
+ }
+ reganch >>= 1;
+ }
+ if(left != 4) {
+ reflags[left] = '-';
+ left = 5;
+ }
+
+ mg->mg_len = re->prelen + 4 + left;
+ /*
+ * If /x was used, we have to worry about a regex ending with a
+ * comment later being embedded within another regex. If so, we don't
+ * want this regex's "commentization" to leak out to the right part of
+ * the enclosing regex, we must cap it with a newline.
+ *
+ * So, if /x was used, we scan backwards from the end of the regex. If
+ * we find a '#' before we find a newline, we need to add a newline
+ * ourself. If we find a '\n' first (or if we don't find '#' or '\n'),
+ * we don't need to add anything. -jfriedl
+ */
+ if (PMf_EXTENDED & re->extflags) {
+ const char *endptr = re->precomp + re->prelen;
+ while (endptr >= re->precomp) {
+ const char c = *(endptr--);
+ if (c == '\n')
+ break; /* don't need another */
+ if (c == '#') {
+ /* we end while in a comment, so we need a newline */
+ mg->mg_len++; /* save space for it */
+ need_newline = 1; /* note to add it */
+ break;
+ }
+ }
+ }
+
+ Newx(mg->mg_ptr, mg->mg_len + 1 + left, char);
+ mg->mg_ptr[0] = '(';
+ mg->mg_ptr[1] = '?';
+ Copy(reflags, mg->mg_ptr+2, left, char);
+ *(mg->mg_ptr+left+2) = ':';
+ Copy(re->precomp, mg->mg_ptr+3+left, re->prelen, char);
+ if (need_newline)
+ mg->mg_ptr[mg->mg_len - 2] = '\n';
+ mg->mg_ptr[mg->mg_len - 1] = ')';
+ mg->mg_ptr[mg->mg_len] = 0;
+ }
+ if (haseval)
+ *haseval = ri->program[0].next_off;
+ if (flags)
+ *flags = ((re->extflags & RXf_UTF8) ? 1 : 0);
+
+ if (lp)
+ *lp = mg->mg_len;
+ return mg->mg_ptr;
+}
+
+
#ifndef PERL_IN_XSUB_RE
/*
- regnext - dig the "next" pointer out of a node
dVAR;
register I32 offset;
- if (p == &PL_regdummy)
+ if (!p)
return(NULL);
offset = (reg_off_by_arg[OP(p)] ? ARG(p) : NEXT_OFF(p));
Perl_sv_catpvf(aTHX_ sv, "%c", c);
}
+
#define CLEAR_OPTSTART \
if (optstart) STMT_START { \
- DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%d nodes)\n", node - optstart)); \
+ DEBUG_OPTIMISE_r(PerlIO_printf(Perl_debug_log, " (%"IVdf" nodes)\n", (IV)(node - optstart))); \
optstart=NULL; \
} STMT_END
-#define DUMPUNTIL(a,b,c,d,e,f) CLEAR_OPTSTART; node=dumpuntil(a,b,c,d,e,f);
+#define DUMPUNTIL(b,e) CLEAR_OPTSTART; node=dumpuntil(r,start,(b),(e),last,sv,indent+1,depth+1);
STATIC const regnode *
S_dumpuntil(pTHX_ const regexp *r, const regnode *start, const regnode *node,
- const regnode *last, SV* sv, I32 l)
+ const regnode *last, const regnode *plast,
+ SV* sv, I32 indent, U32 depth)
{
dVAR;
- register U8 op = EXACT; /* Arbitrary non-END op. */
+ register U8 op = PSEUDO; /* Arbitrary non-END op. */
register const regnode *next;
const regnode *optstart= NULL;
+ RXi_GET_DECL(r,ri);
GET_RE_DEBUG_FLAGS_DECL;
- while (op != END && (!last || node < last)) {
+#ifdef DEBUG_DUMPUNTIL
+ PerlIO_printf(Perl_debug_log, "--- %d : %d - %d - %d\n",indent,node-start,
+ last ? last-start : 0,plast ? plast-start : 0);
+#endif
+
+ if (plast && plast < last)
+ last= plast;
+
+ while (PL_regkind[op] != END && (!last || node < last)) {
/* While that wasn't END last time... */
NODE_ALIGN(node);
op = OP(node);
- if (op == CLOSE)
- l--;
+ if (op == CLOSE || op == WHILEM)
+ indent--;
next = regnext((regnode *)node);
/* Where, what. */
if (OP(node) == OPTIMIZED) {
- if (!optstart && (SvIV(re_debug_flags) & RE_DEBUG_COMPILE_OPTIMISE))
+ if (!optstart && RE_DEBUG_FLAG(RE_DEBUG_COMPILE_OPTIMISE))
optstart = node;
else
goto after_print;
regprop(r, sv, node);
PerlIO_printf(Perl_debug_log, "%4"IVdf":%*s%s", (IV)(node - start),
- (int)(2*l + 1), "", SvPVX_const(sv));
+ (int)(2*indent + 1), "", SvPVX_const(sv));
if (OP(node) != OPTIMIZED) {
if (next == NULL) /* Next ptr. */
PerlIO_printf(Perl_debug_log, "(0)");
+ else if (PL_regkind[(U8)op] == BRANCH && PL_regkind[OP(next)] != BRANCH )
+ PerlIO_printf(Perl_debug_log, "(FAIL)");
else
PerlIO_printf(Perl_debug_log, "(%"IVdf")", (IV)(next - start));
- (void)PerlIO_putc(Perl_debug_log, '\n');
+
+ /*if (PL_regkind[(U8)op] != TRIE)*/
+ (void)PerlIO_putc(Perl_debug_log, '\n');
}
after_print:
: next);
if (last && nnode > last)
nnode = last;
- DUMPUNTIL(r, start, NEXTOPER(NEXTOPER(node)), nnode, sv, l + 1);
+ DUMPUNTIL(NEXTOPER(NEXTOPER(node)), nnode);
}
}
else if (PL_regkind[(U8)op] == BRANCH) {
assert(next);
- DUMPUNTIL(r, start, NEXTOPER(node), next, sv, l + 1);
+ DUMPUNTIL(NEXTOPER(node), next);
}
else if ( PL_regkind[(U8)op] == TRIE ) {
+ const regnode *this_trie = node;
+ const char op = OP(node);
const I32 n = ARG(node);
- const reg_trie_data * const trie = (reg_trie_data*)r->data->data[n];
- const I32 arry_len = av_len(trie->words)+1;
+ const reg_ac_data * const ac = op>=AHOCORASICK ?
+ (reg_ac_data *)ri->data->data[n] :
+ NULL;
+ const reg_trie_data * const trie =
+ (reg_trie_data*)ri->data->data[op<AHOCORASICK ? n : ac->trie];
+#ifdef DEBUGGING
+ AV *const trie_words = (AV *) ri->data->data[n + TRIE_WORDS_OFFSET];
+#endif
+ const regnode *nextbranch= NULL;
I32 word_idx;
- PerlIO_printf(Perl_debug_log,
- "%*s[StS:%"UVuf" Wds:%d Cs:%d Uq:%d #Sts:%"IVdf" Mn:%d Mx:%d",
- (int)(2*(l+3)),
- "",
- trie->startstate,
- TRIE_WORDCOUNT(trie),
- (int)TRIE_CHARCOUNT(trie),
- trie->uniquecharcount,
- (IV)TRIE_LASTSTATE(trie)-1,
- (int)trie->minlen,
- (int)trie->maxlen
- );
- if (trie->bitmap) {
- int i;
- int rangestart= -1;
- sv_setpvn(sv, "", 0);
- for (i = 0; i <= 256; i++) {
- if (i < 256 && TRIE_BITMAP_TEST(trie,i)) {
- if (rangestart == -1)
- rangestart = i;
- } else if (rangestart != -1) {
- if (i <= rangestart + 3)
- for (; rangestart < i; rangestart++)
- put_byte(sv, rangestart);
- else {
- put_byte(sv, rangestart);
- sv_catpvs(sv, "-");
- put_byte(sv, i - 1);
- }
- rangestart = -1;
- }
- }
- PerlIO_printf(Perl_debug_log, " Stcls:%s]\n", SvPVX_const(sv));
- } else
- PerlIO_printf(Perl_debug_log, " No-Stcls]\n");
-
- for (word_idx=0; word_idx < arry_len; word_idx++) {
- SV ** const elem_ptr = av_fetch(trie->words,word_idx,0);
- if (elem_ptr) {
- PerlIO_printf(Perl_debug_log, "%*s<%s%s%s>\n",
- (int)(2*(l+4)), "",
- PL_colors[0],
- SvPV_nolen_const(*elem_ptr),
- PL_colors[1]
- );
+ sv_setpvn(sv, "", 0);
+ for (word_idx= 0; word_idx < (I32)trie->wordcount; word_idx++) {
+ SV ** const elem_ptr = av_fetch(trie_words,word_idx,0);
+
+ PerlIO_printf(Perl_debug_log, "%*s%s ",
+ (int)(2*(indent+3)), "",
+ elem_ptr ? pv_pretty(sv, SvPV_nolen_const(*elem_ptr), SvCUR(*elem_ptr), 60,
+ PL_colors[0], PL_colors[1],
+ (SvUTF8(*elem_ptr) ? PERL_PV_ESCAPE_UNI : 0) |
+ PERL_PV_PRETTY_ELIPSES |
+ PERL_PV_PRETTY_LTGT
+ )
+ : "???"
+ );
+ if (trie->jump) {
+ U16 dist= trie->jump[word_idx+1];
+ PerlIO_printf(Perl_debug_log, "(%"UVuf")\n",
+ (UV)((dist ? this_trie + dist : next) - start));
+ if (dist) {
+ if (!nextbranch)
+ nextbranch= this_trie + trie->jump[0];
+ DUMPUNTIL(this_trie + dist, nextbranch);
+ }
+ if (nextbranch && PL_regkind[OP(nextbranch)]==BRANCH)
+ nextbranch= regnext((regnode *)nextbranch);
+ } else {
+ PerlIO_printf(Perl_debug_log, "\n");
}
}
-
- node = NEXTOPER(node);
- node += regarglen[(U8)op];
-
+ if (last && next > last)
+ node= last;
+ else
+ node= next;
}
- else if ( op == CURLY) { /* "next" might be very big: optimizer */
- DUMPUNTIL(r, start, NEXTOPER(node) + EXTRA_STEP_2ARGS,
- NEXTOPER(node) + EXTRA_STEP_2ARGS + 1, sv, l + 1);
+ else if ( op == CURLY ) { /* "next" might be very big: optimizer */
+ DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS,
+ NEXTOPER(node) + EXTRA_STEP_2ARGS + 1);
}
else if (PL_regkind[(U8)op] == CURLY && op != CURLYX) {
assert(next);
- DUMPUNTIL(r, start, NEXTOPER(node) + EXTRA_STEP_2ARGS,
- next, sv, l + 1);
+ DUMPUNTIL(NEXTOPER(node) + EXTRA_STEP_2ARGS, next);
}
else if ( op == PLUS || op == STAR) {
- DUMPUNTIL(r, start, NEXTOPER(node), NEXTOPER(node) + 1, sv, l + 1);
+ DUMPUNTIL(NEXTOPER(node), NEXTOPER(node) + 1);
}
else if (op == ANYOF) {
/* arglen 1 + class block */
node += regarglen[(U8)op];
}
if (op == CURLYX || op == OPEN)
- l++;
- else if (op == WHILEM)
- l--;
+ indent++;
}
CLEAR_OPTSTART;
+#ifdef DEBUG_DUMPUNTIL
+ PerlIO_printf(Perl_debug_log, "--- %d\n", (int)indent);
+#endif
return node;
}