#endif
#ifdef PERL_UTF8_CACHE_ASSERT
-/* The cache element 0 is the Unicode offset;
- * the cache element 1 is the byte offset of the element 0;
- * the cache element 2 is the Unicode length of the substring;
- * the cache element 3 is the byte length of the substring;
- * The checking of the substring side would be good
- * but substr() has enough code paths to make my head spin;
- * if adding more checks watch out for the following tests:
+/* if adding more checks watch out for the following tests:
* t/op/index.t t/op/length.t t/op/pat.t t/op/substr.t
* lib/utf8.t lib/Unicode/Collate/t/index.t
* --jhi
*/
#define ASSERT_UTF8_CACHE(cache) \
- STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); } } STMT_END
+ STMT_START { if (cache) { assert((cache)[0] <= (cache)[1]); \
+ assert((cache)[2] <= (cache)[3]); \
+ assert((cache)[3] <= (cache)[1]);} \
+ } STMT_END
#else
#define ASSERT_UTF8_CACHE(cache) NOOP
#endif
# define SvARENA_CHAIN(sv) ((sv)->sv_u.svu_rv)
/* Whilst I'd love to do this, it seems that things like to check on
unreferenced scalars
-# define POSION_SV_HEAD(sv) Poison(sv, 1, struct STRUCT_SV)
+# define POSION_SV_HEAD(sv) PoisonNew(sv, 1, struct STRUCT_SV)
*/
-# define POSION_SV_HEAD(sv) Poison(&SvANY(sv), 1, void *), \
- Poison(&SvREFCNT(sv), 1, U32)
+# define POSION_SV_HEAD(sv) PoisonNew(&SvANY(sv), 1, void *), \
+ PoisonNew(&SvREFCNT(sv), 1, U32)
#else
# define SvARENA_CHAIN(sv) SvANY(sv)
# define POSION_SV_HEAD(sv)
void ** const r3wt = &PL_body_roots[sv_type]; \
LOCK_SV_MUTEX; \
xpv = *((void **)(r3wt)) \
- ? *((void **)(r3wt)) : S_more_bodies(aTHX_ sv_type); \
+ ? *((void **)(r3wt)) : more_bodies(sv_type); \
*(r3wt) = *(void**)(xpv); \
UNLOCK_SV_MUTEX; \
} STMT_END
int length = old_type_details->copy;
if (new_type_details->offset > old_type_details->offset) {
- int difference
+ const int difference
= new_type_details->offset - old_type_details->offset;
offset += difference;
length -= difference;
case SVt_PVGV:
if (dtype <= SVt_PVGV) {
- S_glob_assign_glob(aTHX_ dstr, sstr, dtype);
+ glob_assign_glob(dstr, sstr, dtype);
return;
}
/*FALLTHROUGH*/
if ((int)SvTYPE(sstr) != stype) {
stype = SvTYPE(sstr);
if (stype == SVt_PVGV && dtype <= SVt_PVGV) {
- S_glob_assign_glob(aTHX_ dstr, sstr, dtype);
+ glob_assign_glob(dstr, sstr, dtype);
return;
}
}
GvMULTI_on(dstr);
return;
}
- S_glob_assign_glob(aTHX_ dstr, sstr, dtype);
+ glob_assign_glob(dstr, sstr, dtype);
return;
}
if (dtype >= SVt_PV) {
if (dtype == SVt_PVGV) {
- S_glob_assign_ref(aTHX_ dstr, sstr);
+ glob_assign_ref(dstr, sstr);
return;
}
if (SvPVX_const(dstr)) {
case PERL_MAGIC_qr:
vtable = &PL_vtbl_regexp;
break;
+ case PERL_MAGIC_hints:
+ /* As this vtable is all NULL, we can reuse it. */
case PERL_MAGIC_sig:
vtable = &PL_vtbl_sig;
break;
case PERL_MAGIC_backref:
vtable = &PL_vtbl_backref;
break;
+ case PERL_MAGIC_hintselem:
+ vtable = &PL_vtbl_hintselem;
+ break;
case PERL_MAGIC_ext:
/* Reserved for use by extensions not perl internals. */
/* Useful for attaching extension internal data to perl vars. */
/*
* The length is cached in PERL_UTF8_magic, in the mg_len field. Also the
- * mg_ptr is used, by sv_pos_u2b(), see the comments of S_utf8_mg_pos_init().
- * (Note that the mg_len is not the length of the mg_ptr field.)
+ * mg_ptr is used, by sv_pos_u2b() and sv_pos_b2u() - see the comments below.
+ * (Note that the mg_len is not the length of the mg_ptr field.
+ * This allows the cache to store the character length of the string without
+ * needing to malloc() extra storage to attach to the mg_ptr.)
*
*/
}
}
-/*
-=for apidoc sv_pos_u2b
-
-Converts the value pointed to by offsetp from a count of UTF-8 chars from
-the start of the string, to a count of the equivalent number of bytes; if
-lenp is non-zero, it does the same to lenp, but this time starting from
-the offset, rather than from the start of the string. Handles magic and
-type coercion.
-
-=cut
-*/
-
-/*
- * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
- * PERL_UTF8_magic of the sv to store the mapping between UTF-8 and
- * byte offsets. See also the comments of S_utf8_mg_pos().
- *
- */
-
+/* Walk forwards to find the byte corresponding to the passed in UTF-8
+ offset. */
static STRLEN
S_sv_pos_u2b_forwards(pTHX_ const U8 *const start, const U8 *const send,
STRLEN uoffset)
{
const U8 *s = start;
+ PERL_UNUSED_CONTEXT;
+
while (s < send && uoffset--)
s += UTF8SKIP(s);
if (s > send) {
return s - start;
}
-
+/* Given the length of the string in both bytes and UTF-8 characters, decide
+ whether to walk forwards or backwards to find the byte corresponding to
+ the passed in UTF-8 offset. */
static STRLEN
S_sv_pos_u2b_midway(pTHX_ const U8 *const start, const U8 *send,
STRLEN uoffset, STRLEN uend)
return send - start;
}
+/* For the string representation of the given scalar, find the byte
+ corresponding to the passed in UTF-8 offset. uoffset0 and boffset0
+ give another position in the string, *before* the sought offset, which
+ (which is always true, as 0, 0 is a valid pair of positions), which should
+ help reduce the amount of linear searching.
+ If *mgp is non-NULL, it should point to the UTF-8 cache magic, which
+ will be used to reduce the amount of linear searching. The cache will be
+ created if necessary, and the found value offered to it for update. */
static STRLEN
S_sv_pos_u2b_cached(pTHX_ SV *sv, MAGIC **mgp, const U8 *const start,
const U8 *const send, STRLEN uoffset,
STRLEN uoffset0, STRLEN boffset0) {
- STRLEN boffset;
+ STRLEN boffset = 0; /* Actually always set, but let's keep gcc happy. */
bool found = FALSE;
assert (uoffset >= uoffset0);
/* An exact match. */
return cache[1];
}
+ if (cache[2] == uoffset) {
+ /* An exact match. */
+ return cache[3];
+ }
if (cache[0] < uoffset) {
/* The cache already knows part of the way. */
+ S_sv_pos_u2b_forwards(aTHX_ start + boffset0,
send, uoffset - uoffset0);
}
- } else {
+ }
+ else if (cache[2] < uoffset) {
+ /* We're between the two cache entries. */
+ if (cache[2] > uoffset0) {
+ /* and the cache knows more than the passed in pair */
+ uoffset0 = cache[2];
+ boffset0 = cache[3];
+ }
+
boffset = boffset0
+ S_sv_pos_u2b_midway(aTHX_ start + boffset0,
start + cache[1],
uoffset - uoffset0,
cache[0] - uoffset0);
+ } else {
+ boffset = boffset0
+ + S_sv_pos_u2b_midway(aTHX_ start + boffset0,
+ start + cache[3],
+ uoffset - uoffset0,
+ cache[2] - uoffset0);
}
found = TRUE;
}
boffset = real_boffset;
}
- S_utf8_mg_pos_cache_update(aTHX_ sv, mgp, boffset, uoffset);
+ S_utf8_mg_pos_cache_update(aTHX_ sv, mgp, boffset, uoffset, send - start);
return boffset;
}
+
+/*
+=for apidoc sv_pos_u2b
+
+Converts the value pointed to by offsetp from a count of UTF-8 chars from
+the start of the string, to a count of the equivalent number of bytes; if
+lenp is non-zero, it does the same to lenp, but this time starting from
+the offset, rather than from the start of the string. Handles magic and
+type coercion.
+
+=cut
+*/
+
+/*
+ * sv_pos_u2b() uses, like sv_pos_b2u(), the mg_ptr of the potential
+ * PERL_UTF8_magic of the sv to store the mapping between UTF-8 and
+ * byte offsets. See also the comments of S_utf8_mg_pos_cache_update().
+ *
+ */
+
void
Perl_sv_pos_u2b(pTHX_ register SV *sv, I32* offsetp, I32* lenp)
{
return;
}
-/*
-=for apidoc sv_pos_b2u
-
-Converts the value pointed to by offsetp from a count of bytes from the
-start of the string, to a count of the equivalent number of UTF-8 chars.
-Handles magic and type coercion.
-
-=cut
+/* Create and update the UTF8 magic offset cache, with the proffered utf8/
+ byte length pairing. The (byte) length of the total SV is passed in too,
+ as blen, because for some (more esoteric) SVs, the call to SvPV_const()
+ may not have updated SvCUR, so we can't rely on reading it directly.
+
+ The proffered utf8/byte length pairing isn't used if the cache already has
+ two pairs, and swapping either for the proffered pair would increase the
+ RMS of the intervals between known byte offsets.
+
+ The cache itself consists of 4 STRLEN values
+ 0: larger UTF-8 offset
+ 1: corresponding byte offset
+ 2: smaller UTF-8 offset
+ 3: corresponding byte offset
+
+ Unused cache pairs have the value 0, 0.
+ Keeping the cache "backwards" means that the invariant of
+ cache[0] >= cache[2] is maintained even with empty slots, which means that
+ the code that uses it doesn't need to worry if only 1 entry has actually
+ been set to non-zero. It also makes the "position beyond the end of the
+ cache" logic much simpler, as the first slot is always the one to start
+ from.
*/
-
-/*
- * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
- * PERL_UTF8_magic of the sv to store the mapping between UTF-8 and
- * byte offsets.
- *
- */
-
static void
-S_utf8_mg_pos_cache_update(pTHX_ SV *sv, MAGIC **mgp, STRLEN byte, STRLEN utf8)
+S_utf8_mg_pos_cache_update(pTHX_ SV *sv, MAGIC **mgp, STRLEN byte, STRLEN utf8,
+ STRLEN blen)
{
STRLEN *cache;
if (SvREADONLY(sv))
" real %"UVf" for %"SVf, (UV) utf8, (UV) realutf8, sv);
}
}
- cache[0] = utf8;
- cache[1] = byte;
+
+ /* Cache is held with the later position first, to simplify the code
+ that deals with unbounded ends. */
+
+ ASSERT_UTF8_CACHE(cache);
+ if (cache[1] == 0) {
+ /* Cache is totally empty */
+ cache[0] = utf8;
+ cache[1] = byte;
+ } else if (cache[3] == 0) {
+ if (byte > cache[1]) {
+ /* New one is larger, so goes first. */
+ cache[2] = cache[0];
+ cache[3] = cache[1];
+ cache[0] = utf8;
+ cache[1] = byte;
+ } else {
+ cache[2] = utf8;
+ cache[3] = byte;
+ }
+ } else {
+#define THREEWAY_SQUARE(a,b,c,d) \
+ ((float)((d) - (c))) * ((float)((d) - (c))) \
+ + ((float)((c) - (b))) * ((float)((c) - (b))) \
+ + ((float)((b) - (a))) * ((float)((b) - (a)))
+
+ /* Cache has 2 slots in use, and we know three potential pairs.
+ Keep the two that give the lowest RMS distance. Do the
+ calcualation in bytes simply because we always know the byte
+ length. squareroot has the same ordering as the positive value,
+ so don't bother with the actual square root. */
+ const float existing = THREEWAY_SQUARE(0, cache[3], cache[1], blen);
+ if (byte > cache[1]) {
+ /* New position is after the existing pair of pairs. */
+ const float keep_earlier
+ = THREEWAY_SQUARE(0, cache[3], byte, blen);
+ const float keep_later
+ = THREEWAY_SQUARE(0, cache[1], byte, blen);
+
+ if (keep_later < keep_earlier) {
+ if (keep_later < existing) {
+ cache[2] = cache[0];
+ cache[3] = cache[1];
+ cache[0] = utf8;
+ cache[1] = byte;
+ }
+ }
+ else {
+ if (keep_earlier < existing) {
+ cache[0] = utf8;
+ cache[1] = byte;
+ }
+ }
+ }
+ else if (byte > cache[3]) {
+ /* New position is between the existing pair of pairs. */
+ const float keep_earlier
+ = THREEWAY_SQUARE(0, cache[3], byte, blen);
+ const float keep_later
+ = THREEWAY_SQUARE(0, byte, cache[1], blen);
+
+ if (keep_later < keep_earlier) {
+ if (keep_later < existing) {
+ cache[2] = utf8;
+ cache[3] = byte;
+ }
+ }
+ else {
+ if (keep_earlier < existing) {
+ cache[0] = utf8;
+ cache[1] = byte;
+ }
+ }
+ }
+ else {
+ /* New position is before the existing pair of pairs. */
+ const float keep_earlier
+ = THREEWAY_SQUARE(0, byte, cache[3], blen);
+ const float keep_later
+ = THREEWAY_SQUARE(0, byte, cache[1], blen);
+
+ if (keep_later < keep_earlier) {
+ if (keep_later < existing) {
+ cache[2] = utf8;
+ cache[3] = byte;
+ }
+ }
+ else {
+ if (keep_earlier < existing) {
+ cache[0] = cache[2];
+ cache[1] = cache[3];
+ cache[2] = utf8;
+ cache[3] = byte;
+ }
+ }
+ }
+ }
ASSERT_UTF8_CACHE(cache);
- /* Drop the stale "length" cache */
}
/* If we don't know the character offset of the end of a region, our only
return endu;
}
+/*
+=for apidoc sv_pos_b2u
+
+Converts the value pointed to by offsetp from a count of bytes from the
+start of the string, to a count of the equivalent number of UTF-8 chars.
+Handles magic and type coercion.
+
+=cut
+*/
+
+/*
+ * sv_pos_b2u() uses, like sv_pos_u2b(), the mg_ptr of the potential
+ * PERL_UTF8_magic of the sv to store the mapping between UTF-8 and
+ * byte offsets.
+ *
+ */
void
Perl_sv_pos_b2u(pTHX_ register SV* sv, I32* offsetp)
{
const U8* s;
const STRLEN byte = *offsetp;
- STRLEN len;
+ STRLEN len = 0; /* Actually always set, but let's keep gcc happy. */
+ STRLEN blen;
MAGIC* mg = NULL;
const U8* send;
+ bool found = FALSE;
if (!sv)
return;
- s = (const U8*)SvPV_const(sv, len);
+ s = (const U8*)SvPV_const(sv, blen);
- if (len < byte)
+ if (blen < byte)
Perl_croak(aTHX_ "panic: sv_pos_b2u: bad byte offset");
send = s + byte;
if (SvMAGICAL(sv) && !SvREADONLY(sv) && PL_utf8cache
&& (mg = mg_find(sv, PERL_MAGIC_utf8))) {
if (mg->mg_ptr) {
- STRLEN *cache = (STRLEN *) mg->mg_ptr;
+ STRLEN * const cache = (STRLEN *) mg->mg_ptr;
if (cache[1] == byte) {
/* An exact match. */
*offsetp = cache[0];
return;
}
+ if (cache[3] == byte) {
+ /* An exact match. */
+ *offsetp = cache[2];
+ return;
+ }
if (cache[1] < byte) {
/* We already know part of the way. */
/* Actually, we know the end too. */
len = cache[0]
+ S_sv_pos_b2u_midway(aTHX_ s + cache[1], send,
- s + len, mg->mg_len - cache[0]);
+ s + blen, mg->mg_len - cache[0]);
} else {
len = cache[0]
+ S_sv_pos_b2u_forwards(aTHX_ s + cache[1], send);
}
}
- else { /* cache[1] > byte */
- len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[1],
- cache[0]);
+ else if (cache[3] < byte) {
+ /* We're between the two cached pairs, so we do the calculation
+ offset by the byte/utf-8 positions for the earlier pair,
+ then add the utf-8 characters from the string start to
+ there. */
+ len = S_sv_pos_b2u_midway(aTHX_ s + cache[3], send,
+ s + cache[1], cache[0] - cache[2])
+ + cache[2];
}
- ASSERT_UTF8_CACHE(cache);
- if (PL_utf8cache < 0) {
- const STRLEN reallen = S_sv_pos_b2u_forwards(aTHX_ s, send);
-
- if (len != reallen) {
- /* Need to turn the assertions off otherwise we may recurse
- infinitely while printing error messages. */
- SAVEI8(PL_utf8cache);
- PL_utf8cache = 0;
- Perl_croak(aTHX_ "panic: sv_pos_b2u cache %"UVf
- " real %"UVf" for %"SVf,
- (UV) len, (UV) reallen, sv);
- }
+ else { /* cache[3] > byte */
+ len = S_sv_pos_b2u_midway(aTHX_ s, send, s + cache[3],
+ cache[2]);
+
}
+ ASSERT_UTF8_CACHE(cache);
+ found = TRUE;
} else if (mg->mg_len != -1) {
- len = S_sv_pos_b2u_midway(aTHX_ s, send, s + len, mg->mg_len);
- } else {
- len = S_sv_pos_b2u_forwards(aTHX_ s, send);
+ len = S_sv_pos_b2u_midway(aTHX_ s, send, s + blen, mg->mg_len);
+ found = TRUE;
}
}
- else {
- len = S_sv_pos_b2u_forwards(aTHX_ s, send);
+ if (!found || PL_utf8cache < 0) {
+ const STRLEN real_len = S_sv_pos_b2u_forwards(aTHX_ s, send);
+
+ if (found && PL_utf8cache < 0) {
+ if (len != real_len) {
+ /* Need to turn the assertions off otherwise we may recurse
+ infinitely while printing error messages. */
+ SAVEI8(PL_utf8cache);
+ PL_utf8cache = 0;
+ Perl_croak(aTHX_ "panic: sv_pos_b2u cache %"UVf
+ " real %"UVf" for %"SVf,
+ (UV) len, (UV) real_len, sv);
+ }
+ }
+ len = real_len;
}
*offsetp = len;
- S_utf8_mg_pos_cache_update(aTHX_ sv, &mg, byte, len);
+ S_utf8_mg_pos_cache_update(aTHX_ sv, &mg, byte, len, blen);
}
/*
#if defined(USE_ITHREADS)
+/* XXX Remove this so it doesn't have to go thru the macro and return for nothing */
#ifndef GpREFCNT_inc
# define GpREFCNT_inc(gp) ((gp) ? (++(gp)->gp_refcnt, (gp)) : (GP*)NULL)
#endif
if (tblent->oldval == sv)
return tblent;
}
- return 0;
+ return NULL;
}
void *
{
PTR_TBL_ENT_t const *const tblent = ptr_table_find(tbl, sv);
PERL_UNUSED_CONTEXT;
- return tblent ? tblent->newval : (void *) 0;
+ return tblent ? tblent->newval : NULL;
}
/* add a new entry to a pointer-mapping table */
case SAVEt_HINTS:
i = POPINT(ss,ix);
TOPINT(nss,ix) = i;
+ ptr = POPPTR(ss,ix);
+ TOPPTR(nss,ix) = Perl_refcounted_he_dup(aTHX_ ptr, param);
+ if (i & HINT_LOCALIZE_HH) {
+ hv = (HV*)POPPTR(ss,ix);
+ TOPPTR(nss,ix) = hv_dup_inc(hv, param);
+ }
break;
case SAVEt_COMPPAD:
av = (AV*)POPPTR(ss,ix);
PERL_SET_THX(my_perl);
# ifdef DEBUGGING
- Poison(my_perl, 1, PerlInterpreter);
+ PoisonNew(my_perl, 1, PerlInterpreter);
PL_op = NULL;
PL_curcop = NULL;
PL_markstack = 0;
PERL_SET_THX(my_perl);
# ifdef DEBUGGING
- Poison(my_perl, 1, PerlInterpreter);
+ PoisonNew(my_perl, 1, PerlInterpreter);
PL_op = NULL;
PL_curcop = NULL;
PL_markstack = 0;
PL_compiling.cop_warnings = sv_dup_inc(PL_compiling.cop_warnings, param);
if (!specialCopIO(PL_compiling.cop_io))
PL_compiling.cop_io = sv_dup_inc(PL_compiling.cop_io, param);
+ PL_compiling.cop_hints
+ = Perl_refcounted_he_dup(aTHX_ PL_compiling.cop_hints, proto_perl);
PL_curcop = (COP*)any_dup(proto_perl->Tcurcop, proto_perl);
/* pseudo environmental stuff */
PL_formfeed = sv_dup(proto_perl->Iformfeed, param);
PL_maxsysfd = proto_perl->Imaxsysfd;
- PL_multiline = proto_perl->Imultiline;
PL_statusvalue = proto_perl->Istatusvalue;
#ifdef VMS
PL_statusvalue_vms = proto_perl->Istatusvalue_vms;
PL_regnarrate = 0;
PL_regprogram = (regnode*)NULL;
PL_regindent = 0;
- PL_regcc = (CURCUR*)NULL;
PL_reg_call_cc = (struct re_cc_state*)NULL;
PL_reg_re = (regexp*)NULL;
PL_reg_ganch = NULL;
/* attempt to find a match within the aggregate */
if (hash) {
- keysv = S_find_hash_subscript(aTHX_ (HV*)sv, uninit_sv);
+ keysv = find_hash_subscript((HV*)sv, uninit_sv);
if (keysv)
subscript_type = FUV_SUBSCRIPT_HASH;
}
/* index is an expression;
* attempt to find a match within the aggregate */
if (obase->op_type == OP_HELEM) {
- SV * const keysv = S_find_hash_subscript(aTHX_ (HV*)sv, uninit_sv);
+ SV * const keysv = find_hash_subscript((HV*)sv, uninit_sv);
if (keysv)
return varname(gv, '%', o->op_targ,
keysv, 0, FUV_SUBSCRIPT_HASH);
}
else {
- const I32 index = S_find_array_subscript(aTHX_ (AV*)sv, uninit_sv);
+ const I32 index = find_array_subscript((AV*)sv, uninit_sv);
if (index >= 0)
return varname(gv, '@', o->op_targ,
NULL, index, FUV_SUBSCRIPT_ARRAY);