static inline SV *hv_he_store_or_croak(HV *hv, SV *key, SV *val) {
dTHX;
HE *he = hv_fetch_ent(hv, key, TRUE, 0U);
if (!he) {
SvREFCNT_dec(val);
croak("Can't store value");
}
SV *sv = HeVAL(he);
SvSetMagicSV(sv, val);
return sv;
}
SV* THX_MopMcV_get_method(pTHX_ SV* metaclass, SV* name) {
HV* stash = (HV*) SvRV(metaclass);
HE* method_gv_he = hv_fetch_ent(stash, name, 0, 0);
if (method_gv_he != NULL) {
GV* method_gv = (GV*) HeVAL(method_gv_he);
CV* method = GvCV(method_gv);
if (method != NULL && GvSTASH(CvGV(method)) == stash) {
return newRV_inc((SV*) method);
}
}
return NULL;
}
bool THX_MopMcV_has_method(pTHX_ SV* metaclass, SV* name) {
HV* stash = (HV*) SvRV(metaclass);
HE* method_gv_he = hv_fetch_ent(stash, name, 0, 0);
if (method_gv_he != NULL) {
GV* method_gv = (GV*) HeVAL(method_gv_he);
CV* method = GvCV(method_gv);
if (method != NULL && GvSTASH(CvGV(method)) == stash) {
return TRUE;
}
}
return FALSE;
}
void THX_MopMcV_add_method(pTHX_ SV* metaclass, SV* name, SV* code) {
GV* method_gv;
HV* stash = (HV*) SvRV(metaclass);
SV* method = newMopMmV(code, 0);
HE* method_gv_he = hv_fetch_ent(stash, name, 0, 0);
if (method_gv_he != NULL) {
GV* method_gv = (GV*) HeVAL(method_gv_he);
} else {
method_gv = (GV*) newSV(0);
gv_init_sv(method_gv, stash, name, 0);
(void)hv_store_ent(stash, name, (SV*) method_gv, 0);
}
MopMmV_assign_to_stash(method, method_gv, stash);
}
SV* THX_MopMcV_get_attribute(pTHX_ SV* metaclass, SV* name) {
HE* attribute;
SV* attributes = MopOV_get_at_slot(metaclass, CLASS_ATTRIBUTE_SLOT);
if (attributes == NULL) {
attributes = newRV_noinc((SV*) newHV());
MopOV_set_at_slot(metaclass, CLASS_ATTRIBUTE_SLOT, attributes);
// NOTE:
// I know I am not going to
// have the value since I
// only just now created the
// HV to store it.
return NULL;
}
if (SvTYPE(attributes) != SVt_RV && SvTYPE(SvRV(attributes)) != SVt_PVHV) {
croak("attributes is not a HASH ref, this is wrong");
}
attribute = hv_fetch_ent((HV*) SvRV(attributes), name, 0, 0);
return attribute == NULL ? NULL : HeVAL(attribute);
}
/*
=for apidoc mro_isa_changed_in
Takes the necessary steps (cache invalidations, mostly)
when the @ISA of the given package has changed. Invoked
by the C<setisa> magic, should not need to invoke directly.
=cut
*/
void
Perl_mro_isa_changed_in(pTHX_ HV* stash)
{
dVAR;
HV* isarev;
AV* linear_mro;
HE* iter;
SV** svp;
I32 items;
bool is_universal;
struct mro_meta * meta;
const char * const stashname = HvNAME_get(stash);
const STRLEN stashname_len = HvNAMELEN_get(stash);
PERL_ARGS_ASSERT_MRO_ISA_CHANGED_IN;
if(!stashname)
Perl_croak(aTHX_ "Can't call mro_isa_changed_in() on anonymous symbol table");
/* wipe out the cached linearizations for this stash */
meta = HvMROMETA(stash);
if (meta->mro_linear_dfs) {
SvREFCNT_dec(MUTABLE_SV(meta->mro_linear_dfs));
meta->mro_linear_dfs = NULL;
/* This is just acting as a shortcut pointer. */
meta->mro_linear_c3 = NULL;
} else if (meta->mro_linear_c3) {
/* Only the current MRO is stored, so this owns the data. */
SvREFCNT_dec(MUTABLE_SV(meta->mro_linear_c3));
meta->mro_linear_c3 = NULL;
}
if (meta->isa) {
SvREFCNT_dec(meta->isa);
meta->isa = NULL;
}
/* Inc the package generation, since our @ISA changed */
meta->pkg_gen++;
/* Wipe the global method cache if this package
is UNIVERSAL or one of its parents */
svp = hv_fetch(PL_isarev, stashname, stashname_len, 0);
isarev = svp ? MUTABLE_HV(*svp) : NULL;
if((stashname_len == 9 && strEQ(stashname, "UNIVERSAL"))
|| (isarev && hv_exists(isarev, "UNIVERSAL", 9))) {
PL_sub_generation++;
is_universal = TRUE;
}
else { /* Wipe the local method cache otherwise */
meta->cache_gen++;
is_universal = FALSE;
}
/* wipe next::method cache too */
if(meta->mro_nextmethod) hv_clear(meta->mro_nextmethod);
/* Iterate the isarev (classes that are our children),
wiping out their linearization, method and isa caches */
if(isarev) {
hv_iterinit(isarev);
while((iter = hv_iternext(isarev))) {
I32 len;
const char* const revkey = hv_iterkey(iter, &len);
HV* revstash = gv_stashpvn(revkey, len, 0);
struct mro_meta* revmeta;
if(!revstash) continue;
revmeta = HvMROMETA(revstash);
if (revmeta->mro_linear_dfs) {
SvREFCNT_dec(MUTABLE_SV(revmeta->mro_linear_dfs));
revmeta->mro_linear_dfs = NULL;
/* This is just acting as a shortcut pointer. */
revmeta->mro_linear_c3 = NULL;
} else if (revmeta->mro_linear_c3) {
/* Only the current MRO is stored, so this owns the data. */
SvREFCNT_dec(MUTABLE_SV(revmeta->mro_linear_c3));
revmeta->mro_linear_c3 = NULL;
}
if(!is_universal)
revmeta->cache_gen++;
if(revmeta->mro_nextmethod)
hv_clear(revmeta->mro_nextmethod);
if (revmeta->isa) {
SvREFCNT_dec(revmeta->isa);
revmeta->isa = NULL;
}
}
}
/* Now iterate our MRO (parents), and do a few things:
1) instantiate with the "fake" flag if they don't exist
2) flag them as universal if we are universal
3) Add everything from our isarev to their isarev
*/
/* We're starting at the 2nd element, skipping ourselves here */
linear_mro = mro_get_linear_isa(stash);
svp = AvARRAY(linear_mro) + 1;
items = AvFILLp(linear_mro);
while (items--) {
SV* const sv = *svp++;
HV* mroisarev;
HE *he = hv_fetch_ent(PL_isarev, sv, TRUE, 0);
/* That fetch should not fail. But if it had to create a new SV for
us, then we can detect it, because it will not be the correct type.
Probably faster and cleaner for us to free that scalar [very little
code actually executed to free it] and create a new HV than to
copy&paste [SIN!] the code from newHV() to allow us to upgrade the
new SV from SVt_NULL. */
mroisarev = MUTABLE_HV(HeVAL(he));
if(SvTYPE(mroisarev) != SVt_PVHV) {
SvREFCNT_dec(mroisarev);
mroisarev = newHV();
HeVAL(he) = MUTABLE_SV(mroisarev);
}
/* This hash only ever contains PL_sv_yes. Storing it over itself is
almost as cheap as calling hv_exists, so on aggregate we expect to
save time by not making two calls to the common HV code for the
case where it doesn't exist. */
(void)hv_store(mroisarev, stashname, stashname_len, &PL_sv_yes, 0);
if(isarev) {
hv_iterinit(isarev);
while((iter = hv_iternext(isarev))) {
I32 revkeylen;
char* const revkey = hv_iterkey(iter, &revkeylen);
(void)hv_store(mroisarev, revkey, revkeylen, &PL_sv_yes, 0);
}
}
}
}
/*
=for apidoc mro_get_linear_isa_dfs
Returns the Depth-First Search linearization of @ISA
the given stash. The return value is a read-only AV*.
C<level> should be 0 (it is used internally in this
function's recursion).
You are responsible for C<SvREFCNT_inc()> on the
return value if you plan to store it anywhere
semi-permanently (otherwise it might be deleted
out from under you the next time the cache is
invalidated).
=cut
*/
static AV*
S_mro_get_linear_isa_dfs(pTHX_ HV *stash, U32 level)
{
AV* retval;
GV** gvp;
GV* gv;
AV* av;
const HEK* stashhek;
struct mro_meta* meta;
SV *our_name;
HV *stored;
PERL_ARGS_ASSERT_MRO_GET_LINEAR_ISA_DFS;
assert(HvAUX(stash));
stashhek = HvNAME_HEK(stash);
if (!stashhek)
Perl_croak(aTHX_ "Can't linearize anonymous symbol table");
if (level > 100)
Perl_croak(aTHX_ "Recursive inheritance detected in package '%s'",
HEK_KEY(stashhek));
meta = HvMROMETA(stash);
/* return cache if valid */
if((retval = MUTABLE_AV(MRO_GET_PRIVATE_DATA(meta, &dfs_alg)))) {
return retval;
}
/* not in cache, make a new one */
retval = MUTABLE_AV(sv_2mortal(MUTABLE_SV(newAV())));
/* We use this later in this function, but don't need a reference to it
beyond the end of this function, so reference count is fine. */
our_name = newSVhek(stashhek);
av_push(retval, our_name); /* add ourselves at the top */
/* fetch our @ISA */
gvp = (GV**)hv_fetchs(stash, "ISA", FALSE);
av = (gvp && (gv = *gvp) && isGV_with_GP(gv)) ? GvAV(gv) : NULL;
/* "stored" is used to keep track of all of the classnames we have added to
the MRO so far, so we can do a quick exists check and avoid adding
duplicate classnames to the MRO as we go.
It's then retained to be re-used as a fast lookup for ->isa(), by adding
our own name and "UNIVERSAL" to it. */
stored = MUTABLE_HV(sv_2mortal(MUTABLE_SV(newHV())));
if(av && AvFILLp(av) >= 0) {
SV **svp = AvARRAY(av);
I32 items = AvFILLp(av) + 1;
/* foreach(@ISA) */
while (items--) {
SV* const sv = *svp++;
HV* const basestash = gv_stashsv(sv, 0);
SV *const *subrv_p;
I32 subrv_items;
if (!basestash) {
/* if no stash exists for this @ISA member,
simply add it to the MRO and move on */
subrv_p = &sv;
subrv_items = 1;
}
else {
/* otherwise, recurse into ourselves for the MRO
of this @ISA member, and append their MRO to ours.
The recursive call could throw an exception, which
has memory management implications here, hence the use of
the mortal. */
const AV *const subrv
= mro_get_linear_isa_dfs(basestash, level + 1);
subrv_p = AvARRAY(subrv);
subrv_items = AvFILLp(subrv) + 1;
}
while(subrv_items--) {
SV *const subsv = *subrv_p++;
/* LVALUE fetch will create a new undefined SV if necessary
*/
HE *const he = hv_fetch_ent(stored, subsv, 1, 0);
assert(he);
if(HeVAL(he) != &PL_sv_undef) {
/* It was newly created. Steal it for our new SV, and
replace it in the hash with the "real" thing. */
SV *const val = HeVAL(he);
HEK *const key = HeKEY_hek(he);
HeVAL(he) = &PL_sv_undef;
/* Save copying by making a shared hash key scalar. We
inline this here rather than calling Perl_newSVpvn_share
because we already have the scalar, and we already have
the hash key. */
assert(SvTYPE(val) == SVt_NULL);
sv_upgrade(val, SVt_PV);
SvPV_set(val, HEK_KEY(share_hek_hek(key)));
SvCUR_set(val, HEK_LEN(key));
SvREADONLY_on(val);
SvFAKE_on(val);
SvPOK_on(val);
if (HEK_UTF8(key))
SvUTF8_on(val);
av_push(retval, val);
}
}
}
}
(void) hv_store_ent(stored, our_name, &PL_sv_undef, 0);
(void) hv_store(stored, "UNIVERSAL", 9, &PL_sv_undef, 0);
SvREFCNT_inc_simple_void_NN(stored);
SvTEMP_off(stored);
SvREADONLY_on(stored);
meta->isa = stored;
/* now that we're past the exception dangers, grab our own reference to
the AV we're about to use for the result. The reference owned by the
mortals' stack will be released soon, so everything will balance. */
SvREFCNT_inc_simple_void_NN(retval);
SvTEMP_off(retval);
/* we don't want anyone modifying the cache entry but us,
and we do so by replacing it completely */
SvREADONLY_on(retval);
return MUTABLE_AV(Perl_mro_set_private_data(aTHX_ meta, &dfs_alg,
MUTABLE_SV(retval)));
}
STATIC void
S_mro_gather_and_rename(pTHX_ HV * const stashes, HV * const seen_stashes,
HV *stash, HV *oldstash, SV *namesv)
{
XPVHV* xhv;
HE *entry;
I32 riter = -1;
I32 items = 0;
const bool stash_had_name = stash && HvENAME(stash);
bool fetched_isarev = FALSE;
HV *seen = NULL;
HV *isarev = NULL;
SV **svp = NULL;
PERL_ARGS_ASSERT_MRO_GATHER_AND_RENAME;
/* We use the seen_stashes hash to keep track of which packages have
been encountered so far. This must be separate from the main list of
stashes, as we need to distinguish between stashes being assigned
and stashes being replaced/deleted. (A nested stash can be on both
sides of an assignment. We cannot simply skip iterating through a
stash on the right if we have seen it on the left, as it will not
get its ename assigned to it.)
To avoid allocating extra SVs, instead of a bitfield we can make
bizarre use of immortals:
&PL_sv_undef: seen on the left (oldstash)
&PL_sv_no : seen on the right (stash)
&PL_sv_yes : seen on both sides
*/
if(oldstash) {
/* Add to the big list. */
struct mro_meta * meta;
HE * const entry
= (HE *)
hv_common(
seen_stashes, NULL, (const char *)&oldstash, sizeof(HV *), 0,
HV_FETCH_LVALUE|HV_FETCH_EMPTY_HE, NULL, 0
);
if(HeVAL(entry) == &PL_sv_undef || HeVAL(entry) == &PL_sv_yes) {
oldstash = NULL;
goto check_stash;
}
HeVAL(entry)
= HeVAL(entry) == &PL_sv_no ? &PL_sv_yes : &PL_sv_undef;
meta = HvMROMETA(oldstash);
(void)
hv_store(
stashes, (const char *)&oldstash, sizeof(HV *),
meta->isa
? SvREFCNT_inc_simple_NN((SV *)meta->isa)
: &PL_sv_yes,
0
);
CLEAR_LINEAR(meta);
/* Update the effective name. */
if(HvENAME_get(oldstash)) {
const HEK * const enamehek = HvENAME_HEK(oldstash);
if(SvTYPE(namesv) == SVt_PVAV) {
items = AvFILLp((AV *)namesv) + 1;
svp = AvARRAY((AV *)namesv);
}
else {
items = 1;
svp = &namesv;
}
while (items--) {
const U32 name_utf8 = SvUTF8(*svp);
STRLEN len;
const char *name = SvPVx_const(*svp, len);
if(PL_stashcache) {
DEBUG_o(Perl_deb(aTHX_ "mro_gather_and_rename clearing PL_stashcache for '%"SVf"'\n",
SVfARG(*svp)));
(void)hv_delete(PL_stashcache, name, name_utf8 ? -(I32)len : (I32)len, G_DISCARD);
}
++svp;
hv_ename_delete(oldstash, name, len, name_utf8);
if (!fetched_isarev) {
/* If the name deletion caused a name change, then we
* are not going to call mro_isa_changed_in with this
* name (and not at all if it has become anonymous) so
* we need to delete old isarev entries here, both
* those in the superclasses and this class's own list
* of subclasses. We simply delete the latter from
* PL_isarev, since we still need it. hv_delete morti-
* fies it for us, so sv_2mortal is not necessary. */
if(HvENAME_HEK(oldstash) != enamehek) {
if(meta->isa && HvARRAY(meta->isa))
mro_clean_isarev(meta->isa, name, len, 0, 0,
name_utf8 ? HVhek_UTF8 : 0);
isarev = (HV *)hv_delete(PL_isarev, name,
name_utf8 ? -(I32)len : (I32)len, 0);
fetched_isarev=TRUE;
}
}
}
}
}
check_stash:
if(stash) {
if(SvTYPE(namesv) == SVt_PVAV) {
items = AvFILLp((AV *)namesv) + 1;
svp = AvARRAY((AV *)namesv);
}
else {
items = 1;
svp = &namesv;
}
while (items--) {
const U32 name_utf8 = SvUTF8(*svp);
STRLEN len;
const char *name = SvPVx_const(*svp++, len);
hv_ename_add(stash, name, len, name_utf8);
}
/* Add it to the big list if it needs
* mro_isa_changed_in called on it. That happens if it was
* detached from the symbol table (so it had no HvENAME) before
* being assigned to the spot named by the 'name' variable, because
* its cached isa linearisation is now stale (the effective name
* having changed), and subclasses will then use that cache when
* mro_package_moved calls mro_isa_changed_in. (See
* [perl #77358].)
*
* If it did have a name, then its previous name is still
* used in isa caches, and there is no need for
* mro_package_moved to call mro_isa_changed_in.
*/
entry
= (HE *)
hv_common(
seen_stashes, NULL, (const char *)&stash, sizeof(HV *), 0,
HV_FETCH_LVALUE|HV_FETCH_EMPTY_HE, NULL, 0
);
if(HeVAL(entry) == &PL_sv_yes || HeVAL(entry) == &PL_sv_no)
stash = NULL;
else {
HeVAL(entry)
= HeVAL(entry) == &PL_sv_undef ? &PL_sv_yes : &PL_sv_no;
if(!stash_had_name)
{
struct mro_meta * const meta = HvMROMETA(stash);
(void)
hv_store(
stashes, (const char *)&stash, sizeof(HV *),
meta->isa
? SvREFCNT_inc_simple_NN((SV *)meta->isa)
: &PL_sv_yes,
0
);
CLEAR_LINEAR(meta);
}
}
}
if(!stash && !oldstash)
/* Both stashes have been encountered already. */
return;
/* Add all the subclasses to the big list. */
if(!fetched_isarev) {
/* If oldstash is not null, then we can use its HvENAME to look up
the isarev hash, since all its subclasses will be listed there.
It will always have an HvENAME. It the HvENAME was removed
above, then fetch_isarev will be true, and this code will not be
reached.
If oldstash is null, then this is an empty spot with no stash in
it, so subclasses could be listed in isarev hashes belonging to
any of the names, so we have to check all of them.
*/
assert(!oldstash || HvENAME(oldstash));
if (oldstash) {
/* Extra variable to avoid a compiler warning */
const HEK * const hvename = HvENAME_HEK(oldstash);
fetched_isarev = TRUE;
svp = hv_fetchhek(PL_isarev, hvename, 0);
if (svp) isarev = MUTABLE_HV(*svp);
}
else if(SvTYPE(namesv) == SVt_PVAV) {
items = AvFILLp((AV *)namesv) + 1;
svp = AvARRAY((AV *)namesv);
}
else {
items = 1;
svp = &namesv;
}
}
if(
isarev || !fetched_isarev
) {
while (fetched_isarev || items--) {
HE *iter;
if (!fetched_isarev) {
HE * const he = hv_fetch_ent(PL_isarev, *svp++, 0, 0);
if (!he || !(isarev = MUTABLE_HV(HeVAL(he)))) continue;
}
hv_iterinit(isarev);
while((iter = hv_iternext(isarev))) {
HV* revstash = gv_stashsv(hv_iterkeysv(iter), 0);
struct mro_meta * meta;
if(!revstash) continue;
meta = HvMROMETA(revstash);
(void)
hv_store(
stashes, (const char *)&revstash, sizeof(HV *),
meta->isa
? SvREFCNT_inc_simple_NN((SV *)meta->isa)
: &PL_sv_yes,
0
);
CLEAR_LINEAR(meta);
}
if (fetched_isarev) break;
}
}
/* This is partly based on code in hv_iternext_flags. We are not call-
ing that here, as we want to avoid resetting the hash iterator. */
/* Skip the entire loop if the hash is empty. */
if(oldstash && HvUSEDKEYS(oldstash)) {
xhv = (XPVHV*)SvANY(oldstash);
seen = (HV *) sv_2mortal((SV *)newHV());
/* Iterate through entries in the oldstash, adding them to the
list, meanwhile doing the equivalent of $seen{$key} = 1.
*/
while (++riter <= (I32)xhv->xhv_max) {
entry = (HvARRAY(oldstash))[riter];
/* Iterate through the entries in this list */
for(; entry; entry = HeNEXT(entry)) {
const char* key;
I32 len;
/* If this entry is not a glob, ignore it.
Try the next. */
if (!isGV(HeVAL(entry))) continue;
key = hv_iterkey(entry, &len);
if ((len > 1 && key[len-2] == ':' && key[len-1] == ':')
|| (len == 1 && key[0] == ':')) {
HV * const oldsubstash = GvHV(HeVAL(entry));
SV ** const stashentry
= stash ? hv_fetch(stash, key, HeUTF8(entry) ? -(I32)len : (I32)len, 0) : NULL;
HV *substash = NULL;
/* Avoid main::main::main::... */
if(oldsubstash == oldstash) continue;
if(
(
stashentry && *stashentry && isGV(*stashentry)
&& (substash = GvHV(*stashentry))
)
|| (oldsubstash && HvENAME_get(oldsubstash))
)
{
/* Add :: and the key (minus the trailing ::)
to each name. */
SV *subname;
if(SvTYPE(namesv) == SVt_PVAV) {
SV *aname;
items = AvFILLp((AV *)namesv) + 1;
svp = AvARRAY((AV *)namesv);
subname = sv_2mortal((SV *)newAV());
while (items--) {
aname = newSVsv(*svp++);
if (len == 1)
sv_catpvs(aname, ":");
else {
sv_catpvs(aname, "::");
sv_catpvn_flags(
aname, key, len-2,
HeUTF8(entry)
? SV_CATUTF8 : SV_CATBYTES
);
}
av_push((AV *)subname, aname);
}
}
else {
subname = sv_2mortal(newSVsv(namesv));
if (len == 1) sv_catpvs(subname, ":");
else {
sv_catpvs(subname, "::");
sv_catpvn_flags(
subname, key, len-2,
HeUTF8(entry) ? SV_CATUTF8 : SV_CATBYTES
);
}
}
mro_gather_and_rename(
stashes, seen_stashes,
substash, oldsubstash, subname
);
}
(void)hv_store(seen, key, HeUTF8(entry) ? -(I32)len : (I32)len, &PL_sv_yes, 0);
}
}
}
}
/* Skip the entire loop if the hash is empty. */
if (stash && HvUSEDKEYS(stash)) {
xhv = (XPVHV*)SvANY(stash);
riter = -1;
/* Iterate through the new stash, skipping $seen{$key} items,
calling mro_gather_and_rename(stashes,seen,entry,NULL, ...). */
while (++riter <= (I32)xhv->xhv_max) {
entry = (HvARRAY(stash))[riter];
/* Iterate through the entries in this list */
for(; entry; entry = HeNEXT(entry)) {
const char* key;
I32 len;
/* If this entry is not a glob, ignore it.
Try the next. */
if (!isGV(HeVAL(entry))) continue;
key = hv_iterkey(entry, &len);
if ((len > 1 && key[len-2] == ':' && key[len-1] == ':')
|| (len == 1 && key[0] == ':')) {
HV *substash;
/* If this entry was seen when we iterated through the
oldstash, skip it. */
if(seen && hv_exists(seen, key, HeUTF8(entry) ? -(I32)len : (I32)len)) continue;
/* We get here only if this stash has no corresponding
entry in the stash being replaced. */
substash = GvHV(HeVAL(entry));
if(substash) {
SV *subname;
/* Avoid checking main::main::main::... */
if(substash == stash) continue;
/* Add :: and the key (minus the trailing ::)
to each name. */
if(SvTYPE(namesv) == SVt_PVAV) {
SV *aname;
items = AvFILLp((AV *)namesv) + 1;
svp = AvARRAY((AV *)namesv);
subname = sv_2mortal((SV *)newAV());
while (items--) {
aname = newSVsv(*svp++);
if (len == 1)
sv_catpvs(aname, ":");
else {
sv_catpvs(aname, "::");
sv_catpvn_flags(
aname, key, len-2,
HeUTF8(entry)
? SV_CATUTF8 : SV_CATBYTES
);
}
av_push((AV *)subname, aname);
}
}
else {
subname = sv_2mortal(newSVsv(namesv));
if (len == 1) sv_catpvs(subname, ":");
else {
sv_catpvs(subname, "::");
sv_catpvn_flags(
subname, key, len-2,
HeUTF8(entry) ? SV_CATUTF8 : SV_CATBYTES
);
}
}
mro_gather_and_rename(
stashes, seen_stashes,
substash, NULL, subname
);
}
}
}
}
}
}
void
Perl_mro_isa_changed_in(pTHX_ HV* stash)
{
HV* isarev;
AV* linear_mro;
HE* iter;
SV** svp;
I32 items;
bool is_universal;
struct mro_meta * meta;
HV *isa = NULL;
const HEK * const stashhek = HvENAME_HEK(stash);
const char * const stashname = HvENAME_get(stash);
const STRLEN stashname_len = HvENAMELEN_get(stash);
PERL_ARGS_ASSERT_MRO_ISA_CHANGED_IN;
if(!stashname)
Perl_croak(aTHX_ "Can't call mro_isa_changed_in() on anonymous symbol table");
/* wipe out the cached linearizations for this stash */
meta = HvMROMETA(stash);
CLEAR_LINEAR(meta);
if (meta->isa) {
/* Steal it for our own purposes. */
isa = (HV *)sv_2mortal((SV *)meta->isa);
meta->isa = NULL;
}
/* Inc the package generation, since our @ISA changed */
meta->pkg_gen++;
/* Wipe the global method cache if this package
is UNIVERSAL or one of its parents */
svp = hv_fetchhek(PL_isarev, stashhek, 0);
isarev = svp ? MUTABLE_HV(*svp) : NULL;
if((stashname_len == 9 && strEQ(stashname, "UNIVERSAL"))
|| (isarev && hv_exists(isarev, "UNIVERSAL", 9))) {
PL_sub_generation++;
is_universal = TRUE;
}
else { /* Wipe the local method cache otherwise */
meta->cache_gen++;
is_universal = FALSE;
}
/* wipe next::method cache too */
if(meta->mro_nextmethod) hv_clear(meta->mro_nextmethod);
/* Changes to @ISA might turn overloading on */
HvAMAGIC_on(stash);
/* pessimise derefs for now. Will get recalculated by Gv_AMupdate() */
HvAUX(stash)->xhv_aux_flags &= ~HvAUXf_NO_DEREF;
/* DESTROY can be cached in SvSTASH. */
if (!SvOBJECT(stash)) SvSTASH(stash) = NULL;
/* Iterate the isarev (classes that are our children),
wiping out their linearization, method and isa caches
and upating PL_isarev. */
if(isarev) {
HV *isa_hashes = NULL;
/* We have to iterate through isarev twice to avoid a chicken and
* egg problem: if A inherits from B and both are in isarev, A might
* be processed before B and use B's previous linearisation.
*/
/* First iteration: Wipe everything, but stash away the isa hashes
* since we still need them for updating PL_isarev.
*/
if(hv_iterinit(isarev)) {
/* Only create the hash if we need it; i.e., if isarev has
any elements. */
isa_hashes = (HV *)sv_2mortal((SV *)newHV());
}
while((iter = hv_iternext(isarev))) {
HV* revstash = gv_stashsv(hv_iterkeysv(iter), 0);
struct mro_meta* revmeta;
if(!revstash) continue;
revmeta = HvMROMETA(revstash);
CLEAR_LINEAR(revmeta);
if(!is_universal)
revmeta->cache_gen++;
if(revmeta->mro_nextmethod)
hv_clear(revmeta->mro_nextmethod);
if (!SvOBJECT(revstash)) SvSTASH(revstash) = NULL;
(void)
hv_store(
isa_hashes, (const char*)&revstash, sizeof(HV *),
revmeta->isa ? (SV *)revmeta->isa : &PL_sv_undef, 0
);
revmeta->isa = NULL;
}
/* Second pass: Update PL_isarev. We can just use isa_hashes to
* avoid another round of stash lookups. */
/* isarev might be deleted from PL_isarev during this loop, so hang
* on to it. */
SvREFCNT_inc_simple_void_NN(sv_2mortal((SV *)isarev));
if(isa_hashes) {
hv_iterinit(isa_hashes);
while((iter = hv_iternext(isa_hashes))) {
HV* const revstash = *(HV **)HEK_KEY(HeKEY_hek(iter));
HV * const isa = (HV *)HeVAL(iter);
const HEK *namehek;
/* We're starting at the 2nd element, skipping revstash */
linear_mro = mro_get_linear_isa(revstash);
svp = AvARRAY(linear_mro) + 1;
items = AvFILLp(linear_mro);
namehek = HvENAME_HEK(revstash);
if (!namehek) namehek = HvNAME_HEK(revstash);
while (items--) {
SV* const sv = *svp++;
HV* mroisarev;
HE *he = hv_fetch_ent(PL_isarev, sv, TRUE, 0);
/* That fetch should not fail. But if it had to create
a new SV for us, then will need to upgrade it to an
HV (which sv_upgrade() can now do for us). */
mroisarev = MUTABLE_HV(HeVAL(he));
SvUPGRADE(MUTABLE_SV(mroisarev), SVt_PVHV);
/* This hash only ever contains PL_sv_yes. Storing it
over itself is almost as cheap as calling hv_exists,
so on aggregate we expect to save time by not making
two calls to the common HV code for the case where
it doesn't exist. */
(void)
hv_storehek(mroisarev, namehek, &PL_sv_yes);
}
if ((SV *)isa != &PL_sv_undef) {
assert(namehek);
mro_clean_isarev(
isa, HEK_KEY(namehek), HEK_LEN(namehek),
HvMROMETA(revstash)->isa, HEK_HASH(namehek),
HEK_UTF8(namehek)
);
}
}
}
}
/* Now iterate our MRO (parents), adding ourselves and everything from
our isarev to their isarev.
*/
/* We're starting at the 2nd element, skipping ourselves here */
linear_mro = mro_get_linear_isa(stash);
svp = AvARRAY(linear_mro) + 1;
items = AvFILLp(linear_mro);
while (items--) {
SV* const sv = *svp++;
HV* mroisarev;
HE *he = hv_fetch_ent(PL_isarev, sv, TRUE, 0);
/* That fetch should not fail. But if it had to create a new SV for
us, then will need to upgrade it to an HV (which sv_upgrade() can
now do for us. */
mroisarev = MUTABLE_HV(HeVAL(he));
SvUPGRADE(MUTABLE_SV(mroisarev), SVt_PVHV);
/* This hash only ever contains PL_sv_yes. Storing it over itself is
almost as cheap as calling hv_exists, so on aggregate we expect to
save time by not making two calls to the common HV code for the
case where it doesn't exist. */
(void)hv_storehek(mroisarev, stashhek, &PL_sv_yes);
}
/* Delete our name from our former parents' isarevs. */
if(isa && HvARRAY(isa))
mro_clean_isarev(isa, stashname, stashname_len, meta->isa,
HEK_HASH(stashhek), HEK_UTF8(stashhek));
}
/*
=for apidoc mro_get_linear_isa_dfs
Returns the Depth-First Search linearization of C<@ISA>
the given stash. The return value is a read-only AV*.
C<level> should be 0 (it is used internally in this
function's recursion).
You are responsible for C<SvREFCNT_inc()> on the
return value if you plan to store it anywhere
semi-permanently (otherwise it might be deleted
out from under you the next time the cache is
invalidated).
=cut
*/
static AV*
S_mro_get_linear_isa_dfs(pTHX_ HV *stash, U32 level)
{
AV* retval;
GV** gvp;
GV* gv;
AV* av;
const HEK* stashhek;
struct mro_meta* meta;
SV *our_name;
HV *stored = NULL;
PERL_ARGS_ASSERT_MRO_GET_LINEAR_ISA_DFS;
assert(HvAUX(stash));
stashhek
= HvAUX(stash)->xhv_name_u.xhvnameu_name && HvENAME_HEK_NN(stash)
? HvENAME_HEK_NN(stash)
: HvNAME_HEK(stash);
if (!stashhek)
Perl_croak(aTHX_ "Can't linearize anonymous symbol table");
if (level > 100)
Perl_croak(aTHX_
"Recursive inheritance detected in package '%"HEKf"'",
HEKfARG(stashhek));
meta = HvMROMETA(stash);
/* return cache if valid */
if((retval = MUTABLE_AV(MRO_GET_PRIVATE_DATA(meta, &dfs_alg)))) {
return retval;
}
/* not in cache, make a new one */
retval = MUTABLE_AV(sv_2mortal(MUTABLE_SV(newAV())));
/* We use this later in this function, but don't need a reference to it
beyond the end of this function, so reference count is fine. */
our_name = newSVhek(stashhek);
av_push(retval, our_name); /* add ourselves at the top */
/* fetch our @ISA */
gvp = (GV**)hv_fetchs(stash, "ISA", FALSE);
av = (gvp && (gv = *gvp) && isGV_with_GP(gv)) ? GvAV(gv) : NULL;
/* "stored" is used to keep track of all of the classnames we have added to
the MRO so far, so we can do a quick exists check and avoid adding
duplicate classnames to the MRO as we go.
It's then retained to be re-used as a fast lookup for ->isa(), by adding
our own name and "UNIVERSAL" to it. */
if(av && AvFILLp(av) >= 0) {
SV **svp = AvARRAY(av);
I32 items = AvFILLp(av) + 1;
/* foreach(@ISA) */
while (items--) {
SV* const sv = *svp ? *svp : &PL_sv_undef;
HV* const basestash = gv_stashsv(sv, 0);
SV *const *subrv_p;
I32 subrv_items;
svp++;
if (!basestash) {
/* if no stash exists for this @ISA member,
simply add it to the MRO and move on */
subrv_p = &sv;
subrv_items = 1;
}
else {
/* otherwise, recurse into ourselves for the MRO
of this @ISA member, and append their MRO to ours.
The recursive call could throw an exception, which
has memory management implications here, hence the use of
the mortal. */
const AV *const subrv
= mro_get_linear_isa_dfs(basestash, level + 1);
subrv_p = AvARRAY(subrv);
subrv_items = AvFILLp(subrv) + 1;
}
if (stored) {
while(subrv_items--) {
SV *const subsv = *subrv_p++;
/* LVALUE fetch will create a new undefined SV if necessary
*/
HE *const he = hv_fetch_ent(stored, subsv, 1, 0);
assert(he);
if(HeVAL(he) != &PL_sv_undef) {
/* It was newly created. Steal it for our new SV, and
replace it in the hash with the "real" thing. */
SV *const val = HeVAL(he);
HEK *const key = HeKEY_hek(he);
HeVAL(he) = &PL_sv_undef;
sv_sethek(val, key);
av_push(retval, val);
}
}
} else {
/* We are the first (or only) parent. We can short cut the
complexity above, because our @ISA is simply us prepended
to our parent's @ISA, and our ->isa cache is simply our
parent's, with our name added. */
/* newSVsv() is slow. This code is only faster if we can avoid
it by ensuring that SVs in the arrays are shared hash key
scalar SVs, because we can "copy" them very efficiently.
Although to be fair, we can't *ensure* this, as a reference
to the internal array is returned by mro::get_linear_isa(),
so we'll have to be defensive just in case someone faffed
with it. */
if (basestash) {
SV **svp;
stored = MUTABLE_HV(sv_2mortal((SV*)newHVhv(HvMROMETA(basestash)->isa)));
av_extend(retval, subrv_items);
AvFILLp(retval) = subrv_items;
svp = AvARRAY(retval);
while(subrv_items--) {
SV *const val = *subrv_p++;
*++svp = SvIsCOW_shared_hash(val)
? newSVhek(SvSHARED_HEK_FROM_PV(SvPVX(val)))
: newSVsv(val);
}
} else {
/* They have no stash. So create ourselves an ->isa cache
as if we'd copied it from what theirs should be. */
stored = MUTABLE_HV(sv_2mortal(MUTABLE_SV(newHV())));
(void) hv_store(stored, "UNIVERSAL", 9, &PL_sv_undef, 0);
av_push(retval,
newSVhek(HeKEY_hek(hv_store_ent(stored, sv,
&PL_sv_undef, 0))));
}
}
}
} else {
/* We have no parents. */
stored = MUTABLE_HV(sv_2mortal(MUTABLE_SV(newHV())));
(void) hv_store(stored, "UNIVERSAL", 9, &PL_sv_undef, 0);
}
(void) hv_store_ent(stored, our_name, &PL_sv_undef, 0);
SvREFCNT_inc_simple_void_NN(stored);
SvTEMP_off(stored);
SvREADONLY_on(stored);
meta->isa = stored;
/* now that we're past the exception dangers, grab our own reference to
the AV we're about to use for the result. The reference owned by the
mortals' stack will be released soon, so everything will balance. */
SvREFCNT_inc_simple_void_NN(retval);
SvTEMP_off(retval);
/* we don't want anyone modifying the cache entry but us,
and we do so by replacing it completely */
SvREADONLY_on(retval);
return MUTABLE_AV(Perl_mro_set_private_data(aTHX_ meta, &dfs_alg,
MUTABLE_SV(retval)));
}
static AV*
S_mro_get_linear_isa_c3(pTHX_ HV* stash, U32 level)
{
AV* retval;
GV** gvp;
GV* gv;
AV* isa;
const HEK* stashhek;
struct mro_meta* meta;
PERL_ARGS_ASSERT_MRO_GET_LINEAR_ISA_C3;
assert(HvAUX(stash));
stashhek = HvNAME_HEK(stash);
if (!stashhek)
Perl_croak(aTHX_ "Can't linearize anonymous symbol table");
if (level > 100)
Perl_croak(aTHX_ "Recursive inheritance detected in package '%s'",
HEK_KEY(stashhek));
meta = HvMROMETA(stash);
/* return cache if valid */
if((retval = meta->mro_linear_c3)) {
return retval;
}
/* not in cache, make a new one */
gvp = (GV**)hv_fetchs(stash, "ISA", FALSE);
isa = (gvp && (gv = *gvp) && isGV_with_GP(gv)) ? GvAV(gv) : NULL;
if ( isa && ! SvAVOK(isa) ) {
Perl_croak(aTHX_ "@ISA is not an array but %s", Ddesc((SV*)isa));
}
/* For a better idea how the rest of this works, see the much clearer
pure perl version in Algorithm::C3 0.01:
http://search.cpan.org/src/STEVAN/Algorithm-C3-0.01/lib/Algorithm/C3.pm
(later versions go about it differently than this code for speed reasons)
*/
if(isa && AvFILLp(isa) >= 0) {
SV** seqs_ptr;
I32 seqs_items;
HV* const tails = MUTABLE_HV(sv_2mortal(MUTABLE_SV(newHV())));
AV *const seqs = MUTABLE_AV(sv_2mortal(MUTABLE_SV(newAV())));
I32* heads;
/* This builds @seqs, which is an array of arrays.
The members of @seqs are the MROs of
the members of @ISA, followed by @ISA itself.
*/
I32 items = AvFILLp(isa) + 1;
SV** isa_ptr = AvARRAY(isa);
while(items--) {
SV* const isa_item = *isa_ptr++;
if ( ! SvPVOK(isa_item) ) {
Perl_croak(aTHX_ "@ISA element which is not an plain value");
}
{
HV* const isa_item_stash = gv_stashsv(isa_item, 0);
if(!isa_item_stash) {
/* if no stash, make a temporary fake MRO
containing just itself */
AV* const isa_lin = newAV();
av_push(isa_lin, newSVsv(isa_item));
av_push(seqs, (SV*)isa_lin);
}
else {
/* recursion */
AV* const isa_lin = mro_get_linear_isa_c3(isa_item_stash, level + 1);
av_push(seqs, SvREFCNT_inc_NN((SV*)isa_lin));
}
}
}
av_push(seqs, SvREFCNT_inc_NN((SV*)isa));
/* This builds "heads", which as an array of integer array
indices, one per seq, which point at the virtual "head"
of the seq (initially zero) */
Newxz(heads, AvFILLp(seqs)+1, I32);
/* This builds %tails, which has one key for every class
mentioned in the tail of any sequence in @seqs (tail meaning
everything after the first class, the "head"). The value
is how many times this key appears in the tails of @seqs.
*/
seqs_ptr = AvARRAY(seqs);
seqs_items = AvFILLp(seqs) + 1;
while(seqs_items--) {
AV *const seq = MUTABLE_AV(*seqs_ptr++);
I32 seq_items = AvFILLp(seq);
if(seq_items > 0) {
SV** seq_ptr = AvARRAY(seq) + 1;
while(seq_items--) {
SV* const seqitem = *seq_ptr++;
/* LVALUE fetch will create a new undefined SV if necessary
*/
HE* const he = hv_fetch_ent(tails, seqitem, 1, 0);
if(he) {
SV* const val = HeVAL(he);
/* This will increment undef to 1, which is what we
want for a newly created entry. */
sv_inc(val);
}
}
}
}
/* Initialize retval to build the return value in */
retval = newAV();
av_push(retval, newSVhek(stashhek)); /* us first */
/* This loop won't terminate until we either finish building
the MRO, or get an exception. */
while(1) {
SV* cand = NULL;
SV* winner = NULL;
int s;
/* "foreach $seq (@seqs)" */
SV** const avptr = AvARRAY(seqs);
for(s = 0; s <= AvFILLp(seqs); s++) {
SV** svp;
AV * const seq = MUTABLE_AV(avptr[s]);
SV* seqhead;
if(!seq) continue; /* skip empty seqs */
svp = av_fetch(seq, heads[s], 0);
seqhead = *svp; /* seqhead = head of this seq */
if(!winner) {
HE* tail_entry;
SV* val;
/* if we haven't found a winner for this round yet,
and this seqhead is not in tails (or the count
for it in tails has dropped to zero), then this
seqhead is our new winner, and is added to the
final MRO immediately */
cand = seqhead;
if((tail_entry = hv_fetch_ent(tails, cand, 0, 0))
&& (val = HeVAL(tail_entry))
&& (SvIV(val) > 0))
continue;
winner = newSVsv(cand);
av_push(retval, winner);
/* note however that even when we find a winner,
we continue looping over @seqs to do housekeeping */
}
if(!sv_cmp(seqhead, winner)) {
/* Once we have a winner (including the iteration
where we first found him), inc the head ptr
for any seq which had the winner as a head,
NULL out any seq which is now empty,
and adjust tails for consistency */
const int new_head = ++heads[s];
if(new_head > AvFILLp(seq)) {
SvREFCNT_dec(avptr[s]);
avptr[s] = NULL;
}
else {
HE* tail_entry;
SV* val;
/* Because we know this new seqhead used to be
a tail, we can assume it is in tails and has
a positive value, which we need to dec */
svp = av_fetch(seq, new_head, 0);
seqhead = *svp;
tail_entry = hv_fetch_ent(tails, seqhead, 0, 0);
val = HeVAL(tail_entry);
sv_dec(val);
}
}
}
/* if we found no candidates, we are done building the MRO.
!cand means no seqs have any entries left to check */
if(!cand) {
Safefree(heads);
break;
}
/* If we had candidates, but nobody won, then the @ISA
hierarchy is not C3-incompatible */
if(!winner) {
SV *errmsg;
I32 i;
errmsg = newSVpvf(aTHX_ "Inconsistent hierarchy during C3 merge of class '%s':\n\t"
"current merge results [\n", HEK_KEY(stashhek));
for (i = 0; i <= av_len(retval); i++) {
SV **elem = av_fetch(retval, i, 0);
sv_catpvf(aTHX_ errmsg, "\t\t%"SVf",\n", SVfARG(*elem));
}
sv_catpvf(aTHX_ errmsg, "\t]\n\tmerging failed on '%"SVf"'", SVfARG(cand));
/* we have to do some cleanup before we croak */
AvREFCNT_dec(retval);
Safefree(heads);
croak(aTHX_ "%"SVf, SVfARG(errmsg));
}
}
}
else { /* @ISA was undefined or empty */
/* build a retval containing only ourselves */
retval = newAV();
av_push(retval, newSVhek(stashhek));
}
/* we don't want anyone modifying the cache entry but us,
and we do so by replacing it completely */
SvREADONLY_on(retval);
meta->mro_linear_c3 = retval;
return retval;
}
static void
report_event(PSTATE* p_state,
event_id_t event,
char *beg, char *end, U32 utf8,
token_pos_t *tokens, int num_tokens,
SV* self
)
{
struct p_handler *h;
dTHX;
dSP;
AV *array;
STRLEN my_na;
char *argspec;
char *s;
#ifdef UNICODE_HTML_PARSER
#define CHR_DIST(a,b) (utf8 ? utf8_distance((U8*)(a),(U8*)(b)) : (a) - (b))
#else
#define CHR_DIST(a,b) ((a) - (b))
#endif
/* capture offsets */
STRLEN offset = p_state->offset;
STRLEN line = p_state->line;
STRLEN column = p_state->column;
#if 0
{ /* used for debugging at some point */
char *s = beg;
int i;
/* print debug output */
switch(event) {
case E_DECLARATION: printf("DECLARATION"); break;
case E_COMMENT: printf("COMMENT"); break;
case E_START: printf("START"); break;
case E_END: printf("END"); break;
case E_TEXT: printf("TEXT"); break;
case E_PROCESS: printf("PROCESS"); break;
case E_NONE: printf("NONE"); break;
default: printf("EVENT #%d", event); break;
}
printf(" [");
while (s < end) {
if (*s == '\n') {
putchar('\\'); putchar('n');
}
else
putchar(*s);
s++;
}
printf("] %d\n", end - beg);
for (i = 0; i < num_tokens; i++) {
printf(" token %d: %d %d\n",
i,
tokens[i].beg - beg,
tokens[i].end - tokens[i].beg);
}
}
#endif
if (p_state->pending_end_tag && event != E_TEXT && event != E_COMMENT) {
token_pos_t t;
char dummy;
t.beg = p_state->pending_end_tag;
t.end = p_state->pending_end_tag + strlen(p_state->pending_end_tag);
p_state->pending_end_tag = 0;
report_event(p_state, E_END, &dummy, &dummy, 0, &t, 1, self);
SPAGAIN;
}
/* update offsets */
p_state->offset += CHR_DIST(end, beg);
if (line) {
char *s = beg;
char *nl = NULL;
while (s < end) {
if (*s == '\n') {
p_state->line++;
nl = s;
}
s++;
}
if (nl)
p_state->column = CHR_DIST(end, nl) - 1;
else
p_state->column += CHR_DIST(end, beg);
}
if (event == E_NONE)
goto IGNORE_EVENT;
#ifdef MARKED_SECTION
if (p_state->ms == MS_IGNORE)
goto IGNORE_EVENT;
#endif
/* tag filters */
if (p_state->ignore_tags || p_state->report_tags || p_state->ignore_elements) {
if (event == E_START || event == E_END) {
SV* tagname = p_state->tmp;
assert(num_tokens >= 1);
sv_setpvn(tagname, tokens[0].beg, tokens[0].end - tokens[0].beg);
if (utf8)
SvUTF8_on(tagname);
else
SvUTF8_off(tagname);
if (!CASE_SENSITIVE(p_state))
sv_lower(aTHX_ tagname);
if (p_state->ignoring_element) {
if (sv_eq(p_state->ignoring_element, tagname)) {
if (event == E_START)
p_state->ignore_depth++;
else if (--p_state->ignore_depth == 0) {
SvREFCNT_dec(p_state->ignoring_element);
p_state->ignoring_element = 0;
}
}
goto IGNORE_EVENT;
}
if (p_state->ignore_elements &&
hv_fetch_ent(p_state->ignore_elements, tagname, 0, 0))
{
p_state->ignoring_element = newSVsv(tagname);
p_state->ignore_depth = 1;
goto IGNORE_EVENT;
}
if (p_state->ignore_tags &&
hv_fetch_ent(p_state->ignore_tags, tagname, 0, 0))
{
goto IGNORE_EVENT;
}
if (p_state->report_tags &&
!hv_fetch_ent(p_state->report_tags, tagname, 0, 0))
{
goto IGNORE_EVENT;
}
}
else if (p_state->ignoring_element) {
goto IGNORE_EVENT;
}
}
h = &p_state->handlers[event];
if (!h->cb) {
/* event = E_DEFAULT; */
h = &p_state->handlers[E_DEFAULT];
if (!h->cb)
goto IGNORE_EVENT;
}
if (SvTYPE(h->cb) != SVt_PVAV && !SvTRUE(h->cb)) {
/* FALSE scalar ('' or 0) means IGNORE this event */
return;
}
if (p_state->unbroken_text && event == E_TEXT) {
/* should buffer text */
if (!p_state->pend_text)
p_state->pend_text = newSV(256);
if (SvOK(p_state->pend_text)) {
if (p_state->is_cdata != p_state->pend_text_is_cdata) {
flush_pending_text(p_state, self);
SPAGAIN;
goto INIT_PEND_TEXT;
}
}
else {
INIT_PEND_TEXT:
p_state->pend_text_offset = offset;
p_state->pend_text_line = line;
p_state->pend_text_column = column;
p_state->pend_text_is_cdata = p_state->is_cdata;
sv_setpvn(p_state->pend_text, "", 0);
if (!utf8)
SvUTF8_off(p_state->pend_text);
}
#ifdef UNICODE_HTML_PARSER
if (utf8 && !SvUTF8(p_state->pend_text))
sv_utf8_upgrade(p_state->pend_text);
if (utf8 || !SvUTF8(p_state->pend_text)) {
sv_catpvn(p_state->pend_text, beg, end - beg);
}
else {
SV *tmp = newSVpvn(beg, end - beg);
sv_utf8_upgrade(tmp);
sv_catsv(p_state->pend_text, tmp);
SvREFCNT_dec(tmp);
}
#else
sv_catpvn(p_state->pend_text, beg, end - beg);
#endif
return;
}
else if (p_state->pend_text && SvOK(p_state->pend_text)) {
flush_pending_text(p_state, self);
SPAGAIN;
}
/* At this point we have decided to generate an event callback */
argspec = h->argspec ? SvPV(h->argspec, my_na) : "";
if (SvTYPE(h->cb) == SVt_PVAV) {
if (*argspec == ARG_FLAG_FLAT_ARRAY) {
argspec++;
array = (AV*)h->cb;
}
else {
/* start sub-array for accumulator array */
array = newAV();
}
}
else {
array = 0;
if (*argspec == ARG_FLAG_FLAT_ARRAY)
argspec++;
/* start argument stack for callback */
ENTER;
SAVETMPS;
PUSHMARK(SP);
}
for (s = argspec; *s; s++) {
SV* arg = 0;
int push_arg = 1;
enum argcode argcode = (enum argcode)*s;
switch( argcode ) {
case ARG_SELF:
arg = sv_mortalcopy(self);
break;
case ARG_TOKENS:
if (num_tokens >= 1) {
AV* av = newAV();
SV* prev_token = &PL_sv_undef;
int i;
av_extend(av, num_tokens);
for (i = 0; i < num_tokens; i++) {
if (tokens[i].beg) {
prev_token = newSVpvn(tokens[i].beg, tokens[i].end-tokens[i].beg);
if (utf8)
SvUTF8_on(prev_token);
av_push(av, prev_token);
}
else { /* boolean */
av_push(av, p_state->bool_attr_val
? newSVsv(p_state->bool_attr_val)
: newSVsv(prev_token));
}
}
arg = sv_2mortal(newRV_noinc((SV*)av));
}
break;
case ARG_TOKENPOS:
if (num_tokens >= 1 && tokens[0].beg >= beg) {
AV* av = newAV();
int i;
av_extend(av, num_tokens*2);
for (i = 0; i < num_tokens; i++) {
if (tokens[i].beg) {
av_push(av, newSViv(CHR_DIST(tokens[i].beg, beg)));
av_push(av, newSViv(CHR_DIST(tokens[i].end, tokens[i].beg)));
}
else { /* boolean tag value */
av_push(av, newSViv(0));
av_push(av, newSViv(0));
}
}
arg = sv_2mortal(newRV_noinc((SV*)av));
}
break;
case ARG_TOKEN0:
case ARG_TAGNAME:
/* fall through */
case ARG_TAG:
if (num_tokens >= 1) {
arg = sv_2mortal(newSVpvn(tokens[0].beg,
tokens[0].end - tokens[0].beg));
if (utf8)
SvUTF8_on(arg);
if (!CASE_SENSITIVE(p_state) && argcode != ARG_TOKEN0)
sv_lower(aTHX_ arg);
if (argcode == ARG_TAG && event != E_START) {
char *e_type = "!##/#?#";
sv_insert(arg, 0, 0, &e_type[event], 1);
}
}
break;
case ARG_ATTR:
case ARG_ATTRARR:
if (event == E_START) {
HV* hv;
int i;
if (argcode == ARG_ATTR) {
hv = newHV();
arg = sv_2mortal(newRV_noinc((SV*)hv));
}
else {
#ifdef __GNUC__
/* gcc -Wall reports this variable as possibly used uninitialized */
hv = 0;
#endif
push_arg = 0; /* deal with argument pushing here */
}
for (i = 1; i < num_tokens; i += 2) {
SV* attrname = newSVpvn(tokens[i].beg,
tokens[i].end-tokens[i].beg);
SV* attrval;
if (utf8)
SvUTF8_on(attrname);
if (tokens[i+1].beg) {
char *beg = tokens[i+1].beg;
STRLEN len = tokens[i+1].end - beg;
if (*beg == '"' || *beg == '\'') {
assert(len >= 2 && *beg == beg[len-1]);
beg++; len -= 2;
}
attrval = newSVpvn(beg, len);
if (utf8)
SvUTF8_on(attrval);
if (!p_state->attr_encoded) {
#ifdef UNICODE_HTML_PARSER
if (p_state->utf8_mode)
sv_utf8_decode(attrval);
#endif
decode_entities(aTHX_ attrval, p_state->entity2char, 0);
if (p_state->utf8_mode)
SvUTF8_off(attrval);
}
}
else { /* boolean */
if (p_state->bool_attr_val)
attrval = newSVsv(p_state->bool_attr_val);
else
attrval = newSVsv(attrname);
}
if (!CASE_SENSITIVE(p_state))
sv_lower(aTHX_ attrname);
if (argcode == ARG_ATTR) {
if (hv_exists_ent(hv, attrname, 0) ||
!hv_store_ent(hv, attrname, attrval, 0)) {
SvREFCNT_dec(attrval);
}
SvREFCNT_dec(attrname);
}
else { /* ARG_ATTRARR */
if (array) {
av_push(array, attrname);
av_push(array, attrval);
}
else {
XPUSHs(sv_2mortal(attrname));
XPUSHs(sv_2mortal(attrval));
}
}
}
}
else if (argcode == ARG_ATTRARR) {
push_arg = 0;
}
break;
case ARG_ATTRSEQ: /* (v2 compatibility stuff) */
if (event == E_START) {
AV* av = newAV();
int i;
for (i = 1; i < num_tokens; i += 2) {
SV* attrname = newSVpvn(tokens[i].beg,
tokens[i].end-tokens[i].beg);
if (utf8)
SvUTF8_on(attrname);
if (!CASE_SENSITIVE(p_state))
sv_lower(aTHX_ attrname);
av_push(av, attrname);
}
arg = sv_2mortal(newRV_noinc((SV*)av));
}
break;
case ARG_TEXT:
arg = sv_2mortal(newSVpvn(beg, end - beg));
if (utf8)
SvUTF8_on(arg);
break;
case ARG_DTEXT:
if (event == E_TEXT) {
arg = sv_2mortal(newSVpvn(beg, end - beg));
if (utf8)
SvUTF8_on(arg);
if (!p_state->is_cdata) {
#ifdef UNICODE_HTML_PARSER
if (p_state->utf8_mode)
sv_utf8_decode(arg);
#endif
decode_entities(aTHX_ arg, p_state->entity2char, 1);
if (p_state->utf8_mode)
SvUTF8_off(arg);
}
}
break;
case ARG_IS_CDATA:
if (event == E_TEXT) {
arg = boolSV(p_state->is_cdata);
}
break;
case ARG_SKIPPED_TEXT:
arg = sv_2mortal(p_state->skipped_text);
p_state->skipped_text = newSVpvn("", 0);
break;
case ARG_OFFSET:
arg = sv_2mortal(newSViv(offset));
break;
case ARG_OFFSET_END:
arg = sv_2mortal(newSViv(offset + CHR_DIST(end, beg)));
break;
case ARG_LENGTH:
arg = sv_2mortal(newSViv(CHR_DIST(end, beg)));
break;
case ARG_LINE:
arg = sv_2mortal(newSViv(line));
break;
case ARG_COLUMN:
arg = sv_2mortal(newSViv(column));
break;
case ARG_EVENT:
assert(event >= 0 && event < EVENT_COUNT);
arg = sv_2mortal(newSVpv(event_id_str[event], 0));
break;
case ARG_LITERAL:
{
int len = (unsigned char)s[1];
arg = sv_2mortal(newSVpvn(s+2, len));
if (SvUTF8(h->argspec))
SvUTF8_on(arg);
s += len + 1;
}
break;
case ARG_UNDEF:
arg = sv_mortalcopy(&PL_sv_undef);
break;
default:
arg = sv_2mortal(newSVpvf("Bad argspec %d", *s));
break;
}
if (push_arg) {
if (!arg)
arg = sv_mortalcopy(&PL_sv_undef);
if (array) {
/* have to fix mortality here or add mortality to
* XPUSHs after removing it from the switch cases.
*/
av_push(array, SvREFCNT_inc(arg));
}
else {
XPUSHs(arg);
}
}
}
if (array) {
if (array != (AV*)h->cb)
av_push((AV*)h->cb, newRV_noinc((SV*)array));
}
else {
PUTBACK;
if ((enum argcode)*argspec == ARG_SELF && !SvROK(h->cb)) {
char *method = SvPV(h->cb, my_na);
perl_call_method(method, G_DISCARD | G_EVAL | G_VOID);
}
else {
perl_call_sv(h->cb, G_DISCARD | G_EVAL | G_VOID);
}
if (SvTRUE(ERRSV)) {
RETHROW;
}
FREETMPS;
LEAVE;
}
if (p_state->skipped_text)
SvCUR_set(p_state->skipped_text, 0);
return;
IGNORE_EVENT:
if (p_state->skipped_text) {
if (event != E_TEXT && p_state->pend_text && SvOK(p_state->pend_text))
flush_pending_text(p_state, self);
#ifdef UNICODE_HTML_PARSER
if (utf8 && !SvUTF8(p_state->skipped_text))
sv_utf8_upgrade(p_state->skipped_text);
if (utf8 || !SvUTF8(p_state->skipped_text)) {
#endif
sv_catpvn(p_state->skipped_text, beg, end - beg);
#ifdef UNICODE_HTML_PARSER
}
else {
SV *tmp = newSVpvn(beg, end - beg);
sv_utf8_upgrade(tmp);
sv_catsv(p_state->pend_text, tmp);
SvREFCNT_dec(tmp);
}
#endif
}
#undef CHR_DIST
return;
}
