There is a further complication with non-closure anonymous subs (i.e. those that do not refer to any lexicals outside that sub). In this case, the anonymous prototype is shared rather than being cloned. This has the consequence that the parent may be freed while there are still active children, eg
BEGIN { $a = sub { eval '$x' } }
In this case, the BEGIN is freed immediately after execution since there are no active references to it: the anon sub prototype has "CvWEAKOUTSIDE" set since it's not a closure, and $a points to the same CV, so it doesn't contribute to BEGIN's refcount either. When $a is executed, the "eval '$x'" causes the chain of "CvOUTSIDE"s to be followed, and the freed BEGIN is accessed.
To avoid this, whenever a CV and its associated pad is freed, any "&" entries in the pad are explicitly removed from the pad, and if the refcount of the pointed-to anon sub is still positive, then that child's "CvOUTSIDE" is set to point to its grandparent. This will only occur in the single specific case of a non-closure anon prototype having one or more active references (such as $a above).
One other thing to consider is that a CV may be merely undefined rather than freed, eg "undef &foo". In this case, its refcount may not have reached zero, but we still delete its pad and its "CvROOT" etc. Since various children may still have their "CvOUTSIDE" pointing at this undefined CV, we keep its own "CvOUTSIDE" for the time being, so that the chain of lexical scopes is unbroken. For example, the following should print 123:
my $x = 123;
sub tmp { sub { eval '$x' } }
my $a = tmp();
undef &tmp;
print $a->();
bool CvWEAKOUTSIDE(CV *cv)
void CX_CURPAD_SAVE(struct context)
SV * CX_CURPAD_SV(struct context, PADOFFSET po)
SV * PAD_BASE_SV(PADLIST padlist, PADOFFSET po)
void PAD_CLONE_VARS(PerlInterpreter *proto_perl \)
U32 PAD_COMPNAME_FLAGS(PADOFFSET po)
STRLEN PAD_COMPNAME_GEN(PADOFFSET po)
STRLEN PAD_COMPNAME_GEN_set(PADOFFSET po, int gen)
HV * PAD_COMPNAME_OURSTASH(PADOFFSET po)
char * PAD_COMPNAME_PV(PADOFFSET po)
HV * PAD_COMPNAME_TYPE(PADOFFSET po)
void PAD_DUP(PADLIST dstpad, PADLIST srcpad, CLONE_PARAMS* param)
void PAD_RESTORE_LOCAL(PAD *opad)
void PAD_SAVE_LOCAL(PAD *opad, PAD *npad)
void PAD_SAVE_SETNULLPAD()
SV * PAD_SETSV(PADOFFSET po, SV* sv)
SAVECOMPPAD();
PAD_SET_CUR_NOSAVE(padlist,n);
void PAD_SET_CUR(PADLIST padlist, I32 n)
void PAD_SET_CUR_NOSAVE(PADLIST padlist, I32 n)
void PAD_SV(PADOFFSET po)
SV * PAD_SVl(PADOFFSET po)
void SAVECLEARSV(SV **svp)
void SAVECOMPPAD()
XXX DAPM it would make more sense to make the arg a PADOFFSET
void SAVEPADSV(PADOFFSET po)
Currently only useful internally when determining if a GV should be created even in rvalue contexts.
"flags" is not used at present but available for future extension to allow selecting particular classes of magical variable.
Currently assumes that "name" is NUL terminated (as well as len being valid). This assumption is met by all callers within the perl core, which all pass pointers returned by SvPV.
bool is_gv_magical(const char *name, STRLEN len, U32 flags)
bool is_gv_magical_sv(SV *name, U32 flags)
HV * refcounted_he_chain_2hv(const struct refcounted_he *c)
void refcounted_he_free(struct refcounted_he *he)
struct refcounted_he * refcounted_he_new(struct refcounted_he *const parent, SV *const key, SV *const value)
PerlIO* start_glob(SV* pattern, IO *io)
int magic_sethint(SV* sv, MAGIC* mg)
void mg_localize(SV* sv, SV* nsv)
You are responsible for "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).
AV* mro_get_linear_isa_c3(HV* stash, I32 level)
You are responsible for "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).
AV* mro_get_linear_isa_dfs(HV* stash, I32 level)
void mro_isa_changed_in(HV* stash)
For these purposes ``forms'' are a kind-of CV, eval"``s are too (except they're not callable at will and are always thrown away after the eval''" is done executing). Require'd files are simply evals without any outer lexical scope.
XSUBs don't have CvPADLIST set - dXSTARG fetches values from PL_curpad, but that is really the callers pad (a slot of which is allocated by every entersub).
The CvPADLIST AV has does not have AvREAL set, so REFCNT of component items is managed ``manual'' (mostly in pad.c) rather than normal av.c rules. The items in the AV are not SVs as for a normal AV, but other AVs:
0'th Entry of the CvPADLIST is an AV which represents the ``names'' or rather the ``static type information'' for lexicals.
The CvDEPTH'th entry of CvPADLIST AV is an AV which is the stack frame at that depth of recursion into the CV. The 0'th slot of a frame AV is an AV which is @_. other entries are storage for variables and op targets.
During compilation: "PL_comppad_name" is set to the names AV. "PL_comppad" is set to the frame AV for the frame CvDEPTH == 1. "PL_curpad" is set to the body of the frame AV (i.e. AvARRAY(PL_comppad)).
During execution, "PL_comppad" and "PL_curpad" refer to the live frame of the currently executing sub.
Iterating over the names AV iterates over all possible pad items. Pad slots that are SVs_PADTMP (targets/GVs/constants) end up having &PL_sv_undef ``names'' (see pad_alloc()).
Only my/our variable (SVs_PADMY/SVs_PADOUR) slots get valid names. The rest are op targets/GVs/constants which are statically allocated or resolved at compile time. These don't have names by which they can be looked up from Perl code at run time through eval"`` like my/our variables can be. Since they can't be looked up by ''name" but only by their index allocated at compile time (which is usually in PL_op->op_targ), wasting a name SV for them doesn't make sense.
The SVs in the names AV have their PV being the name of the variable. xlow+1..xhigh inclusive in the NV union is a range of cop_seq numbers for which the name is valid. For typed lexicals name SV is SVt_PVMG and SvSTASH points at the type. For "our" lexicals, the type is also SVt_PVMG, with the SvOURSTASH slot pointing at the stash of the associated global (so that duplicate "our" declarations in the same package can be detected). SvUVX is sometimes hijacked to store the generation number during compilation.
If SvFAKE is set on the name SV, then that slot in the frame AV is a REFCNT'ed reference to a lexical from ``outside''. In this case, the name SV does not use xlow and xhigh to store a cop_seq range, since it is in scope throughout. Instead xhigh stores some flags containing info about the real lexical (is it declared in an anon, and is it capable of being instantiated multiple times?), and for fake ANONs, xlow contains the index within the parent's pad where the lexical's value is stored, to make cloning quicker.
If the 'name' is '&' the corresponding entry in frame AV is a CV representing a possible closure. (SvFAKE and name of '&' is not a meaningful combination currently but could become so if "my sub foo {}" is implemented.)
Note that formats are treated as anon subs, and are cloned each time write is called (if necessary).
The flag SVf_PADSTALE is cleared on lexicals each time the my() is executed, and set on scope exit. This allows the 'Variable $x is not available' warning to be generated in evals, such as
{ my $x = 1; sub f { eval '$x'} } f();
AV * CvPADLIST(CV *cv)
CV* cv_clone(CV* proto)
void cv_dump(const CV *cv, const char *title)
void do_dump_pad(I32 level, PerlIO *file, PADLIST *padlist, int full)
U32 intro_my()
PADOFFSET pad_add_anon(SV* sv, OPCODE op_type)
If fake, it means we're cloning an existing entry
PADOFFSET pad_add_name(const char *name, HV* typestash, HV* ourstash, bool clone, bool state)
PADOFFSET pad_alloc(I32 optype, U32 tmptype)
void pad_block_start(int full)
void pad_check_dup(const char* name, bool is_our, const HV* ourstash)
Returns the offset in the bottom pad of the lex or the fake lex. cv is the CV in which to start the search, and seq is the current cop_seq to match against. If warn is true, print appropriate warnings. The out_* vars return values, and so are pointers to where the returned values should be stored. out_capture, if non-null, requests that the innermost instance of the lexical is captured; out_name_sv is set to the innermost matched namesv or fake namesv; out_flags returns the flags normally associated with the IVX field of a fake namesv.
Note that pad_findlex() is recursive; it recurses up the chain of CVs, then comes back down, adding fake entries as it goes. It has to be this way because fake namesvs in anon protoypes have to store in xlow the index into the parent pad.
PADOFFSET pad_findlex(const char *name, const CV* cv, U32 seq, int warn, SV** out_capture, SV** out_name_sv, int *out_flags)
PADOFFSET pad_findmy(const char* name)
void pad_fixup_inner_anons(PADLIST *padlist, CV *old_cv, CV *new_cv)
void pad_free(PADOFFSET po)
void pad_leavemy()
padnew_CLONE this pad is for a cloned CV
padnew_SAVE save old globals
padnew_SAVESUB also save extra stuff for start of sub
PADLIST* pad_new(int flags)
void pad_push(PADLIST *padlist, int depth)
void pad_reset()
void pad_setsv(PADOFFSET po, SV* sv)
void pad_swipe(PADOFFSET po, bool refadjust)
void pad_tidy(padtidy_type type)
(This function should really be called pad_free, but the name was already taken)
void pad_undef(CV* cv)
SV * PL_DBsingle
GV * PL_DBsub
SV * PL_DBtrace
bool PL_dowarn
GV* PL_last_in_gv
SV* PL_ofs_sv
SV* PL_rs
djSP;
void sv_add_arena(char* ptr, U32 size, U32 flags)
I32 sv_clean_all()
void sv_clean_objs()
void sv_free_arenas()
SV* sv_2num(SV* sv)
The name is returned as a mortal SV.
Assumes that PL_op is the op that originally triggered the error, and that PL_comppad/PL_curpad points to the currently executing pad.
SV* find_uninit_var(OP* obase, SV* uninit_sv, bool top)
void report_uninit(SV* uninit_sv)