p5pRT
commented
10 years ago From @bulk88
Created by @bulk88
I ran across some strange looking code so I investigated. The SPAGAINs and PUTBACKs can be removed since magic calls now swap stacks. But removing the X from XPUSH is challenging. There is
" EXTEND(SP\, HvUSEDKEYS(keys) * (dokeys + dovalues));"
But it DOESNT extend the stack enough to avoid the X being called in XPUSHs when executing
"our %Config = %Config::Config;"
in ExtUtils\MakeMaker\Config.pm . I tried doing ------------------------ use Config; use Data::Dumper; %h = ('a' => 1\, 'b' =>2); print(Dumper(scalar(%h))); print(Dumper(scalar(%Config))); ------------------------ which gave
#############5.12 C:\Documents and Settings\Owner\Desktop>perl n11.pl $VAR1 = "2/8"; $VAR1 = 1; #############5.19 C:\Documents and Settings\Owner\Desktop>perl n11.pl $VAR1 = '2/8'; $VAR1 = 1;
But %Config clearly doesn't have 1 key in it. Config_heavy.pl does not have a SCALAR method\, and it seem optional to have one per Perl_magic_scalarpack(pTHX_ HV *hv\, MAGIC *mg) in mg.c. I found https://rt-archive.perl.org/perl5/Ticket/Display.html?id=18186 and https://rt-archive.perl.org/perl5/Ticket/Display.html?id=24798 which seem related to the fact that SCALAR returns an IMMORTAL SV when there is no SCALAR sub in the package\, not the key and bucket counts. Is this intended behavior or should it be revisted?
So is there a way to fix " EXTEND(SP\, HvUSEDKEYS(keys) * (dokeys + dovalues));" to remove the X from XPUSHs later in the sub and correctly EXTEND for tied HVs? Calling SCALAR doesn't work on %Config.
Also Perl_magic_scalarpack seems to do the same stash lookup twice for no good reason.
------------------------------------------------------------- SV * Perl_magic_scalarpack(pTHX_ HV *hv\, MAGIC *mg) { dVAR; SV *retval; SV * const tied = SvTIED_obj(MUTABLE_SV(hv)\, mg); HV * const pkg = SvSTASH((const SV *)SvRV(tied));
PERL_ARGS_ASSERT_MAGIC_SCALARPACK;
if (!gv_fetchmethod_autoload(pkg\, "SCALAR"\, FALSE)) {\<\<\<\<\<\<\<\<\<\<\<\<\<\<\<FIRST TIME SV *key; if (HvEITER_get(hv)) /* we are in an iteration so the hash cannot be empty */ return &PL_sv_yes; /* no xhv_eiter so now use FIRSTKEY */ key = sv_newmortal(); magic_nextpack(MUTABLE_SV(hv)\, mg\, key); HvEITER_set(hv\, NULL); /* need to reset iterator */ return SvOK(key) ? &PL_sv_yes : &PL_sv_no; }
/* there is a SCALAR method that we can call */ retval = Perl_magic_methcall(aTHX_ MUTABLE_SV(hv)\, mg\, SV_CONST(SCALAR)\, 0\, 0);\<\<\<\<\<\<\<\<\<\<\<\<SECOND TIME if (!retval) retval = &PL_sv_undef; return retval; } -------------------------------------------------------------
I did testing with the following patch. That is how I know the SPAGAIN and PUTBACK can go. The extend check assert were so frequent they were commented out so the stack realloc asserts can trip without noise. harness did pass all tests with no crashes from stack realloc assert. ("DebugBreak()" is a C breakpoint function so harness stops because a child perl proc crashed and I can examine the .t perl process with a C debugger).
toddr
Migrated from rt.perl.org#121348 (status was 'open')
Searchable as RT121348$