In perl.git, the branch sprout/padlist has been created
<http://perl5.git.perl.org/perl.git/commitdiff/a8408c3a0c51a76ab29c5e7091dc6e016706974b?hp=0000000000000000000000000000000000000000>
at a8408c3a0c51a76ab29c5e7091dc6e016706974b (commit)
- Log -----------------------------------------------------------------
commit a8408c3a0c51a76ab29c5e7091dc6e016706974b
Author: Father Chrysostomos <[email protected]>
Date: Fri Aug 17 13:01:49 2012 -0700
Fix format closure bug with redefined outer sub
CVs close over their outer CVs. So, when you write:
my $x = 52;
sub foo {
sub bar {
sub baz {
$x
}
}
}
bazâs CvOUTSIDE pointer points to bar, barâs CvOUTSIDE points to foo,
and fooâs to the main cv.
When the inner reference to $x is looked up, the CvOUTSIDE chain is
followed, and each subâs pad is looked at to see if it has an $x.
(This happens at compile time.)
It can happen that bar is undefined and then redefined:
undef &bar;
eval 'sub bar { my $x = 34 }';
After this, baz will still refer to the main cvâs $x (52), but, if baz
had âeval '$x'â instead of just $x, it would see the new barâs $x.
(Itâs not really a new bar, as its refaddr is the same, but it has a
new body.)
This particular case is harmless, and is obscure enough that we could
define it any way we want, and it could still be considered correct.
The real problem happens when CVs are cloned.
When a CV is cloned, its name pad already contains the offsets into
the parent pad where the values are to be found. If the outer CV
has been undefined and redefined, those pad offsets can be com-
pletely bogus.
Normally, a CV cannot be cloned except when its outer CV is running.
And the outer CV cannot have been undefined without also throwing
away the op that would have cloned the prototype.
But formats can be cloned when the outer CV is not running. So it
is possible for cloned formats to close over bogus entries in a new
parent pad.
In this example, \$x gives us an array ref. It shows ARRAY(0xbaff1ed)
instead of SCALAR(0xdeafbee):
sub foo {
my $x;
format =
@
($x,warn \$x)[0]
.
}
undef &foo;
eval 'sub foo { my @x; write }';
foo
__END__
And if the offset that the formatâs pad closes over is beyond the end
of the parentâs new pad, we can even get a crash, as in this case:
eval
'sub foo {' .
'{my ($a,$b,$c,$d,$e,$f,$g,$h,$i,$j,$k,$l,$m,$n,$o,$p,$q,$r,$s,$t,$u)}'x999
. q|
my $x;
format =
@
($x,warn \$x)[0]
.
}
|;
undef &foo;
eval 'sub foo { my @x; my $x = 34; write }';
foo();
__END__
So now, instead of using CvROOT to identify clones of
CvOUTSIDE(format), we use the padlist ID instead. Padlists donât
actually have an ID, so we give them one. Any time a sub is cloned,
the new padlist gets the same ID as the old. The format needs to
remember what its outer subâs padlist ID was, so we put that in the
padlist struct, too.
M embed.fnc
M embedvar.h
M intrpvar.h
M pad.c
M pad.h
M pp.c
M pp.h
M pp_ctl.c
M proto.h
M t/comp/form_scope.t
M toke.c
commit f23bc75841348cd07856392e4d4903200f2b53f3
Author: Father Chrysostomos <[email protected]>
Date: Thu Aug 16 23:44:11 2012 -0700
pp_ctl.c:pp_dbstate: Donât adjust CvDEPTH for XSUBs
Commit c127bd3aaa5c5 made XS DB::DB subs work. Before that,
pp_dbstate assumed DB::DB was written it perl. It adjusts CvDEPTH
when calling the XSUB, which serves no purpose. It was presumably
just copied from the pure-Perl-calling code. pp_entersub does-
nât do this.
M pp_ctl.c
commit 728fada04b4d19213310eeb3c043757927ce7a78
Author: Father Chrysostomos <[email protected]>
Date: Thu Aug 16 21:54:53 2012 -0700
[perl #113718] Add inline.h
We can put static inline functions here, and they can depend on
function prototypes and struct definitions from other header
files.
M MANIFEST
A inline.h
M perl.h
commit bac9ebfe07516d102a911cb11bd309b18a41869e
Author: Father Chrysostomos <[email protected]>
Date: Thu Aug 16 16:47:38 2012 -0700
Increase $B::Xref::VERSION from 1.03 to 1.04
M ext/B/B/Xref.pm
commit f395734db0899dc11af236f9c0fb6ac64fbdcf5f
Author: Father Chrysostomos <[email protected]>
Date: Thu Aug 16 16:46:20 2012 -0700
Stop padlists from being AVs
In order to fix a bug, I need to add new fields to padlists. But I
cannot easily do that as long as they are AVs.
So I have created a new padlist struct.
This not only allows me to extend the padlist struct with new members
as necessary, but also saves memory, as we now have a three-pointer
struct where before we had a whole SV head (3-4 pointers) + XPVAV (5
pointers).
This will unfortunately break half of CPAN, but the pad API docs
clearly say this:
NOTE: this function is experimental and may change or be
removed without notice.
This would have broken B::Debug, but a patch sent upstream has already
been integrated into blead with commit 9d2d23d981.
M av.c
M dump.c
M embed.fnc
M embed.h
M ext/B/B.xs
M ext/B/B/Xref.pm
M ext/B/typemap
M ext/XS-APItest/APItest.xs
M pad.c
M pad.h
M perl.h
M proto.h
M sv.c
commit 33dd83b52606334a0d1313fe4c9f8096868481b1
Author: Father Chrysostomos <[email protected]>
Date: Wed Aug 15 22:27:54 2012 -0700
Use PADLIST in more places
Much code relies on the fact that PADLIST is typedeffed as AV.
PADLIST should be treated as a distinct type. (This also means cast-
ing PADLISTs to AV when passing them to av_fetch.)
M cop.h
M dump.c
M ext/XS-APItest/APItest.xs
M op.c
M pad.c
M pp_ctl.c
M pp_hot.c
M pp_sort.c
M sv.c
M sv.h
commit 47b7605e418a55530e7aab5e3d0a33eb2ccd331b
Author: Father Chrysostomos <[email protected]>
Date: Wed Aug 15 22:11:46 2012 -0700
Move PAD(LIST) typedefs to perl.h
otherwise they can only be used in some header files.
M pad.h
M perl.h
-----------------------------------------------------------------------
--
Perl5 Master Repository
