Father Chrysostomos [Wed, 31 Oct 2012 17:02:03 +0000 (10:02 -0700)]
Stop run-time regexp blocks from leaking regexps
This was leaking like a sieve: $var = '(?{})'; /stuff$var/;
When a run-time regular expression has code blocks in it,
those are compiled separately inside their own qr thingy (see
S_compile_runtime_code in regcomp.c).
In re_op_compile, the newly-compiled code blocks are stored in
pRExC_state->code_blocks, which is a mallocked array. That array also
holds reference counts on the regular expressions from which the code
blocks derive their existence. When the whole regular expression is
compiled, the code blocks are fetched from that array, and the new
regular expression ends up holding a reference count on those code
block’s originating regular expressions.
The reference counts that pRExC_state->code_blocks had were not low-
ered when pRExC_state->code_blocks was freed, except for qr/stuff$var/
(because the qr// would take ownership of those reference counts,
which would be lowered when the outer qr// itself was freed).
Father Chrysostomos [Tue, 30 Oct 2012 23:41:27 +0000 (16:41 -0700)]
Stop / $looks_like_block/ from leaking
If an interpolated string looks as though it contains a regexp code
block, the regexp compiler will evaluate it inside qr'...' and then
extract the code blocks from the resulting regexp object.
If it turned out to be a false positive (e.g., "[(?{})]"), then
the code to handle this returned without freeing the temporary reg-
exp object.
Ricardo Signes [Thu, 1 Nov 2012 14:20:20 +0000 (10:20 -0400)]
add perl5.16.2 to perlhist
Ricardo Signes [Thu, 1 Nov 2012 14:19:47 +0000 (10:19 -0400)]
add perl5162delta
Ricardo Signes [Thu, 1 Nov 2012 14:07:46 +0000 (10:07 -0400)]
add the 5.16.2 epigraph
Konovalov, Vadim (Vadim)** CTR ** [Thu, 1 Nov 2012 14:03:34 +0000 (14:03 +0000)]
Initial (incomplete) patch to start restoring WinCE build
Subject: RE: status of WinCE Perl port in 2012
From: "Konovalov, Vadim (Vadim)** CTR **" <vadim.konovalov@alcatel-lucent.com>
Date: Tue, 23 Oct 2012 14:26:49 +0200
Message-ID: <35BF8D9716175C43BB9D67CA60CC345E028EE0C899@FRMRSSXCHMBSC2.dc-m.alcatel-lucent.com>
Steve Hay [Thu, 1 Nov 2012 13:43:39 +0000 (13:43 +0000)]
Remove __attribute__malloc__ from MYSWAP functions
These functions are only used when the native sockets functions are not
available, e.g. when building miniperl on Windows following commit
19253ae62c, so gcc's warning about ignoring the __malloc__ attribute here
is not normally seen.
The addition of "a" to these functions in embed.fnc by
commit
f54cb97a39 was presumably wrong since none of them actually
allocate any memory (nor did so at the time), so change it to just "R"
(which is implied by the "a" and is still appropriate).
Daniel Dragan [Sat, 13 Oct 2012 23:37:33 +0000 (19:37 -0400)]
Win32 miniperl: delay loading for Winsock, and then remove it
Slim down the image and speed up start up time for Win32 miniperl by
removing Winsock. Also if the build process on Win32 in the future
requires sockets, commenting one line in win32.h will turn sockets back on
for miniperl, but this time with delay loading on VC Perl. The only casulty
of no sockets for Win32 miniperl was figuring out the computer's name in
win32/config_sh.PL. A workaround by using an ENV var was implemented. The
purpose of this commit is to speed up the build process of Perl.
As said in the comment in win32.h, the WIN32_NO_SOCKETS macro is
incomplete in implementation. It is only removed winsock from being linked
in in miniperl, not full Perl. PERL_IMPLICIT_SYS (specifically PerlSock in
win32/perlhost.h) and makedef.pl's hard coded list of win32_* function
exports cause winsock to still be linked in with even with
WIN32_NO_SOCKETS on full perl. Both PERL_IMPLICIT_SYS (win32/perlhost.h)
and makedef.pl would require changes to remove winsock from being linked
in on full perl in the future.
Steve Hay [Wed, 31 Oct 2012 08:37:41 +0000 (08:37 +0000)]
Add the DynaLoader upgrade to perldelta
Steve Hay [Wed, 31 Oct 2012 08:36:58 +0000 (08:36 +0000)]
Use correct type to avoid a cast added by
fe1c5936a5
(Suggested by Tony Cook.)
Steve Hay [Wed, 31 Oct 2012 08:25:25 +0000 (08:25 +0000)]
Bump DynaLoader's $VERSION after commit
fe1c5936a5
Daniel Dragan [Wed, 17 Oct 2012 15:56:29 +0000 (11:56 -0400)]
consting in perl.c:S_Internals_V and Win32 DynaLoader
These assorted static allocated variables were in RW memory in the perl
image. Move them to RO memory so they are sharable between different
Perl processes by the OS. The lack of consting in Win32 Dynaloader traces
to commit
0a753a76406 . S_Internals_V traces to commit
4a5df386486 .
James E Keenan [Tue, 30 Oct 2012 23:25:39 +0000 (19:25 -0400)]
Three spelling corrections.
Father Chrysostomos [Tue, 30 Oct 2012 21:38:27 +0000 (14:38 -0700)]
utf8.c: Stop _core_swash_init from leaking
If an %INC hook or $@ assignment dies, then a scalar is leaked. I
don’t know that it is possible to test this.
Father Chrysostomos [Tue, 30 Oct 2012 16:44:26 +0000 (09:44 -0700)]
Allow regexp-to-pvlv assignment
Since the xpvlv and regexp structs conflict, we have to find somewhere
else to put the regexp struct.
I was going to sneak it in SvPVX, allocating a buffer large
enough to fit the regexp struct followed by the string, and have
SvPVX - sizeof(regexp) point to the struct. But that would make all
regexp flag-checking macros fatter, and those are used in hot code.
So I came up with another method. Regexp stringification is not
speed-critical. So we can move the regexp stringification out of
re->sv_u and put it in the regexp struct. Then the regexp struct
itself can be pointed to by re->sv_u. So SVt_REGEXPs will have
re->sv_any and re->sv_u pointing to the same spot. PVLVs can then
have sv->sv_any point to the xpvlv body as usual, but have sv->sv_u
point to a regexp struct. All regexp member access can go through
sv_u instead of sv_any, which will be no slower than before.
Regular expressions will no longer be SvPOK, so we give sv_2pv spec-
ial logic for regexps. We don’t need to make the regexp struct
larger, as SvLEN is currently always 0 iff mother_re is set. So we
can replace the SvLEN field with the pv.
SvFAKE is never used without SvPOK or SvSCREAM also set. So we can
use that to identify regexps.
Father Chrysostomos [Mon, 29 Oct 2012 07:01:02 +0000 (00:01 -0700)]
regcomp.c: Really stop regexp-to-pv assignment from leaking
edd9fea2b8 was not enough. A scalar may hold a PV even with the
SvPOKp flag off:
$ ./perl -Ilib -e 'warn $$; while(1){ $x = "a"; $x = 1; $x = ${qr//}}'
Father Chrysostomos [Mon, 29 Oct 2012 06:56:01 +0000 (23:56 -0700)]
Turn off OK flags when creating a regexp.
$ perl5.16.0 -le '$x = 1.1; $x = ${qr//}; print 0+$x'
4.
15336878055219e-317
$ perl5.16.0 -le '$x = 1; $x = ${qr//}; print 0+$x'
1645024
Very strange.
Under debugging builds, both produce assertion failures.
By turning off all OK flags, we also prevent the destination’s utf8-
ness from sticking.
Father Chrysostomos [Mon, 29 Oct 2012 06:49:48 +0000 (23:49 -0700)]
sv.c: Drop PV when assigning over regexp
$x = ${qr//};
$x = 3;
On the second line, we don’t need to copy the stringification of the
regexp, since we are just going to clobber it anyway.
Father Chrysostomos [Mon, 29 Oct 2012 06:47:22 +0000 (23:47 -0700)]
Prune dead code in sv.c:sv_force_normal_flags
When a regexp is unregexped, a new SV (temp) is created, so it
can swap bodies with the regular expression (sv), and then temp
can be freed.
If SvLEN is 0, then a scalar does not own its string buffer. Copied
regexps use that mechanism to share strings; only the original regexp
owns the string.
This little bit of code for handling the SvPVX field is strange:
/* Remember that SvPVX is in the head, not the body. */
if (SvLEN(temp)) {
SvLEN_set(temp, SvLEN(sv));
/* This signals "buffer is owned by someone else" in sv_clear,
which is the least effort way to stop it freeing the buffer.
*/
SvLEN_set(sv, SvLEN(sv)+1);
} else {
/* Their buffer is already owned by someone else. */
SvPVX(sv) = savepvn(SvPVX(sv), SvCUR(sv));
SvLEN_set(temp, SvCUR(sv)+1);
}
Checking SvLEN(temp) is pointless if we have just created temp. That
check is always false. Presumably it was meant to be SvLEN(sv). But
the original regexp scalar (i.e., not a copy) can never make it to
this function. So SvLEN(sv) is always 0, which is why this has
not caused any problem. The SvLEN_set inside the apodosis is also
strange. ‘This signals "buffer is owned by someone else"’. No it
certainly does not! It is not setting SvLEN to 0, but definitely to
non-zero. I can only assume this is a copy-and-paste error, which has
never caused a problem because it is unreachable.
I hereby excise this code (leaving the contents of the else).
Father Chrysostomos [Mon, 29 Oct 2012 06:37:21 +0000 (23:37 -0700)]
regcomp.c: Don’t point mother_re to regexp copy
In code like this:
$x = ${qr//};
$y = $x
undef $x;
We end up with $y’s mother_re pointer pointing to something that is
not a regexp.
This can cause thread cloning to create a new regexp with its SvPVX
pointing to the string buffer of the original regexp:
$x = ${qr/abcd/};
$y = $x;
use Devel::Peek;
Dump($y);
$x = *3;
use threads;
async { Dump $y; print $y, "\n" }->join;
The dump shows that both $y’s share the same string buffer, and nei-
ther claims ownership to it.
I have not been able to make this crash or reuse the string for some-
thing else, but still this is walking a fine line. Theoretically, it
should be possible for that string to be freed and reused in the par-
ent thread while the child thread is still using it.
Instead of pointing mother_re to the rhs of the assignment, point it
to the original re from which the rhs derives its existence. I.e.,
copy the mother_re field.
Father Chrysostomos [Sun, 28 Oct 2012 21:41:55 +0000 (14:41 -0700)]
regcomp.c: Stop regexp-to-pv assignemnt from leaking
SvPV_set will just set SvPVX, allowing the existing value to leak.
This leak was caused by
f082678508, which allowed reg_temp_copy to
be called with an existing SV, but without modifying the contents of
reg_temp_copy to account.
Father Chrysostomos [Sun, 28 Oct 2012 14:00:23 +0000 (07:00 -0700)]
sv.c: Fix code-before-declarations
Father Chrysostomos [Sun, 28 Oct 2012 08:48:18 +0000 (01:48 -0700)]
Don’t crash with $tied[-1] when array is tied to non-obj
The code for checking to see whether $NEGATIVE_INDICES is defined in
the tie package was very fragile, and was repeated four times.
Father Chrysostomos [Sun, 28 Oct 2012 08:44:31 +0000 (01:44 -0700)]
Don’t skip tied EXISTS for negative array indices
This was broken in 5.14.0 for those cases where $NEGATIVE_INDICES is
not true:
sub TIEARRAY{bless[]};
sub FETCHSIZE { 50 }
sub EXISTS { print "does $_[1] exist?\n" }
tie @a, "";
exists $a[1];
exists $a[-1];
$NEGATIVE_INDICES=1;
exists $a[-1];
$ pbpaste|perl5.12.0
does 1 exist?
does 49 exist?
does -1 exist?
$ pbpaste|perl5.14.0
does 1 exist?
does -1 exist?
This was broken by
54a4274e3c.
Daniel Dragan [Sat, 27 Oct 2012 09:18:49 +0000 (05:18 -0400)]
yyerror->yyerror_pvn in toke.c:S_new_constant
Avoids a strlen.
Daniel Dragan [Sat, 27 Oct 2012 05:44:13 +0000 (01:44 -0400)]
rmv a sv_2mortal and unused var in toke.c:Perl_yyerror_pvn
newSVpvn_flags is capable of mortalizing already, use that, is_utf8 is used
only once, waste of an auto var stack slot to calculate it so early,
instead create the flags arg to newSVpvn_flags at the point of usage.
flags param of yyerror_pvn will always be on the C stack.
Father Chrysostomos [Sun, 28 Oct 2012 06:53:46 +0000 (23:53 -0700)]
sv.c: Allow blessed cows
There is no reason kine should not receive blessings, too.
Father Chrysostomos [Sun, 28 Oct 2012 06:51:42 +0000 (23:51 -0700)]
sv.c: Remove redundant Sv[INP]OK checks on fbm/regexps
These two code paths, sv_2iv and sv_2uv, used to be shared with gmag-
ical scalars in general, but now only apply to scalars that cannot
hold numeric values and are always SvPOK.
Father Chrysostomos [Sun, 28 Oct 2012 06:42:09 +0000 (23:42 -0700)]
sv.c: Remove redundant sv_force_normal calls from sv_2[iun]v
The previous commit made it possible to nummify a string whose SvLEN
is 0. So we don’t need to run shared hash key scalars through
sv_force_normal before nummifying them. We still need to run COW sca-
lars through sv_force_normal under PERL_OLD_COPY_ON_WRITE, as it uses
the IVX field for COW bookkeeping. For simplicity’s sake, I’m not
bothering to distinguish shared hash keys scalars from other COW sca-
lars under PERL_OLD_COPY_ON_WRITE.
Father Chrysostomos [Sun, 28 Oct 2012 06:30:28 +0000 (23:30 -0700)]
sv.c: !SvLEN does not mean undefined
There are various SvPOKp(sv) && SvLEN(sv) checks in numeric
conversion routines in sv.c, which date back to perl 1. (See
<http://perl5.git.perl.org/perl.git/blob/
8d063cd8450e59e:/str.c#l89>.)
Back then it did not matter, as str->len (later SvLEN) was always set
when there was a PV. It was not until perl 5.003_01 (
1edc1566d5) that
we got the SvLEN==0 mechanism for PVs not owned by the scalar. (I
don’t believe it was actually used till later, so when this became a
problem I don’t know--but that’s enough digging.)
A regexp returned by ${qr//} is POK but does not own its string. This
means that nummifying a regexp will result in a uninitialized warning.
The SvLEN check is redundant and problematic, so I am removing it.
(This also means I can remove the sv_force_normal calls in the next
commit, since shared hash key scalars, which also have SvLEN==0 will
no longer need it to pass the SvLEN checks.)
This does mean, however, that SVt_REGEXP can reach code paths that
expect to be able to use Sv[IN]VX (not valid for regexps), so I actu-
ally have to check that the type != SVt_REGEXP as well. We already
have code for handling fbm scalars (for which Sv[IN]VX fields are also
unusable), so we can send regexps through those paths.
Father Chrysostomos [Sun, 28 Oct 2012 01:18:35 +0000 (18:18 -0700)]
Stop regexp assignment from clobbering magic
$ perl5.10.0 -le '$1 = ${qr||}; print "ok"'
Modification of a read-only value attempted at -e line 1.
$ perl5.12.0 -le '$1 = ${qr||}; print "ok"'
ok
Wonderful!
It can also cause blessings to be lost, or so I thought:
sub curse {
for my $obj ( ${$_[0]} ) {
my $save = $obj;
$obj = ${qr||};
$obj = $save;
}
}
$y = bless \$x;
print $y, "\n"; # main=SCALAR(0x825b70)
curse $y;
print $y, "\n"; # Bus error
The OBJECT flag gets left on, but SvSTASH is null.
Commit
b9ad13acb set SvSTASH to null after copying the regexp struct.
Commit
703c388dc did the same with SvMAGIC. In both cases, this was
to avoid bugs involving magic and blessings being copied by = which
should not happen. But both changes caused other bugs.
Three months later,
6e1287864cd changed the order of the struct, such
that SvMAGIC and SvSTASH are no longer copied from the parent regexp,
rendering the aforementioned changes no longer necessary.
Father Chrysostomos [Sat, 27 Oct 2012 05:59:10 +0000 (22:59 -0700)]
Fix assertion failure with $float = $regexp assignment
Commit
b9ad13acb3 caused case SVt_REGEXP in sv_upgrade to fall
through to the assertions under case SVt_PVIV that are not relevant to
SVt_REGEXP.
We should really be setting the FAKE flag when actually making a sca-
lar a regexp, rather than in sv_upgrade. (I will probably need it
there in future commits, too, since it really should be possible for
SVt_PVLVs to hold regular expressions.)
Father Chrysostomos [Sat, 27 Oct 2012 05:42:51 +0000 (22:42 -0700)]
sv.c: No need to de-COW COWs on upgrade
I’ve taken the conservative approach of still de-COWing COWs for reg-
exps and anything above. I’m not confident that it would be safe
for those.
Father Chrysostomos [Sat, 27 Oct 2012 01:20:16 +0000 (18:20 -0700)]
Don’t sv_force_normal in mg.c:S_save_magic
This was added to make SvREADONLY_off safe. (I think read-only is
turned off during magic so the magic scalar itself can be set without
the sv_set* functions getting upset.) Since SvREADONLY doesn’t mean
read-only for COWs, we don’t actually need to do sv_force_normal, but
can simply skip SvREADONLY_off for COWs.
By leaving it to sv_set* functions to do sv_force_normal, we avoid
having to copy the string buffer if it is just going to be thrown away
anyway. S_save_magic can’t know whether the scalar will actually be
overwritten, so it has to copy the buffer.
Craig A. Berry [Sat, 27 Oct 2012 18:23:46 +0000 (13:23 -0500)]
Workaround for VAX compiler optimizer bug in Digest::SHA.
This was [perl #85932] and has been forwarded upstream as
[rt.cpan.org #80157]. The code is a near-verbatim copy of how
the same problem has been solved in Digest::MD5 since 2001.
After this change, the core build is now working again (slowly!)
on OpenVMS VAX.
Craig A. Berry [Sat, 27 Oct 2012 12:53:10 +0000 (07:53 -0500)]
Fix use of non-existent bareword filehandle in t/TEST.
The refactoring done in
84650816efdc42d6 was incomplete and left
a couple of VMS-specific instances of RESULT while replacing all
other occurrences with $result.
Spotted by Jim Cromie.
Yves Orton [Fri, 26 Oct 2012 09:41:10 +0000 (11:41 +0200)]
silence warning in toke.c charnames support
In a C sprintf the expectation is that * parameters are type "int".
Reini Urban [Fri, 26 Oct 2012 21:44:57 +0000 (16:44 -0500)]
RT #115488 cxstack -1 at nested scan_heredoc
print <<E1 eq "foo\n\n" ? "ok 19\n" : "not ok 19\n";
@{[ <<E2 ]}
foo
E2
E1
checked for a CxTYPE(cx) == CXt_EVAL with an invalid cxstack -1, detected by asan.
Father Chrysostomos [Fri, 26 Oct 2012 23:23:11 +0000 (16:23 -0700)]
Allow cow with $magic = $hashkey
This was brought up in
<https://rt.perl.org/rt3/Ticket/Display.html?id=114820#txn-1165898>.
There is no reason we cannot assigned a shared hash key to a magical
scalar. The only destination flag in CAN_COW_MASK that makes COW
assignment questionable is SVf_BREAK. If such an assignment can hap-
pen (and I don’t believe it actually can), we will end up with unbal-
anced string table warnings. So change the CAN_COW_MASK check to an
SVf_BREAK check.
Craig A. Berry [Fri, 26 Oct 2012 15:01:13 +0000 (10:01 -0500)]
Make private variable static in regexec.c.
Craig A. Berry [Fri, 26 Oct 2012 21:33:26 +0000 (16:33 -0500)]
De-globalize regcomp inversion lists.
These lists are declared at file scope so will be global unless
made static. Actual use of these lists is via the various PL_xxx
global variables that point to them and that (except for
NonL1_Perl_Non_Final_Folds_invlist) are initialized in
Perl_re_op_compile in regcomp.c (but not in its incarnation as
ext/re/re_comp.c).
So change the lists to be static, and also skip declaring and
initializing them in ext/re/re_comp.c except for the one case that
is actually used in the extension version.
David Mitchell [Fri, 26 Oct 2012 09:14:06 +0000 (10:14 +0100)]
start to make ext/B work with 5.14.x
This fixes up a couple of test files to work under 5.14.x.
Lots more needs fixing up to make the whole distribution work
under 5.14.x, but I've lost the will for now.,
David Mitchell [Thu, 25 Oct 2012 13:48:06 +0000 (14:48 +0100)]
B.xs: move all B::*OP methods to B::OP::next
The previous commit moved all B::*OP methods capable of using direct field
offsets into next(). This commit moves the remaining B::*OP methods onto
it too (apart from oplist(), which returns a list rather than a single
item).
This simplifies the code, reduces the object size, and will also make it
easier to add an overlay facility, which will be coming soon.
David Mitchell [Wed, 24 Oct 2012 20:53:38 +0000 (21:53 +0100)]
B.xs: rationalise all methods aliased to next()
The code for B::OP::next() actually implements all B::*OP::* methods
that work by directly returning a field at a known offset in the OP
structure. Methods that can't do direct access usually have their own
body, rather than sharing with next().
However, whether a method can do direct field access is often dependent on
threading and/or perl version; so the same method is sometimes implemented
by next(), and sometimes by one or more individual method bodies. This is
all very confusing.
This commit takes all methods that *may* be implemented within next(),
and makes them always implemented by next(), using a table of data that
describes each method's offset, or -1 if it needs special handling.
This makes it a lot easier to see what's going on, and will also make it
easier to add an overlay facility, which will be coming soon.
The following commit will consolidate the remaining B::*OP methods within
next().
David Mitchell [Wed, 24 Oct 2012 17:10:22 +0000 (18:10 +0100)]
ext/B: remove pre-5.10 support
Expunge all conditional code that supports 5.6.x through 5.9.x,
making 5.10.0 the oldest release notionally supported.
This simplifies things considerably.
See p5p thread starting at
Message-ID: <
20121018122941.GE1908@iabyn.com>
David Mitchell [Wed, 24 Oct 2012 14:50:25 +0000 (15:50 +0100)]
make ext/B work with 5.16.x
The modules and tests under ext/B are notionally supposed to be
portable to older perl versions; in practice, extensive bit-rot
has occurred; often attempts have been made to add version-specific
code, which haven't actually been tested against older perl versions.
This commit does the minimum necessary to get the tests under ext/B
working with 5.16.0 and 5.16.1, threaded and unthreaded. It makes no
assertions as to whether it will work with the rest of the 5.16.x test
suite.
The side effects of this fix-up are:
* a facility has been added to OptreeCheck.pm (the test module that
checks the Concise output of various constructs) that allows
version-specific matching, e.g.:
# 4 <$> const(PV "junk") s* < 5.017002
# 4 <$> const(PV "junk") s*/FOLD >=5.017002
* OptreeCheck.pm's skip mechanism was found to be broken: checkOptree()
allows you to specify skipping, but only skipped one test, even though
a single call to checkOptree() could generate multiple lines of test
output.
Father Chrysostomos [Fri, 26 Oct 2012 15:32:23 +0000 (08:32 -0700)]
Better documentation for internal SV types
Yves Orton [Sun, 21 Oct 2012 13:04:47 +0000 (15:04 +0200)]
fix a compile warning and refactor some diagnostics in regexec.c
Yves Orton [Wed, 29 Aug 2012 07:38:07 +0000 (09:38 +0200)]
improve diagnostics of dbm_filter_util.pl by using Data::Dumper::qquote
We are testing things like packed strings. If we output the bytes raw
via diag we upset terminal layers expecting utf8, and generally output
unreadable garbage regardless. So use Data::Dumper::qqoute() to
preprocess diagnositics output.
Yves Orton [Wed, 29 Aug 2012 07:47:03 +0000 (09:47 +0200)]
Fix hash ordering dependency in DBM_Filter/t/int32.t
Under the filtering rules in place undef() and "" and 0 map to a
packed representation of 0.
In the StoreData call we pass in an anonymous perl (untied) hash
containing an "undef" key (which is actually treated as "") with a
value of undef(), along with a key 0 with a value of 1. This hash
will store both values as distinct key/value pairs.
When this hash is used to set up the *tied* %h1 hash both the "" key
and the 0 key will be converted into the same packed value "\0\0\0\0",
which means that whichever is last in the each() of the input hashref
will be the one stored in %h1.
This means the test breaks if we change the PL_hash_seed or the hash
implementation in such a way that "" comes before 0 in the keys of
the hash.
This patch changes the input test hash to verify that undef() => 1 is
treated the same as 0 => 1, and eliminates the potential key collision.
The reason this test was reliable in the wild is that pretty well all
perls use a 0 hash seed and the same hash function.
This test probably would have broken in other enviornments as well.
Yves Orton [Mon, 27 Aug 2012 06:52:51 +0000 (08:52 +0200)]
fix hash key ordering dependency in t/warnings.t
Hash seed randomization causes these tests to fail occasionally.
Yves Orton [Tue, 28 Aug 2012 07:23:15 +0000 (09:23 +0200)]
fix a hash order dependency in t/re_funcs_u.t
Yves Orton [Mon, 27 Aug 2012 06:53:54 +0000 (08:53 +0200)]
fix hash order dependency in ext/B/t/b.t
Hash seed randomization causes these tests to fail occasionally.
Yves Orton [Tue, 28 Aug 2012 08:15:40 +0000 (10:15 +0200)]
fix a very subtle hash ordering dependency in op/smartkve.t
Currently our hash implementation is order dependent on insertion.
When two keys collide and have to be stored in the same bucket the
order in which they are inserted into the hash will govern the order
in which they are fetched out by things like keys() and values().
This means that a copy of such a hash may be different. It is possible
this can be fixed with a low cost, but until then you cannot rely on
two hashes with the same keys having the same ordering of those keys
Depending on the hash algorithm and the seed values used this test
would fail. By changing it so there is one initial hash and then all
tests are done on copies of that hash we avoid the problem.
Yves Orton [Mon, 27 Aug 2012 06:51:39 +0000 (08:51 +0200)]
fix hash key ordering dependency in t/op/defins.t
These tests break if we change the hash function or
randomly initialize the hash seed.
Father Chrysostomos [Fri, 26 Oct 2012 01:11:11 +0000 (18:11 -0700)]
perl5180delta: List mods broken by padlist changes
that have at least five dependents.
Reini Urban [Sun, 21 Oct 2012 23:56:11 +0000 (18:56 -0500)]
Re-enable static op allocation with obslab
obslab and the removal of the op_latefree logic, which allowed static
ops, removed support for the compiler modules, which allocates ops statically.
Add an op_static flag to replace the old latefree(d) op_free logic.
Daniel Dragan [Wed, 24 Oct 2012 20:15:51 +0000 (16:15 -0400)]
optimize memory wrap croaks, often used in MEM_WRAP_CHECK
Most perls are built with PERL_MALLOC_WRAP. This causes MEM_WRAP_CHECK
macro to perform some checks on the requested allocation size in macro
Newx. The checks are performed at the caller, not in the callee (for me
on Win32 perl the callee in Newx is Perl_safesysmalloc) of Newx.
If the check fails a "Perl_croak_nocontext("%s",PL_memory_wrap)" is done.
In x86 machine code,
"if(bad_alloc) Perl_croak_nocontext("%s",PL_memory_wrap); will be written
as "cond jmp ahead ~15 bytes", "push const pointer", "push const pointer",
"call const pointer". For each Newx where the allocation amount was not a
constant (constant folding would remove the croak memory wrap branch
compleatly), the branch takes 15-19 bytes depending on x86 compiler. There
are about 80 Newx'es in the interp (win32 dynamic linking perl) that do
the memory wrap check and have a
"Perl_croak_nocontext("%s",PL_memory_wrap)" in them after all optimizations
by the compiler.
This patch reduces the memory wrap branch from 15-19 to
5 bytes on x86. Since croak_memory_wrap is a static and a noreturn, a
compiler with IPO may optimize the whole branch to "cond jmp 32 bits
relative" at each callsite. A less optimal complier may do "cond jmp 8 bits
relative (jump past the "call S_croak_memory_wrap" instruction),
then "call S_croak_memory_wrap". Both ways are better than the current
situation. The reason why croak_memory_wrap is a static and not an export
is that the compiler has more opportunity to optimize/reduce the impact of
the memory wrap branch at the call site if the target is in the same image
rather than in a different image, which would require using the platform
specific dynamic linking mechanism/export table/etc, which often requires
a new stack frame per ABI of the platform. If a dynamic linked XS module
does not use S_croak_memory_wrap it will be removed from the image by the
C compiler. If it is included in the XS image, it is a very small block
of code and a 3 byte string litteral. A CPU cache line is typically
32 or 64 bytes and a memory read is typically 16. Cutting the
instructions by 10 to 16 bytes out of "hot code" (10 of the ~80 call
sites are pp_*) is a worthy goal. In a few places the memory wrap croak is
used explictly, not from a MEM_WRAP_CHECK, this patch converts those to use
the static. If PERL_MALLOC_WRAP is undef, there are still a couple uses of
croak memory wrap, so do not keep S_croak_memory_wrap in a ifdef
PERL_MALLOC_WRAP. Also see
http://www.nntp.perl.org/group/perl.perl5.porters/2012/10/msg194383.html
and [perl #115456].
Father Chrysostomos [Thu, 25 Oct 2012 20:00:55 +0000 (13:00 -0700)]
[perl #115440] Fix various leaks with fatal FETCH
Various pieces of code were creating an SV and then assigning to it
from a value that might be magical. If the source scalar is magical,
it could die when magic is called, leaking the scalar that would have
been assigned to.
So we call get-magic before creating the new scalar, and then use a
non-magical assignment.
Also, anonhash and anonlist were doing nothing to protect the aggre-
gate if an argument should die on FETCH, resulting in a leak.
Ruslan Zakirov [Tue, 23 Oct 2012 16:04:37 +0000 (20:04 +0400)]
test memory leaks around magic get dieing
Leaks happen when newSV is allocated, but then
copy operaton dies in get magic leaving not freed
scalar around.
Most of new tests check leaks in code path executing
sv_mortalcopy which has such problem. Two cases has
the same pattern, but don't use sv_mortalcopy. Can be
found with the following command:
grep -n -A3 'newSV\>' *.c | grep -B3 sv_set
Karl Williamson [Thu, 25 Oct 2012 03:35:06 +0000 (21:35 -0600)]
toke.c: Avoid unnecessary uninitialized value msgs
\N{uknown character} is now a syntax error. It also generates a "Use of
uninitialized value" message that is redundant (and confusing) with the
unknown character message.
Karl Williamson [Wed, 24 Oct 2012 23:17:35 +0000 (17:17 -0600)]
charnames pod: Note that \N{} doesn't accept interpolated $vars
Ricardo Signes [Wed, 24 Oct 2012 22:51:06 +0000 (18:51 -0400)]
add missing closing parens to documentation
reported by Joaquin Ferrero in [perl #115460] and [perl #115458]
Karl Williamson [Wed, 24 Oct 2012 16:38:53 +0000 (10:38 -0600)]
test.pl: Fix description of how PREFIX works
Karl Williamson [Wed, 24 Oct 2012 16:32:26 +0000 (10:32 -0600)]
charnames.t: Add names for some tests
Karl Williamson [Wed, 24 Oct 2012 16:28:55 +0000 (10:28 -0600)]
test.pl: Allow NAME to be used with --FILE--
Prior to this patch the --FILE-- feature of test.pl could not be used on
tests that had a name. This is because --FILE-- is expecting a \n
before it, and NAME strips that off. This commit just makes the \n
optional.
Karl Williamson [Wed, 24 Oct 2012 16:02:54 +0000 (10:02 -0600)]
Make \N{unknown char} a syntax error
Previously, it was a warning with the REPLACEMENT CHARACTER substituted.
Unicode recommends that it be a syntax error, and any code that used
this had to be buggy since the REPLACEMENT CHARACTER has no other use in
Unicode.
Karl Williamson [Wed, 24 Oct 2012 15:30:33 +0000 (09:30 -0600)]
charnames.t: Fix erroneous interpolation of \N{}
This is supposed to print as-is, not interpolate.
Karl Williamson [Sun, 21 Oct 2012 17:10:32 +0000 (11:10 -0600)]
toke.c: Indent properly
Karl Williamson [Wed, 24 Oct 2012 03:39:09 +0000 (21:39 -0600)]
regcomp.c: White-space only; no code changes
This outdents code that was previously in an inner block
Karl Williamson [Wed, 24 Oct 2012 03:36:00 +0000 (21:36 -0600)]
regcomp.c: Remove dead code
Multi-char folds in bracketed character classes are now handled
differently than before commit
9d53c4576e551530162e7cd79ab72ed81b1e1a0f.
This code was overlooked in the clean up commit
e0193e472b025d41438e251be622aad42c9af9cc.
Karl Williamson [Wed, 24 Oct 2012 03:32:46 +0000 (21:32 -0600)]
regexec.c: Nano optimization: reverse 'if' order
It's more likely that the code point is invariant than that the target
is UTF-8, so test it first. This works even if the target isn't UTF-8.
Karl Williamson [Wed, 24 Oct 2012 03:25:30 +0000 (21:25 -0600)]
regexec.c: Remove dead code
An ANYOF node now no longer matches more than one character, since
9d53c4576e551530162e7cd79ab72ed81b1e1a0f. This code was overlooked in
the clean up commit
e0193e472b025d41438e251be622aad42c9af9cc. Since the
maximum match is 1 character, there is no point in passing a ptr that
was set to indicate how far the match went, so that parameter is
removed.
Karl Williamson [Wed, 24 Oct 2012 03:17:32 +0000 (21:17 -0600)]
regexec.c: Add/modify NOTREACHED asserts.
Karl Williamson [Wed, 24 Oct 2012 03:11:49 +0000 (21:11 -0600)]
regexec.c: Silence compiler warning
Some compilers are saying that c1 and c2 in this routine are not
initialized. I hand-verified that there really is no path through the
routine that uses either of these before setting, so this is a compiler
issue. Nonetheless, its best to silence the warning.
Jesse Luehrs [Tue, 23 Oct 2012 19:32:22 +0000 (14:32 -0500)]
fork() should return undef on failure, even in list context
Steffen Mueller [Mon, 22 Oct 2012 18:56:45 +0000 (20:56 +0200)]
Move more X's off the PUSH into EXTEND's
Generally speaking, if there's branching and each branch may contain a
varying number of PUSH operations, EXTENDing to the longest number of
items should generally be more efficient since the stack will eventually
grow to a relatively steady state, but the extra branching incurred by
the X's will cost every time.
Steffen Mueller [Mon, 22 Oct 2012 18:20:05 +0000 (20:20 +0200)]
ext/re: Optimize XPUSH's to EXTEND(), PUSH,...
Steffen Mueller [Mon, 22 Oct 2012 18:19:29 +0000 (20:19 +0200)]
Data::Dumper: Optimize XPUSH's to EXTEND(), PUSH,...
Steffen Mueller [Mon, 22 Oct 2012 13:47:04 +0000 (15:47 +0200)]
Pushing to stack after EXTEND doesn't require *X*PUSH
This just converts a few (m)?XPUSH(s|p|...) to (m)?PUSH(s|p|...) that
follow an explicit call to EXTEND.
Chris 'BinGOs' Williams [Sun, 21 Oct 2012 20:26:55 +0000 (21:26 +0100)]
Update CPANPLUS-Dist-Build to CPAN version 0.64
[DELTA]
0.64 Sat Oct 20 23:21:36 BST 2012
- Update Module::Install, no functional changes.
Chris 'BinGOs' Williams [Sun, 21 Oct 2012 20:24:14 +0000 (21:24 +0100)]
Synchronilise Module-CoreList in Maintainers.pl with CPAN
Craig A. Berry [Sun, 21 Oct 2012 14:33:53 +0000 (09:33 -0500)]
Proper skippage for sanity check in B::Deparse's core.t.
The check was being skipped when PERL_CORE was not defined or if
running on VMS, but the test count was not being updated, so it
would be wrong and cause the test to fail in these cases. So use
the standard skip facility to handle the skipping. Plus the test
works fine on VMS so go ahead and run it there as well.
Dagfinn Ilmari Mannsåker [Sun, 21 Oct 2012 03:43:20 +0000 (21:43 -0600)]
PATCH: [perl #115242] Test
This commit is a test for this bug.
Karl Williamson [Sat, 20 Oct 2012 22:55:26 +0000 (16:55 -0600)]
PATCH: [perl #115242]: /m and regex optimizer bug.
This commit turns off string length checking for /m. A string longer
than the calculated maximum can match under /m because, for example,
trailing new lines in it can come after the $ anchor.
A test for this condition is in the next commit.
Karl Williamson [Sat, 20 Oct 2012 22:22:05 +0000 (16:22 -0600)]
perlapi.pod: Clarify what a parameter means
Karl Williamson [Sat, 20 Oct 2012 22:19:49 +0000 (16:19 -0600)]
fold_grind.t: Set $| if debugging
Karl Williamson [Sat, 20 Oct 2012 22:18:32 +0000 (16:18 -0600)]
Add comment
Karl Williamson [Sat, 20 Oct 2012 22:15:48 +0000 (16:15 -0600)]
regen/mk_PL_charclass.pl: Move code to subroutine
This code is for just this property and was kludged in to be executed in
the general loop. It makes more sense to it to be in the subroutine
that handles the property that was just added in a prior commit.
It also changes the output slightly. The Latin1 sharp S isn't a
non-final fold, unlike what was said previously
Karl Williamson [Sat, 20 Oct 2012 22:00:21 +0000 (16:00 -0600)]
mktables: Don't generate no-longer needed tables
These internal tables were only used in regen code, and those have been
modified to not use them; so can be removed.
Karl Williamson [Sat, 20 Oct 2012 21:59:33 +0000 (15:59 -0600)]
regen/mk_PL_charclass.pl: White-space only; no code changes
Indent a newly formed block
Karl Williamson [Sat, 20 Oct 2012 21:25:18 +0000 (15:25 -0600)]
regen/mk_PL_charclass.pl: Generate our own tables for certain properties
The two affected inversion lists are used only in regen. It is wasteful
to have mktables generate these, as they aren't used elsewhere and just
take up disk space.
Karl Williamson [Sat, 20 Oct 2012 21:13:22 +0000 (15:13 -0600)]
regen/mk_invlists.pl: Generate our own tables for certain properties
The two affected inversion lists are used only in regen. It is wasteful
to have mktables generate these, as they aren't used elsewhere and just
take up disk space.
Craig A. Berry [Sat, 20 Oct 2012 22:27:01 +0000 (17:27 -0500)]
Exclude signal-related symbols on VMS without home-grown signalling.
config.h defines KILL_BY_SIGPRC when we are using our own signal
handling functions based on the sys$sigprc system service. But
when *not* doing that, and since using makedef.pl to define our
exportable symbols, we have not properly excluded the symbols
involved in the home-grown implementation. It's not a common
configuration, but since it's configurable, it should work, and
now it will.
Karl Williamson [Sat, 20 Oct 2012 19:10:41 +0000 (13:10 -0600)]
regen/regcharclass.pl: White space only; no code changes
Indent as a result of a new block
Karl Williamson [Sat, 20 Oct 2012 19:04:51 +0000 (13:04 -0600)]
regen/regcharclass.pl: Generate better code for some macros
This commit revamps the recently added function calculate_mask() to not
just work to give a single mask/compare value for its input and fail if
there are none, but to return a list of masks/compares when the set can
be split up into subsets that each can be represented by a mask/compare.
If this list taken as a whole yields fewer branches than what we get
otherwise, it is better code, and is used.
Said another way, what we had there before was all or nothing; this
works to improve things even if we can't do it all.
Karl Williamson [Sat, 20 Oct 2012 17:27:44 +0000 (11:27 -0600)]
regen/regcharclass.pl: White-space only; no code changes
This outdents some code that is no longer contained in a block, and
reflows the comments to 80 columns;
Karl Williamson [Sat, 20 Oct 2012 17:25:50 +0000 (11:25 -0600)]
regen/regcharclass.pl: Refactor a 'return' slightly
Instead of having a common return at the end of the function, this does
that same return at an intermediate point. The reason is that a future
commit will change the final return somewhat, and this seemed the
clearest way to go.
Craig A. Berry [Sat, 20 Oct 2012 03:16:17 +0000 (22:16 -0500)]
Tweak psect attributes in VMS initialization code.
noexe is the default and older compilers can't process it, so don't
explicitly specify it.
nopic is mandatory but non-default on VAX and is silently ignored
on Alpha and Itanium, so go ahead and specify it explicitly.
Craig A. Berry [Sat, 20 Oct 2012 02:38:04 +0000 (21:38 -0500)]
Shorten command length when generating linker options file.
The MINIPERL macro expands to 40 characters longer than the
MINIPERLQ macro in order to include Cwd, but Cwd isn't needed
when doing a simple one-liner, and the extra length can make us
exceed the command buffer on older VMS systems.