Father Chrysostomos [Wed, 27 Jun 2012 13:03:35 +0000 (06:03 -0700)]
diag.t: Don’t emit erroneous passing to-do tests
The __CATEGORIES__ entries by definition exist in perldiag.
Father Chrysostomos [Wed, 27 Jun 2012 13:01:29 +0000 (06:01 -0700)]
diagnostics.t: Tweak to get tests passing
I thought I had run it, hmm.
Father Chrysostomos [Wed, 27 Jun 2012 07:53:14 +0000 (00:53 -0700)]
[Merge] perldiag.pod severity/category cleanup
Father Chrysostomos [Wed, 27 Jun 2012 07:47:24 +0000 (00:47 -0700)]
diag.t: Check categories and severity
There was code already started for this, but never finished. This
commit makes it work for the most part.
This is not smart enough yet to understand nested categories. There
is a new exceptions list in the __DATA__ section for cases it can’t
handle, which are precious few.
Father Chrysostomos [Wed, 27 Jun 2012 07:45:34 +0000 (00:45 -0700)]
diag.t: Use correct variable
Multiline messages are listed in the __DATA__ section with spaces
instead of line breaks. When adding proper support for multiline err-
ors earlier, I made a slight mistake and caused multiline to-do mess-
ages to skip the to-do code path, because I was looking them up via
the name with line breaks ($name), rather than the name with line
breaks modified ($key).
This was causing it to fall down to the pass($key) below.
I want to add more tests down there, which will fail for multiline
to-do items.
Father Chrysostomos [Wed, 27 Jun 2012 07:41:49 +0000 (00:41 -0700)]
diag.t: Load test.pl in BEGIN
so that parentheses can be omitted in tests to be added.
Father Chrysostomos [Wed, 27 Jun 2012 07:36:24 +0000 (00:36 -0700)]
perldiag: ‘Unicode surrogate is illegal’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 07:35:11 +0000 (00:35 -0700)]
perldiag: ‘Code point is not Unicode’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 07:32:27 +0000 (00:32 -0700)]
perldiag: ‘Operation returns its argument’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 07:30:36 +0000 (00:30 -0700)]
perldiag: ‘...illegal for interchange’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 07:11:09 +0000 (00:11 -0700)]
perldiag: ‘UTF-16 surrogate U+%X’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 07:07:22 +0000 (00:07 -0700)]
perldiag: ‘Ambiguous use resolved as’ is always S
Father Chrysostomos [Wed, 27 Jun 2012 06:17:43 +0000 (23:17 -0700)]
perldiag: Add cat for Lost precision
Father Chrysostomos [Wed, 27 Jun 2012 06:14:20 +0000 (23:14 -0700)]
perldiag: ‘Regexp modifier after "-"’ should have no cat
Only warnings have categories.
Father Chrysostomos [Wed, 27 Jun 2012 06:09:37 +0000 (23:09 -0700)]
perldiag: ‘Opening fh also as dir’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 06:08:59 +0000 (23:08 -0700)]
perldiag: ‘Opening dirhandle also as file’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 06:07:54 +0000 (23:07 -0700)]
perldiag: Add cat for Duplicate modifier
Father Chrysostomos [Wed, 27 Jun 2012 05:35:51 +0000 (22:35 -0700)]
perldiag: ‘invalid option -D%c’ is a warning
Father Chrysostomos [Wed, 27 Jun 2012 05:34:28 +0000 (22:34 -0700)]
perldiag: Don’t use L<<>>
diagnostics.pm isn’t smart enough to handle it, and programming it
for one case seems like overkill.
Father Chrysostomos [Wed, 27 Jun 2012 05:34:20 +0000 (22:34 -0700)]
perldiag: Add cat for PERL_SIGNALS illegal
Father Chrysostomos [Wed, 27 Jun 2012 05:26:53 +0000 (22:26 -0700)]
perldiag: ‘Unbalanced context’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 05:26:15 +0000 (22:26 -0700)]
perldiag: ‘Unbalanced tmps’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 05:25:54 +0000 (22:25 -0700)]
perldiag: ‘Unbalanced saves’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 05:25:37 +0000 (22:25 -0700)]
perldiag: ‘Unbalanced scopes’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 05:24:56 +0000 (22:24 -0700)]
perldiag: ‘Variable is not imported’ is a default warning
Father Chrysostomos [Wed, 27 Jun 2012 05:22:41 +0000 (22:22 -0700)]
perldiag: ‘Scalars leaked’ is a warning
Father Chrysostomos [Wed, 27 Jun 2012 05:17:15 +0000 (22:17 -0700)]
perldiag: Add cat for ‘Attempt to set length of freed array’
Father Chrysostomos [Wed, 27 Jun 2012 04:31:53 +0000 (21:31 -0700)]
Squash repetitive code in pp.c:S_delete_local
The two branches of this were almost identical. Not only is it unnec-
essary to repeat the code, but it is error-prone, as bug fixes have to
be done in both branches.
Father Chrysostomos [Wed, 27 Jun 2012 03:21:29 +0000 (20:21 -0700)]
Null HeVAL and local delete → crash
The XS API allows hash entries to be created with null values. perl
itself uses these internally, too.
Though they should rarely be seen by Perl code, Perl ops should still
be able to handle them without crashing (by croaking). I fixed helem
and hslice in 5.16 (commit
746f6409), but missed localised deletions.
Father Chrysostomos [Wed, 27 Jun 2012 03:05:55 +0000 (20:05 -0700)]
Make pp.c:pp_srand slightly less repetitive
Father Chrysostomos [Wed, 27 Jun 2012 01:04:02 +0000 (18:04 -0700)]
Make srand treat "-1" as -1
It was returning U+FFFD for negative numbers, unless they
were strings.
The SvOK is to avoid double uninit warnings. The !SvIsUV is for
efficiency. We don’t need to coerce it to an NV if it is a UV
already, because we know it won’t pass that test.
Father Chrysostomos [Wed, 27 Jun 2012 00:41:40 +0000 (17:41 -0700)]
Make srand respect magic
It was returning U+FFFD for negative numbers, but only for non-magical
variables.
Darin McBride [Mon, 25 Jun 2012 23:36:32 +0000 (17:36 -0600)]
AIX Hints file change
Ok, I think my earlier problem came from not cleaning up properly between
multiple compiles. So I don't need to modify any test.
Proposed log entry:
When compiling many of the perl modules available from CPAN, an expectation
of C99 support seems implied by many authors. Unfortunately, xlC doesn't
do this by default, we have to add -qlanglvl=extc99 to the ccflags to get
this to work. And by the time we get to installing modules, this flag
is already set in Config.
So, add the flag unconditionally at the outset. Also, remove
-qlonglong and -qlanglvl=extended from the flags to silence the warning that
extc99 includes them and that xlC is going to ignore it, using extc99 instead.
Signed-off-by: H.Merijn Brand <h.m.brand@xs4all.nl>
Jesse Luehrs [Tue, 26 Jun 2012 21:47:51 +0000 (16:47 -0500)]
fix off-by-one when restoring hashes [perl #73972]
Storable tries to preallocate enough space for all of the elements it's
going to receive, both for efficiency reasons and because reallocation
triggers throwing away all of the placeholders in the hash (which are
used for restricted hashes) if the hash isn't already READONLY, and
since Storable rebuilds restricted hashes by first populating all of the
placeholders and then setting it READONLY at the end, this would break
things.
Unfortunately, it was allocating just slightly less than enough space -
hashes reallocate when they hit their limit, not when they exceed it,
and so if you tried to store a restricted hash with a number of keys
right on the boundary, it would trigger a reallocation and lose all of
the allowed keys that it had just stored. This fixes the issue by
allocating the correct amount of space to ensure that reallocation
doesn't happen.
Steve Hay [Tue, 26 Jun 2012 19:52:25 +0000 (20:52 +0100)]
Remove C++isms from doio.c
(Introduced by
c2fd40cb02 and not tolerated by VC++ compiling a C file.)
Jesse Luehrs [Tue, 26 Jun 2012 18:31:28 +0000 (13:31 -0500)]
perlrun typos from argrath@ub32.org [perl #113852]
Darin McBride [Tue, 26 Jun 2012 16:24:56 +0000 (09:24 -0700)]
[perl #112990] Simplify kill implementation and docs
Clean up kill implementation and clear up the docs in perlfunc to be less
ambiguous and encompass more of its behaviour.
a) kill -INT => $pid;
was surprisingly doing a "kill 0, $pid" instead of being the same as "kill -2,
$pid" (killing the process group for $pid with an interrupt). Now negative
signal names will be allowed and be the same as if the name was replaced with
the signal number it represents.
b) remove all calls to killpg() as killpg is defined in terms of kill anyway.
c) Clarify the use of signal names vs numbers in perlfunc so that using names
is not so well hidden, as well as explaining the usage of negative signal
numbers as well as negative process IDs.
Father Chrysostomos [Tue, 26 Jun 2012 16:23:06 +0000 (09:23 -0700)]
[perl #113812] Handle null CvOUTSIDE in cv_clone
Commit a0d2bbd stopped closures from hanging on to their
enclosing subs.
This means that the outer sub can be freed before the closure. Since
it is only the outer sub that references the closure prototype (in a
'&' entry in its pad), when the outer sub is freed the closure proto-
type goes with it. Now if that closure itself contains a closure
(more precisely, a closure prototype in its pad, referenced by an
anoncode op), that inner closure prototype’s CvOUTSIDE points to the
first closure prototype. When that first closure prototype is freed
the innermost closure prototype has its CvOUTSIDE set to whatever the
outermost sub’s CvOUTSIDE was set to, which could be null if it is
in its own file. So when the first closure is called, it passes to
cv_clone a closure prototype with no CvOUTSIDE.
cv_clone used to crash if CvOUTSIDE was null.
Example:
$ cat foo.pl
# the main CV of the file is the outer sub in this case
my $x
$first_closure = sub {
$inner_closure = sub { $x }
}
$ perl -e 'require "./foo.pl"; $first_closure->()'
Bus error
This commit makes it use find_runcv when CvOUTSIDE is null.
Father Chrysostomos [Tue, 26 Jun 2012 15:28:13 +0000 (08:28 -0700)]
perldiag: Two space after dots
Also rewrap for slightly better splain output
Jesse Luehrs [Tue, 26 Jun 2012 10:33:05 +0000 (05:33 -0500)]
fix test from previous commit
Jesse Luehrs [Tue, 26 Jun 2012 10:12:38 +0000 (05:12 -0500)]
fix memory leak in @INC filters that die [perl #92252]
Nicholas Clark [Tue, 5 Jun 2012 14:53:43 +0000 (16:53 +0200)]
In t/op/stat.t, ensure that the output of ls is not localised.
t/op/stat.t parses the output of ls, and will get confused if localisation
happens to change the header line containing the total to a word starting
with the letter c. This fixes the build issue on a Czech locale reported
in http://perlmonks.org/?node=973257 and RT #113472.
Tony Cook [Tue, 26 Jun 2012 07:56:44 +0000 (17:56 +1000)]
[perl #45173] port the test to MSWin32
Jesse Luehrs [Tue, 26 Jun 2012 08:13:20 +0000 (03:13 -0500)]
add comment about the atof change [perl #109318]
Jesse Luehrs [Tue, 26 Jun 2012 07:16:57 +0000 (02:16 -0500)]
warn on -i with no input files given [perl #113410]
Jesse Luehrs [Tue, 26 Jun 2012 06:35:31 +0000 (01:35 -0500)]
fix c++ compilation
Jesse Luehrs [Tue, 26 Jun 2012 05:13:54 +0000 (00:13 -0500)]
use a less broken test for locale radix in atof [perl #109318]
Jesse Luehrs [Tue, 26 Jun 2012 05:33:40 +0000 (00:33 -0500)]
fix some language/country codes
Father Chrysostomos [Tue, 26 Jun 2012 05:12:39 +0000 (22:12 -0700)]
diag.t: Two to-done entries I missed
Father Chrysostomos [Tue, 26 Jun 2012 05:10:12 +0000 (22:10 -0700)]
perldelta: make podcheck happy
Father Chrysostomos [Tue, 26 Jun 2012 05:08:46 +0000 (22:08 -0700)]
perldelta for #73690, scalar(write)
Father Chrysostomos [Tue, 26 Jun 2012 04:32:25 +0000 (21:32 -0700)]
perldelta for #112966/
2c5f48c251e
Father Chrysostomos [Tue, 26 Jun 2012 03:47:06 +0000 (20:47 -0700)]
perldelta for #113798, glob and PATH
Father Chrysostomos [Tue, 26 Jun 2012 03:43:02 +0000 (20:43 -0700)]
perldelta for #112962, reg_temp_copy and null offs
Father Chrysostomos [Tue, 26 Jun 2012 03:40:43 +0000 (20:40 -0700)]
perldelta: Oops, duplicate entry
Father Chrysostomos [Tue, 26 Jun 2012 03:39:39 +0000 (20:39 -0700)]
perldelta for #101666, regexp error error
Father Chrysostomos [Tue, 26 Jun 2012 03:39:27 +0000 (20:39 -0700)]
perldelta for #112826, Tie::StdHandle and read offset
Father Chrysostomos [Tue, 26 Jun 2012 03:36:04 +0000 (20:36 -0700)]
perldelta for #113764, duping scalar handles
Father Chrysostomos [Tue, 26 Jun 2012 03:33:40 +0000 (20:33 -0700)]
perldelta for #89544, nested closure leak
Father Chrysostomos [Tue, 26 Jun 2012 03:30:57 +0000 (20:30 -0700)]
perldelta for #113730, do and $@
Father Chrysostomos [Tue, 26 Jun 2012 03:29:22 +0000 (20:29 -0700)]
pp_sys.c: Squelch compiler warning
Father Chrysostomos [Tue, 26 Jun 2012 03:28:46 +0000 (20:28 -0700)]
perldelta for #113712, stubs after errors
Father Chrysostomos [Mon, 25 Jun 2012 00:58:54 +0000 (17:58 -0700)]
diag.t: Tolerate trailing space
Father Chrysostomos [Mon, 25 Jun 2012 00:53:27 +0000 (17:53 -0700)]
perldiag: Correct various regexp msgs
Many of these are listed in a slightly different way from what perl actually emits. E.g., this entry:
Sequence (?#... not terminated in regex; marked by <-- HERE in m/%s/
is incorrect, as the actual message is:
$ perl -e '/(?#/'
Sequence (?#... not terminated in regex m/(?#/ at -e line 1.
Father Chrysostomos [Mon, 25 Jun 2012 00:24:49 +0000 (17:24 -0700)]
diag.t: Support regcomp’s FAIL macros
Jesse Luehrs [Tue, 26 Jun 2012 01:43:38 +0000 (20:43 -0500)]
fix stack handling in write() [perl #73690]
I'm not sure about that POPs at the beginning of pp_leavewrite, but it
seems to work. As far as I can tell, executing the format always leaves
an extra value on the stack, but I'm not sure where that happens
exactly, so I'm just fixing it up there.
Steve Hay [Mon, 25 Jun 2012 17:03:18 +0000 (18:03 +0100)]
Import CCHOME from the environment when building with GCC on Windows
Using the environment for such things is preferable to always having to
specify a CCHOME= command-line argument, and is akin to setting various
environment variables via vcvars32.bat for VC++ builds.
Also remove a blank line to make it more obvious that copying DLLs to the
t/ folder is part of the test-prep target. (This is one thing which
doesn't work if CCHOME has not been set and MinGW isn't installed in its
default location.)
Vincent Pit [Mon, 25 Jun 2012 14:07:25 +0000 (16:07 +0200)]
Fix tab completion of module names in the debugger
Also broken by 'use strict'.
Vincent Pit [Mon, 25 Jun 2012 13:56:27 +0000 (15:56 +0200)]
Fix restarting a debugging session opened for a one-liner
It was broken by 'use strict' as well.
Vincent Pit [Mon, 25 Jun 2012 13:49:06 +0000 (15:49 +0200)]
Fix (and test) module listing in the debugger
This was also broken when 'use strict' was added.
Vincent Pit [Mon, 25 Jun 2012 13:06:49 +0000 (15:06 +0200)]
Fix (and test) breakpoints on subroutines
They were broken since 'use strict' was added to the debugger.
Jesse Luehrs [Mon, 25 Jun 2012 05:53:55 +0000 (00:53 -0500)]
perldelta for 8b8c1fb
Jesse Luehrs [Mon, 25 Jun 2012 05:39:57 +0000 (00:39 -0500)]
allow "my ()" [perl #113554]
it looks like this was an unrelated bugfix during development of mad,
but which was merged into core with #ifdef PERL_MAD just because it was
a change from the existing perl code... i don't see any reason why this
should have been ifdef'd out to begin with
Jesse Luehrs [Mon, 25 Jun 2012 00:00:54 +0000 (19:00 -0500)]
perldelta for 100c03a, f914a68, and 61a3fb8
Father Chrysostomos [Sun, 24 Jun 2012 21:14:01 +0000 (14:14 -0700)]
Add E. Choroba to AUTHORS
E. Choroba [Sun, 24 Jun 2012 20:28:29 +0000 (22:28 +0200)]
Fix Exporter.pm documentation: Notes are not code.
Father Chrysostomos [Sun, 24 Jun 2012 21:05:18 +0000 (14:05 -0700)]
perldiag: Document ‘Integer overflow in srand’
Father Chrysostomos [Sun, 24 Jun 2012 21:00:03 +0000 (14:00 -0700)]
diag.t: Cull exceptions
Father Chrysostomos [Sun, 24 Jun 2012 20:52:09 +0000 (13:52 -0700)]
diag.t: Don’t skip lines starting with spaces
Father Chrysostomos [Sun, 24 Jun 2012 20:42:59 +0000 (13:42 -0700)]
perldiag: Remove initial %s: from regexp messages
Commit
b45f050a81 moved the initial /%s/: and changed it to a final
‘in regex m/%s/’, but some perldiag entries were not updated.
Father Chrysostomos [Sun, 24 Jun 2012 20:26:48 +0000 (13:26 -0700)]
perldiag: document ‘Invalid negative number’ in chr
Father Chrysostomos [Sun, 24 Jun 2012 20:22:59 +0000 (13:22 -0700)]
Unbreak diag.t
In speeding in up in commit
de53a0eaa2de, I actually broke it such
that it only looked for DIE.
Father Chrysostomos [Sun, 24 Jun 2012 19:22:13 +0000 (12:22 -0700)]
perldiag: Help splain find version overflow msg
There are two variants, and only one was listed here.
Father Chrysostomos [Sun, 24 Jun 2012 19:20:21 +0000 (12:20 -0700)]
perldiag: ‘Integer overflow in %s number’ is S
Father Chrysostomos [Sun, 24 Jun 2012 18:22:10 +0000 (11:22 -0700)]
perldiag: ‘Integer overflow in version’ is W
Father Chrysostomos [Sun, 24 Jun 2012 18:20:47 +0000 (11:20 -0700)]
perldiag: typo
Father Chrysostomos [Sun, 24 Jun 2012 18:14:29 +0000 (11:14 -0700)]
Don’t crash with ()=&CORE::srand
Ops that don’t allow arbitrarily long argument lists are flagged at compile-time with the number of children.
Coresubs have to call the same op with differing numbers of arguments,
so they push nulls on to the stack to make up the number of items the
pp function expects to pop.
Commit
f914a6829 stopped popping the null of the stack.
Before:
$ ./perl -le 'print &CORE::srand'
1240765685
After:
$ ./perl -le 'print &CORE::srand'
Bus error
List assignment does the same thing, and makes it easier to
write a test.
Jesse Luehrs [Sun, 24 Jun 2012 08:47:39 +0000 (03:47 -0500)]
add warning for negative argument to chr() [perl #83048]
Jesse Luehrs [Sun, 24 Jun 2012 08:10:05 +0000 (03:10 -0500)]
warn when srand overflows [perl #40605]
Father Chrysostomos [Sun, 24 Jun 2012 07:10:55 +0000 (00:10 -0700)]
[perl #113796] lib/File/stat.t failures when -l $^X
It seems that gcc produces this:
$ ./perl -Ilib -e 'warn $^X'
/Users/sprout/Perl/perl.git/perl at -e line 1.
while g++ produces this:
$ ./perl -Ilib -e 'warn $^X'
./perl at -e line 1.
(I may be misdiagnosing this, but I have two blead builds that give
different values for $^X.)
The script is using $^X for testing, which is why it could produce
different results.
In any case, this produces the same output for both compilers:
use File::stat;
my $stat = File::stat::stat('./perl');
warn eval '-l $stat';
warn eval '-l "./perl"';
__END__
Warning: something's wrong at - line 3.
1 at - line 4.
The test is wrong, as -l _ will fail after a stat. You have to do
lstat for -l _ to work. Similarly, -l $stat_obj should only return
true if the $stat_obj was returned by File::stat::lstat.
This commit adjusts the test accordingly.
Father Chrysostomos [Sun, 24 Jun 2012 01:29:03 +0000 (18:29 -0700)]
perlfunc: (caller)[9] is ${^WARNING_BITS}
Technically, (caller)[9] returns ${^WARNING_BITS} if defined, or, if
${^WARNING_BITS} is not defined, whatever ‘use warnings’ or ‘no
warnings’ would set ${^WARNING_BITS} to, depending on the value of
$^W. But that is subject to change, and is a bit of a mouthful.
Jesse Luehrs [Sun, 24 Jun 2012 06:23:49 +0000 (01:23 -0500)]
don't let arriving signals reset $@ [perl #45173]
since signals can arrive at any point, clearing $@ isn't a safe
thing to do
Tony Cook [Sat, 23 Jun 2012 03:41:19 +0000 (13:41 +1000)]
[perl #64772] make the read-only-sync test a bit more portable
- Win32 simply doesn't support fsync, and %Config reflects that
- sync() on a file seems more portable to me than on a directory, and
the first is enough to test the code path we want to test.
- AIX is documented to fail fsync() on a read-only handle, so skip
there
Father Chrysostomos [Sat, 23 Jun 2012 20:59:34 +0000 (13:59 -0700)]
[perl #112966] Crash on delete local; other local bugs
Commit
bee7c5743fa appears to have fixed this. But what it does is
barely significant:
diff --git a/sv.c b/sv.c
index b96f7c1..a4994f5 100644
--- a/sv.c
+++ b/sv.c
@@ -9525,6 +9525,11 @@ Perl_sv_bless(pTHX_ SV *const sv, HV *const stash)
SvUPGRADE(tmpRef, SVt_PVMG);
SvSTASH_set(tmpRef, MUTABLE_HV(SvREFCNT_inc_simple(stash)));
+ if (Gv_AMG(stash))
+ SvAMAGIC_on(sv);
+ else
+ (void)SvAMAGIC_off(sv);
+
if(SvSMAGICAL(tmpRef))
if(mg_find(tmpRef, PERL_MAGIC_ext) || mg_find(tmpRef, PERL_MAGIC_uvar))
mg_set(tmpRef);
The crash can still be triggered another way. Instead of a blessing,
we need to modify a method (to turn on the potentially-overloaded
flag) and then use an operator that respects overloading. This exam-
ple crashes before and after
bee7c5743fa:
eval 'sub Sample::foo {}';
"".bless {},'Sample';
delete local $Sample::{ '()' };
It is the recalculation of overload caches before a localised deletion
that causes the crash. And it only happens when the '()' key does
not exist.
Actually, it turns out that S_delete_local doesn’t behave correctly
for rmagical aggregates, except for %ENV:
$ ./perl -Ilib -MDevel::Peek -e 'delete local $ISA[0]'
Bus error
$ ./perl -XIlib -MDevel::Peek -e '??; delete local $::{foo}'
Bus error
It’s this line, which occurs twice in pp.c:S_do_delete_local, which
is at fault:
const bool can_preserve = SvCANEXISTDELETE(osv)
|| mg_find((const SV *)osv, PERL_MAGIC_env);
When can_preserve is true, the ‘preeminent’ variable is set based on
whether the element exists. Otherwise it is set to true.
Why the term ‘preeminent’ was chosen I don’t know, but in this case it
means that the element already exists, so it has to be restored after-
wards. We can’t just do save_delete.
The code for saving a hash element assumes it is non-null, and crashes
otherwise.
The logic for setting can_preserve is wrong. SvCANEXISTDELETE returns
true for non-magical variables and for variables with those tie meth-
ods implemented. For magical variables that are not tied, it returns
the wrong answer. PERL_MAGIC_env seems to have been added as an
exception, to keep it working. But other magical aggregates were not
accounted for.
This logic was copied from other functions (aslice, hslice, etc.),
which are similarly buggy, but they don’t crash:
$ ./perl -Ilib -le ' { local $::{foo} } print exists $::{foo}'
$ ./perl -Ilib -le 'm??; { local $::{foo} } print exists $::{foo}'
1
In all these cases, it is SvCANEXISTDELETE that is buggy. So this
commit fixes it and adds tests for all the code paths that use it.
Now no exception needs to be made for PERL_MAGIC_env.
Jesse Luehrs [Sat, 23 Jun 2012 19:45:09 +0000 (14:45 -0500)]
clarify the $hints and $bitmask values in caller()
Jesse Luehrs [Sat, 23 Jun 2012 18:01:59 +0000 (13:01 -0500)]
perldelta for 61c2a93 and 68b4061
Jesse Luehrs [Fri, 22 Jun 2012 18:56:15 +0000 (13:56 -0500)]
also make sure ->isa works on undeclared packages
Jesse Luehrs [Fri, 22 Jun 2012 18:34:49 +0000 (13:34 -0500)]
all packages can do methods in UNIVERSAL [perl #47113]
Foo->can("can") should be true even if "package Foo" hasn't been seen
yet (obviously, since that method call doesn't die with a method not
found error).
Father Chrysostomos [Sat, 23 Jun 2012 16:36:51 +0000 (09:36 -0700)]
regen pod issues
Father Chrysostomos [Sat, 23 Jun 2012 16:34:26 +0000 (09:34 -0700)]
[perl #113798] Don’t hide PATH from perlglob
To fix another bug, miniperl was changed to clear out %ENV before
shelling out to call the underlying glob program (csh on Unix;
perlglob on Windows), in commit
a3342be368.
That proved slightly problematic, as it stopped <~> from working on
Unix, so commit
93b2dae1 changed it to preserve just $ENV{HOME}.
That turns out not to have been enough. For some compilers, Win-
dows needs PATH preserved for perlglob to find certain DLLs it
needs to load.