Paul "LeoNerd" Evans [Tue, 4 Jan 2011 17:04:47 +0000 (17:04 +0000)]
Support emulation of AI_NUMERICSERV even though it's not strictly RFC 2553, because most OSes support it anyway, and it's easy to do
brian d foy [Wed, 5 Jan 2011 05:00:30 +0000 (23:00 -0600)]
perlfaq1: less "more faster"
Karl Williamson [Wed, 5 Jan 2011 00:01:09 +0000 (17:01 -0700)]
mktables: Small performance enhancement
Instead of generating a separate list of missing ranges in a table, it's
less work to use the fill-in the missing ranges feature.
Karl Williamson [Tue, 4 Jan 2011 23:59:45 +0000 (16:59 -0700)]
mktables: white space-only change
A previous commit removed the 'if' around this code, so can outdent it.
Karl Williamson [Tue, 4 Jan 2011 23:56:55 +0000 (16:56 -0700)]
mktables: remove unneeded test
When I was originally writing this code, I thought it was faster to do
an extra substr to avoid a regex match. I have since found out that is
not true.
Karl Williamson [Tue, 4 Jan 2011 23:56:31 +0000 (16:56 -0700)]
mktables: typos in comments
Karl Williamson [Tue, 4 Jan 2011 23:53:13 +0000 (16:53 -0700)]
mktables: improve debug statement
Don't output an empty list.
Karl Williamson [Tue, 4 Jan 2011 23:50:04 +0000 (16:50 -0700)]
mktables: Add error check
This makes sure that in certain cases when adding to a table that we
know should not have elements in the area that it actually does.
Paul "LeoNerd" Evans [Tue, 4 Jan 2011 16:38:08 +0000 (16:38 +0000)]
Use Zero() rather than rely on C99 struct initialiser behaviour
Signed-off-by: Ævar Arnfjörð Bjarmason <avar@cpan.org>
Nicholas Clark [Tue, 4 Jan 2011 16:59:46 +0000 (16:59 +0000)]
Convert tied WRITE to using S_tied_handle_method()
Nicholas Clark [Tue, 4 Jan 2011 15:31:03 +0000 (15:31 +0000)]
Convert tied READ to using S_tied_handle_method()
Nicholas Clark [Tue, 4 Jan 2011 12:23:30 +0000 (12:23 +0000)]
Convert tied PRINTF to using S_tied_handle_method()
Nicholas Clark [Tue, 4 Jan 2011 11:54:25 +0000 (11:54 +0000)]
Convert tied OPEN to using S_tied_handle_method()
Add a new flag ARGUMENTS_ON_STACK to S_tied_handle_method(), which methods
which take a list (OPEN, PRINTF, READ and WRITE) will need.
Nicholas Clark [Tue, 4 Jan 2011 11:36:48 +0000 (11:36 +0000)]
In pp_sys.c, move S_tied_handle_method() before pp_open.
David Mitchell [Tue, 4 Jan 2011 14:36:37 +0000 (14:36 +0000)]
stash.t: make TODO tests only TODO under threads
.. since they only fail under threads.
Nicholas Clark [Tue, 4 Jan 2011 10:13:42 +0000 (10:13 +0000)]
Remove a long-vestigial PUTBACK from the tied OPEN code in pp_send.
Commit
4592e6caefc41a75, from 1999, added the PUTBACK as part of the initial
implementation of tied OPEN. At this time, pp_open (as then was) popped the
arguments off the stack (to local variables), so the new tie code pushed them
back onto the stack, before calling the method. The stack position was
manipulated, hence the local variable sp had potentially diverged from the
global PL_stack_sp, hence the latter needed updating.
Commit
a567e93b903bc984, from 2001, which added support for
open $fh, '|-', @array; refactored pp_open to avoid all stack manipulation.
Hence sp (the local variable) never changes along this code path, hence it
never needs to be written back to PL_stack_sp.
Nicholas Clark [Tue, 4 Jan 2011 09:48:27 +0000 (09:48 +0000)]
Converge the implementation of tied OPEN with PRINTF, READ and WRITE.
Consistency will make it easier to refactor.
Reini Urban [Tue, 14 Sep 2010 16:04:22 +0000 (18:04 +0200)]
build man pages on cygwin too
Reini Urban [Tue, 14 Sep 2010 15:54:15 +0000 (17:54 +0200)]
Improve cygwin rebase behaviour
If a dll is updated on cygwin reuse the old imagebase address.
This solves most rebase errors, esp when updating on core dll's.
See http://www.tishler.net/jason/software/rebase/rebase-2.4.2.README
Nicholas Clark [Tue, 4 Jan 2011 08:29:35 +0000 (08:29 +0000)]
Refactor filenames.t to use unlike() and cmp_ok() for better diagnostics.
Also, use regexs in place of the loops over @dont and @more_dont, the lists
of Win32 (was DOS (was CP/M)) reserved device filenames and of disallowed
characters. This makes the intent clearer, at the cost of some runtime.
Jan Dubois [Tue, 4 Jan 2011 02:40:17 +0000 (18:40 -0800)]
Don't install Cygwin code into other platforms
DynaLoader already has its own preprocessor to filter out code that
is only relevant on other platforms, so use it for the change from
commit
cc7e77fd
Jesse Vincent [Mon, 3 Jan 2011 17:24:05 +0000 (01:24 +0800)]
update t/porting/filenames.t to check for path components contaning two
"." characters or with lenghts exceeding 32 characters.
David Mitchell [Mon, 3 Jan 2011 16:49:02 +0000 (16:49 +0000)]
regen/overload.pl: add comments to overload.[ch]
In the auto-generated overload.c and overload.h files, add a comment to
each line mapping from the method name to the enumeration name and
vice-versa.
Jesse Vincent [Mon, 3 Jan 2011 16:41:22 +0000 (00:41 +0800)]
Actuall excise 'Apollo DomainOS' support. We officially killed it in 5.11.0. It
hadn't worked for years before that.
Zsbán Ambrus [Mon, 3 Jan 2011 16:41:35 +0000 (08:41 -0800)]
Add link for v5.8.5-RC1 announcement
Reini Urban [Tue, 14 Sep 2010 15:48:32 +0000 (17:48 +0200)]
CYG14 Dynaloader without USEIMPORTLIB, and search cyg prefix
part1: Support the standard cyg dll prefix, which is e.g. needed for FFI's.
Ctypes and C::DynaLib use DynaLoader to find dlls.
part2: With -DUSEIMPORTLIB DynaLoader symbols link against the prefixed
symbol names for the .dll.a importlib, but we need to link against the
symbols directly. We don't link Dynaloader against libperl.dll.a.
Otherwise:
$ g++-4 -o cygperl5_13_4.dll --shared perlsrc.o cygwin.o DynaLoader.o -ldl -lcrypt
Creating library file: libperl.dll.a
DynaLoader.o: In function `XS_DynaLoader_dl_undef_symbols':
ext/DynaLoader/DynaLoader.c:346: undefined reference to `__imp__PL_stack_sp'
ext/DynaLoader/DynaLoader.c:346: undefined reference to `__imp__PL_markstack_ptr'
ext/DynaLoader/DynaLoader.c:346: undefined reference to `__imp__PL_stack_base'
Reini Urban [Tue, 14 Sep 2010 16:06:38 +0000 (18:06 +0200)]
Update cygwin hints
do not use usemymalloc (double size + slow)
remove deprecated libcygipc info
remove overlarge stack size
David Mitchell [Mon, 3 Jan 2011 12:56:51 +0000 (12:56 +0000)]
reindent tryAMAGICunTARGET after previous change
David Mitchell [Mon, 3 Jan 2011 12:54:21 +0000 (12:54 +0000)]
simplify tryAMAGICunTARGET
Expecting the targ in sp[-1] rather than sp[0] is accomplished
cleanly using dATARGET.
David Mitchell [Mon, 3 Jan 2011 12:51:27 +0000 (12:51 +0000)]
fix a few typos in lib/overload.t
Vincent Pit [Mon, 3 Jan 2011 11:05:57 +0000 (12:05 +0100)]
Reindent pp_rv2av
Father Chrysostomos [Mon, 3 Jan 2011 07:49:17 +0000 (23:49 -0800)]
missing article
Father Chrysostomos [Mon, 3 Jan 2011 07:38:32 +0000 (23:38 -0800)]
Father Chrysostomos [Mon, 3 Jan 2011 07:21:14 +0000 (23:21 -0800)]
Silence a compiler warning
Father Chrysostomos [Mon, 3 Jan 2011 07:14:37 +0000 (23:14 -0800)]
Restore the old description of $[
The new entry is a bit too futuristic: assignment to $[ still works
and it is not read-only.
This does not fully restore the old description, as it contained
grammatical errors and parts of it were not updated when 5.10 changed
the scoping.
Father Chrysostomos [Mon, 3 Jan 2011 07:00:57 +0000 (23:00 -0800)]
Revert "[perl #68654] next::method doesn't see UNIVERSAL"
This reverts commit
a5cd004dbd757df2bcf9e17aab6a8ed1272157d7.
Father Chrysostomos [Mon, 3 Jan 2011 07:00:53 +0000 (23:00 -0800)]
Revert "[perl #80098] Bleadperl breaks Attribute::Lexical"
This reverts commit
1726bc11330f7a943b1e12c6dd5fa5454b90abd6.
Peter Martini [Mon, 3 Jan 2011 06:54:04 +0000 (22:54 -0800)]
[perl #45147] Issue with the exists function
Perl_av_exists tested to see if regdata magic was present,
but did not have any logic to fetch that data in the positive
key case. Additionally, in the negative key case, if AvFILL
indicated the key existed, it wouldn't return, and would then
fall through to the logic that treated it like a real array.
Zsbán Ambrus [Mon, 3 Jan 2011 04:25:55 +0000 (20:25 -0800)]
[perl #81032] Overhaul Porting/epigraphs.pod
This patch makes multiple changes to Porting/epigraphs.pod and
pod/perlhist.pod.
For those that don't know, Porting/epigraphs.pod is a new document that
collects the quotes (chosen by Pumpkins) in perl release announcements.
The changes are the following.
1. Add a link pointing to each release announcement in the mailing
list archives. These are from ysth's list, the source from which
Porting/epigraphs.pod was originally compiled, but they weren't in
Porting/epigraphs.pod so far.
2. Reorder Porting/epigraphs.pod chronologically, because I believe
that makes more sense -- pod/perlhist.pod is still sorted by version numbers.
3. Incidentally, some missing releases are added to pod/perlhist.pod too.
4. Fix a mistake where Porting/epigraphs.pod gives the wrong version number.
5. Add some epigraphs that appear in ysth's list but not in
Porting/epigraphs.pod.
6. I did some research in perl history before the part that ysth's list
covers, and added older perl announcements I found. This work is not
complete: I stopped somewhere in 2000.
Paul "LeoNerd" Evans [Thu, 30 Dec 2010 20:50:13 +0000 (20:50 +0000)]
Give a warning if pack_sockaddr_in6 is asked to pack a non-zero scope_id on machines that can't support it
Paul "LeoNerd" Evans [Thu, 30 Dec 2010 19:26:19 +0000 (19:26 +0000)]
Use HAS_SIN6_SCOPE_ID rather than glibc version testing
Paul "LeoNerd" Evans [Thu, 16 Dec 2010 11:49:49 +0000 (11:49 +0000)]
Bugfix to Socket.pm 's GAI fallback constants - use $name rather than $_
Paul "LeoNerd" Evans [Wed, 15 Dec 2010 16:02:19 +0000 (16:02 +0000)]
Capture newly-created unit tests to MANIFEST
Paul "LeoNerd" Evans [Wed, 15 Dec 2010 15:13:19 +0000 (15:13 +0000)]
Make Socket.pm 'use strict' clean
Paul "LeoNerd" Evans [Wed, 15 Dec 2010 14:37:51 +0000 (14:37 +0000)]
Provide fallback implementation of getaddrinfo and getnameinfo in pure perl if libc doesn't provide one
Paul "LeoNerd" Evans [Mon, 13 Dec 2010 18:13:10 +0000 (18:13 +0000)]
Initial documentation of getaddrinfo and getnameinfo functions
Paul "LeoNerd" Evans [Mon, 13 Dec 2010 17:50:56 +0000 (17:50 +0000)]
Unit tests for Socket::getaddrinfo() and Socket::getnameinfo()
Paul "LeoNerd" Evans [Mon, 13 Dec 2010 17:50:06 +0000 (17:50 +0000)]
Implement Socket::getaddrinfo() and Socket::getnameinfo(), with related constants
Paul "LeoNerd" Evans [Fri, 10 Dec 2010 15:02:30 +0000 (15:02 +0000)]
Conditionally set sa_len-type fields when packing sockaddr addresses
Nuno Carvalho [Fri, 24 Dec 2010 23:10:49 +0000 (23:10 +0000)]
Add new test file t/porting/filenames.t
Check portability of filenames from the MANIFEST. Modified by
committer to remove check for multiple dots -- there are already
too many violations and there are workarounds to handle them.
Also tweaked to report full path so it's easier to find culprits.
Still need checks for dots in directory names and paths that
differ only by case.
Craig A. Berry [Mon, 3 Jan 2011 00:19:51 +0000 (18:19 -0600)]
Clever tests may fool with $^O.
So make t/test.pl use its private version throughout.
David Golden [Fri, 31 Dec 2010 04:43:44 +0000 (23:43 -0500)]
Add JSON::PP to the Perl core
Per discussions with Jesse Vincent, JSON::PP has been added
to the Perl core to support the new CPAN meta file specification
Father Chrysostomos [Mon, 3 Jan 2011 02:30:17 +0000 (18:30 -0800)]
tests for [perl #36347] Object destruction incomplete
Father Chrysostomos [Sun, 2 Jan 2011 22:51:21 +0000 (14:51 -0800)]
[perl #36347] Object destruction incomplete
do_clean_objs only looks for objects referenced by RVs, so blessed
array references and lexical variables (and probably other SVs, too)
are not DESTROYed.
This commit adds a new visit() call to sv_clean_objs, which curses
(DESTROYs and un-blesses, leaving the reference count as it is) any
objects that are still left after do_clean_named_io_objs. The new
do_curse routine (a pointer to which is passeds to visit()) follows
do_clean_named_io_objs’ example and explicitly skips the STDOUT and
STDERR handles, in case destructors need to use them.
The cursing code, which is now called from two places, is moved out of
sv_clear and put in its own routine. The check that the reference
count is zero does not apply when called from sv_clean_objs, so the
new S_curse routine takes a boolean argument that determines whether
that check should take place.
brian d foy [Mon, 3 Jan 2011 02:27:39 +0000 (20:27 -0600)]
Merge branch 'briandfoy/perlfaq' into blead
brian d foy [Mon, 3 Jan 2011 02:26:01 +0000 (20:26 -0600)]
perlfaq1: Adjust some of the language around the Perl 6 references
David Leadbeater [Sun, 2 Jan 2011 22:33:16 +0000 (14:33 -0800)]
[perl #73630] Build separate DTrace for miniperl
As outlined in the RT ticket due to miniperl's dependencies differing to
the final perl binary dtrace -G needs to be called separately for each.
Build tested on Mac OS X 10.6, Solaris 11 and Scientific Linux with
SystemTap from git.
(Solaris is the only system I have access to where DTRACE_O actually gets
defined. Neither the dtrace binary on OS X nor SystemTap's dtrace
compatibility layer accept the -G option.)
David Golden [Sun, 2 Jan 2011 22:34:44 +0000 (17:34 -0500)]
Update CPAN::Meta::YAML to 0.003
This is a minor documentation change at Ingy's request. Also,
the 0.003 CPAN release omits a test file that we were excluding
so I've removed it from the EXCLUDES list in Maintainers.pl
David Leadbeater [Sun, 2 Jan 2011 21:38:21 +0000 (13:38 -0800)]
[perl #81500] Make compilation work when NO_ENV_ARRAY_IN_MAIN is defined
This is only normally defined if OEMVS is defined; the commit that added it
(
2f3efc97) claims z/OS support. I guess no-one has tried this for awhile as
dd374669 broke this in 2005.
Peter John Acklam [Sun, 2 Jan 2011 21:13:31 +0000 (13:13 -0800)]
[perl #81400] Fix bmodinv() part of RT 63237
The following standard definition is used: z is the modular inverse of
x (mod y) if and only if x*z (mod y) = 1 (mod y).
- dist/Math-BigInt/lib/Math/BigInt.pm: Fix the code in bmodinv() so it can
handle negative arguments. The code can be optimized further for speed,
but correctnes first.
- dist/Math-BigInt/t/bigintpm.inc: Fix the test case for modinv(-3, -5).
The output should be -3, since -3 * -3 (mod -5) = -9 (mod -5) = -4, and
1 (mod -5) = -4.
- dist/Math-BigRat/t/bigratpm.inc: Fix same test case as above.
Math::BigRat::bmodinv() only handles integers, and is essentially just a
front-end to Math::BigInt::bmodinv().
Craig A. Berry [Sun, 2 Jan 2011 21:16:11 +0000 (15:16 -0600)]
Don't clobber $^O when you're going to use it again.
Even if using it means relying on t/test.pl functionality such as
fresh_perl_is().
Nicholas Clark [Sun, 2 Jan 2011 20:56:21 +0000 (20:56 +0000)]
GvIO(gv) returns NULL for a NULL gv, so refactor to take advantage of this.
Simplify tests of !gv || !io to just !io, avoid calling GvIO(gv) more than
once, and where possible initialise io at declaration time, to allow it to be
const.
David Mitchell [Sun, 2 Jan 2011 19:38:30 +0000 (19:38 +0000)]
make <expr> always overload if expr is overloaded
Due to the way that '<> as glob' was parsed differently from
'<> as filehandle' from 5.6 onwards, something like <$foo[0]>
didn't handle overloading, even where $foo[0] was an overloaded object.
This was contrary to the docs for overload, and meant that <> couldn't
be used as a general overloaded iterator operator.
David Mitchell [Sun, 2 Jan 2011 15:04:50 +0000 (15:04 +0000)]
overloaded <> sometimes left an extra stack arg
David Mitchell [Fri, 31 Dec 2010 18:55:36 +0000 (18:55 +0000)]
call pp_glob() even when its being skipped
Currently when an external Perl glob function is used (which is most of
the time), the OP_GLOB op is removed and replaced with the pair:
GV("CORE::GLOBAL::glob"), ENTERSUB.
This commit re-adds the OP_GLOB to the op tree, but with OPf_SPECIAL set;
and pp_glob() is updated to just return if OPf_SPECIAL is set.
Thus there's no change in outward functionality with this commit. However,
by always calling pp_glob(), it will allow us (in the next commit) to
handle iterator overloading consistently, regardless of whether the
internal globbing function is used or not.
David Mitchell [Fri, 31 Dec 2010 14:53:24 +0000 (14:53 +0000)]
ck_glob() - add a little code comment
Nicholas Clark [Sun, 2 Jan 2011 17:01:49 +0000 (18:01 +0100)]
1; is not a valid no-op for /bin/sh, but true is.
b99c822f5c5b52c3 was not actually correct in identify the cause of the errors.
(although still a valid change to make for the sake of consistency)
Florian Ragwitz [Sun, 2 Jan 2011 17:08:26 +0000 (18:08 +0100)]
Update META.yml's no_index after the Time-Local move
We should probably write a porting test for that at some point.
Florian Ragwitz [Sun, 2 Jan 2011 16:19:37 +0000 (17:19 +0100)]
CPAN is upstream for Time::Local again
Florian Ragwitz [Sun, 2 Jan 2011 15:25:52 +0000 (16:25 +0100)]
Make Time::Local run on perls older than 5.12 again
With 5.12, this module was made y2038-safe, even on systems with a 32-bit
time_t. Unfortunately, that broke things on older perl versions and blead became
this module's upstream as the y2038-safety couldn't easily be backported.
This change also doesn't attempt to backport y2038 support. It merely restores
the old behaviour including the y2038 limitations on older versions of perl on
platforms without a 64-bit time_t.
With this, we can at least move Time::Locale's upstream back to CPAN. Making it
y2038-safe for old perls on 32-bit time_t platforms probably shouldn't be a
priority anyway. There's Time::y2038, which does everything this module does,
and then some, while being y2038-safe.
At some point, this module should probably be deprecated.
Nicholas Clark [Sun, 2 Jan 2011 10:40:25 +0000 (10:40 +0000)]
chmod +x metaconfig.SH.
Otherwise ./Configure -S generates a diagnostic, as it expects to be able to
execute it.
Jesse Vincent [Sun, 2 Jan 2011 02:50:21 +0000 (10:50 +0800)]
perlbug did not previously generate a From: header. While some MTAs do
the "right" thing and insert a valid "From:", not all of them do,
potentially resulting in dropped mail.
Nicholas Clark [Sat, 1 Jan 2011 15:33:18 +0000 (15:33 +0000)]
Ensure ExtUtils::CBuilder's 04-base.t cleans up its temporary directories.
The tempdir() function has the rather confusing default of *not* cleaning up
("because of issues with backwards compatibility") so one needs the CLEANUP
flag. (The analogous newdir() method in the OO interface doesn't suffer this
gotcha.)
Jesse Vincent [Sat, 1 Jan 2011 10:46:20 +0000 (18:46 +0800)]
Document 'test_porting' and start a section on how committing to blead
Chris 'BinGOs' Williams [Sat, 1 Jan 2011 10:37:04 +0000 (10:37 +0000)]
Update Encode to CPAN version 2.42
[DELTA]
$Revision: 2.42 $ $Date: 2010/12/31 22:48:48 $
! Encode.xs
! Unicode/Unicode.xs
Applied: RT#64371: Update for 5.14 API changes
http://rt.cpan.org/Ticket/Display.html?id=64371
Nicholas Clark [Fri, 31 Dec 2010 16:20:11 +0000 (16:20 +0000)]
Avoid using OP_NAME() in unimplemented_op(), as it won't give useful strings.
Using OP_NAME() isn't going to be helpful here. Firstly, it doesn't cope
with out of range op numbers - it only "special" cases op_custom. Secondly,
as the three ops we "panic" on are padmy, mapstart and custom, if we get
here for a custom op then that means that the custom op didn't have an
implementation. Given that OP_NAME() looks up the custom op by its pp_addr,
likely it will return NULL, unless someone (unhelpfully) registers
&PL_unimplemented_op as the address of their custom op. NULL doesn't
generate a useful error message. "custom" does. */
David Golden [Fri, 31 Dec 2010 16:29:33 +0000 (11:29 -0500)]
Porting/Maintainers.pl: add (MY)META.json to ignored files
David Mitchell [Thu, 30 Dec 2010 16:29:12 +0000 (16:29 +0000)]
standardise amagic method naming
Some amagic-related macros take the full method enumeration name,
(e.g. "add_amg"); while others "helpfully" allow you to pass a shortened
version, ("add"), and do a CAT2(meth,_amg) behind the scenes.
Standardise on passing the full name; this makes it less confusing and
allows you to grep for the enumeration name in the source.
It updates two macros to accept full enumeration names: tryAMAGICunTARGET
(which isn't used outside the core apparently), and AMG_CALLun, which is
replaced by a new AMG_CALLunary (since AMG_CALLun is used outside the
core).
Craig A. Berry [Fri, 31 Dec 2010 01:45:13 +0000 (19:45 -0600)]
VMS nits in cpan/Test-Harness/t/source.t
1.) Can't get the basename of a directory in native format.
2.) Symlinks still a bit dodgy -- skip tests for now.
Submitted upstream at:
https://rt.cpan.org/Ticket/Display.html?id=64353
Craig A. Berry [Thu, 30 Dec 2010 22:29:50 +0000 (16:29 -0600)]
Skip ExtUtils-CBuilder base tests expected and known to fail on VMS.
Nicholas Clark [Thu, 30 Dec 2010 17:45:53 +0000 (17:45 +0000)]
Merge the opcode bodies for pp_bind and pp_connect.
Nicholas Clark [Thu, 30 Dec 2010 17:30:24 +0000 (17:30 +0000)]
Most socket ops weren't warning for unopened handles unless 'closed' was enabled
They were checking that category 'closed' was enabled for warnings before
calling report_evil_fh(), which in turn was (correctly) checking category
'unopened'.
Nicholas Clark [Thu, 30 Dec 2010 17:14:15 +0000 (17:14 +0000)]
Convert a pp_sys test to use warnings 'closed', instead of the broader 'io'.
It's actually only testing warnings enabled by 'closed', so test using the
tighter category.
Nicholas Clark [Thu, 30 Dec 2010 16:52:22 +0000 (16:52 +0000)]
pp_leavewrite wasn't warning for unopened handles unless 'closed' was enabled.
It was checking that category 'closed' was enabled for warnings before calling
report_evil_fh(), which in turn was (correctly) checking category 'unopened'.
David Mitchell [Thu, 30 Dec 2010 10:32:44 +0000 (10:32 +0000)]
Better handling of magic methods freeing the SV
This is a fix for RT #81230 (and more). Currently, mg_get() works around
the case where the called magic (e.g. FETCH) frees the magic SV. It does
this by unconditionally pushing the SV on the tmps stack before invoking
the method.
There are two issues with this. Firstly, it may artificially extend the
life of the SV. This was the root of the problem with #81230. There, the
DB_File code, under -T, created a tainted tied object. Accessing the
object (within FETCH as it happens), caused mg_get() to be invoked on the
object (due to the taint magic), and thus extend the life of the object.
This then caused c<untie %h if $h{k}> to give the warning
untie attempted while 1 inner references still exist.
This only became noticeable after
efaf36747029c85b4d8825318cb4d485a0bb350e,
which stopped wrapping magic method calls in SAVETMPS/FREETMPS.
The second issue issue that this protection only applies to mg_get();
functions like mg_set() can still segfault if the SV is deleted.
This commit fixes both problems as follows:
First, the protection mechanism is moved out of mg_get() and into
save_magic() / restore_magic(), so that it protects more things.
Secondly, the protection is now:
* in save_magic(), SvREFCNT_inc() the SV, thus protecting it from being
freed during FETCH (or whatever)
* in restore_magic(), SvREFCNT_dec() the SV, undoing the protection
without extending the life of the SV, *except* if the refcount is
1 (ie FETCH tried to free it), then push it on the mortals stack
to extend it life a bit so our callers wont choke on it.
Renee Baecker [Wed, 29 Dec 2010 09:10:32 +0000 (01:10 -0800)]
Update Renee Baeckers mail address
Signed-off-by: David Golden <dagolden@cpan.org>
Tony Cook [Wed, 29 Dec 2010 05:44:58 +0000 (16:44 +1100)]
properly clean up work files from testing GDBM_File
Tony Cook [Tue, 28 Dec 2010 11:39:14 +0000 (22:39 +1100)]
close the produced test file so win32 can unlink it
Previously runs would leave a GLOB(0x...) file in t/
Tony Cook [Tue, 28 Dec 2010 11:09:31 +0000 (22:09 +1100)]
skip ExtUtils-CBuilder base tests expected and known to fail on Win32
Nicholas Clark [Tue, 28 Dec 2010 09:50:49 +0000 (09:50 +0000)]
As report_evil_fh() checks WARN_{CLOSED,UNOPENED}, don't duplicate this.
This trades reduced code size for an extra function call in the error path with
warnings disabled. (And removes a duplicated check for the case of taking the
error path *with* warnings enabled.)
Removing the check from Perl_do_close() does not change behaviour, as io is
NULL there, hence Perl_report_evil_fh() will always be checking WARN_UNOPENED
and setting vile to "unopened".
Nicholas Clark [Tue, 28 Dec 2010 09:30:31 +0000 (09:30 +0000)]
As report_wrongway_fh() checks ckWARN(WARN_IO) internally, don't duplicate this.
This trades reduced code size for an extra function call in the error path with
warnings disabled. (And removes a duplicated check for the case of taking the
error path *with* warnings enabled.)
Nicholas Clark [Tue, 28 Dec 2010 09:20:43 +0000 (09:20 +0000)]
Remove OP_phoney_{IN,OUT}PUT_ONLY, as they are no longer used or supported.
Nicholas Clark [Tue, 28 Dec 2010 09:09:44 +0000 (09:09 +0000)]
Argument op to report_evil_fh() is always PL_op->op_type, so need not be passed
Nicholas Clark [Tue, 28 Dec 2010 08:56:36 +0000 (08:56 +0000)]
The io argument to report_evil_fh() is always GvIO(gv), so need not be passed.
Nicholas Clark [Tue, 28 Dec 2010 07:55:45 +0000 (07:55 +0000)]
Extract the OP_phoney_* code from report_evil_fh() into report_wrongway_fh()
Previously Perl_report_evil_fh()'s body was just an if/else at the top level -
a good sign that it is actually implementing two disjoint functions.
Nicholas Clark [Tue, 28 Dec 2010 08:32:44 +0000 (08:32 +0000)]
In pp_sockpair, remove duplication of code to close the supplied file handles.
It's not necessary to (also) test gv1 and gv2 before returning undef as an
error, because io1 will automatically be NULL if gv1 is NULL, and similarly
io2 if gv2 is.
Nicholas Clark [Tue, 28 Dec 2010 08:19:03 +0000 (08:19 +0000)]
Fix typo in warning code added to pp_sockpair in
c289d2f7288798f8.
That should be gv2, not gv1. However, I believe that it's impossible to reach
the warning code, given the structure of the optree that the perl 5
implementation produces, as gv1 and gv2 will never be NULL, and GvIOn() will
always return non-NULL. (Or croak, but that won't return).
Tony Cook [Tue, 28 Dec 2010 02:58:29 +0000 (13:58 +1100)]
avoid undefined behaviour in fetch and modify of SP from
0a0e3cc51e
pp_sys.c: In function ‘Perl_pp_send’:
pp_sys.c:1845: warning: operation on ‘sp’ may be undefined
David Golden [Mon, 27 Dec 2010 19:50:31 +0000 (14:50 -0500)]
Add CPAN::Meta::YAML as a dual-life core module
Nicholas Clark [Mon, 27 Dec 2010 19:11:26 +0000 (19:11 +0000)]
Simplify return handling for tied handle OPEN/PRINTF/READ/WRITE.
Their code used to explicitly move the return value from its current position
on the stack, to the position that was top of top of the stack just before the
call was made. However, the POPBLOCK() in pp_leavesub will restore the stack
pointer, and passing G_SCALAR to call_method() will force exactly one return
value, so all of this is needless defensiveness.