Steve Hay [Thu, 19 Sep 2013 15:47:45 +0000 (16:47 +0100)]
Upgrade CPAN::Meta from version 2.132510 to 2.132620
Steve Hay [Thu, 19 Sep 2013 15:27:46 +0000 (16:27 +0100)]
perldelta - 'nonexistent' should be 'non-existent'
Steve Hay [Thu, 19 Sep 2013 15:16:27 +0000 (16:16 +0100)]
perldelta - Remove all but one XXX notices
Steve Hay [Thu, 19 Sep 2013 15:10:36 +0000 (16:10 +0100)]
perldelta - Fill in currently remaining TODO items
9d32676e doesn't need one since it only fixes things that were added post
5.19.3, but [perl #117265] is worth mentioning somewhere so I've added that
to the warnings entry.
Steve Hay [Thu, 19 Sep 2013 14:32:17 +0000 (15:32 +0100)]
perldelta - Update entry for
b29f65fce6
Steve Hay [Thu, 19 Sep 2013 14:25:44 +0000 (15:25 +0100)]
Steve Hay [Thu, 19 Sep 2013 14:20:00 +0000 (15:20 +0100)]
perldelta - More copy-editing
Steve Hay [Thu, 19 Sep 2013 14:04:58 +0000 (15:04 +0100)]
perldelta - Un-TODO
9c7618be
As noted by Karl Williamson, it's very unlikely that any user would have
encountered the bug which this fixed, so it's not worth mentioning in
perldelta.
Steve Hay [Thu, 19 Sep 2013 13:27:10 +0000 (14:27 +0100)]
Add a USING_MSVC6 macro to identify Microsoft Visual C++ 6.0
This simplifies some of the logic necessary for coping with its various
problems.
Suggested by Nicholas Clark.
Nicholas Clark [Thu, 19 Sep 2013 13:13:24 +0000 (15:13 +0200)]
Remove Cwd from the "actually architecture dependant files" special cases.
This probably hasn't been needed since the file Cwd.pm was moved from lib/
into the directory containing its XS code.
Neil Bowers [Wed, 18 Sep 2013 19:47:53 +0000 (20:47 +0100)]
Added is_core(), which returns true if the module was/is in core
Default to checking against $^V, but you can optionally specify
the perl release, and can also optionally specify a minimum
version of the module.
Signed-off-by: Chris 'BinGOs' Williams <chris@bingosnet.co.uk>
Nicholas Clark [Thu, 19 Sep 2013 09:32:02 +0000 (11:32 +0200)]
Remove 4 redundant #undefs, missed by commit
79be8fb4ac8fa995.
That commit removed all the code that defined PERL_NEED_MY_HTOLE64,
PERL_NEED_MY_LETOH64, PERL_NEED_MY_HTOBE64 and PERL_NEED_MY_BETOH64, but
missed removing code in perl.h which conditionally undefined them.
Craig A. Berry [Wed, 18 Sep 2013 02:16:21 +0000 (21:16 -0500)]
Steve Hay [Wed, 18 Sep 2013 22:56:21 +0000 (23:56 +0100)]
Fix the VC6 build on Windows following commit
e25d460c74
As noted in commit
38aa66aabf, VC6 on Windows does not support the
64-bit-int build option, and likewise is broken if we don't undef HAS_QUAD
in the core.
Steve Hay [Wed, 18 Sep 2013 17:09:47 +0000 (18:09 +0100)]
Update win32/config_H.[gv]c canned configuration files
The win32/config.[gv]c files are generally kept up to date these days (and
we have tests to check that) so no changes are required in them.
The win32/config_H.[gv]c files are regenerated as per instructions in the
win32/Makefile and win32/makefile.mk, being careful to restore a couple of
things otherwise lost from the config_H.gc file. The files are now in sync
with the top-level master configuration file, config_h.SH.
Father Chrysostomos [Wed, 18 Sep 2013 16:10:31 +0000 (09:10 -0700)]
Add Toby Inkster to AUTHORS
Toby Inkster [Wed, 18 Sep 2013 12:29:47 +0000 (13:29 +0100)]
Exclude perlfaq1, perlglossary, etc from the list of pragmata - these just pod, not modules.
Steve Hay [Wed, 18 Sep 2013 13:11:21 +0000 (14:11 +0100)]
Really fix precedence problem in IO::Socket::connect() from
80d2c56d79
Thanks to Dagfinn Ilmari Mannsåker for paying more attention than me.
Steve Hay [Wed, 18 Sep 2013 12:58:27 +0000 (13:58 +0100)]
perldelta - Correction to new slice syntax
Thanks to Michael Schroeder for the spot.
Chris 'BinGOs' Williams [Wed, 18 Sep 2013 09:52:57 +0000 (10:52 +0100)]
Make changes_between() in Module::CoreList API the same as the other functions
Steve Hay [Wed, 18 Sep 2013 10:09:21 +0000 (11:09 +0100)]
Fix precedence problem in IO::Socket::connect() from
80d2c56d79
Thanks to Andreas Koenig for the spot.
Brian Fraser [Wed, 18 Sep 2013 08:42:12 +0000 (05:42 -0300)]
Merge branch 'hugmeir/disallow_cntrl' into blead
[perl #119123] disallow literal control character variables
This branch introduces a deprecation warning on variables like $^T,
where ^T is a literal control character in the source code.
It also reworks how the handling of whitespace in ${...} works,
fixing a number of bugs, such as ${\nfoo\n} not increasing the
line number.
Brian Fraser [Wed, 18 Sep 2013 08:20:43 +0000 (05:20 -0300)]
toke.c, scan_ident: No need for a while loop when using PEEKSPACE
PEEKSPACE already gobbles up all the remaining spaces until it
either finds a non-space character, or EOF.
Brian Fraser [Sun, 1 Sep 2013 23:41:26 +0000 (20:41 -0300)]
toke.c, scan_ident(): use PEEKSPACE() to skip over whitespace.
This fixes a number of bugs regarding whitespace and line numbers
in scan_ident(), such as ${\nfoo\n} not increasing the line number,
or ${\ntime\n[1]} not working.
It goes through a number of hoops to get the correct line number for
warnings emmitted from scan_ident, and reverts CopLINE to its
original value if scan_ident() is giving up and returning from the
point of the opening bracket, like in the case of ${\n\nfoo()}.
Brian Fraser [Sun, 1 Sep 2013 22:32:17 +0000 (19:32 -0300)]
toke.c, S_scan_ident(): Don't take a "end of buffer" argument, use PL_bufend
All but one of scan_ident()'s callers already passed PL_bufend as
the removed argument; The one deviant was intuit_more(), which was
setting the "end of buffer" argument, to the next close-bracket.
This commit modifies intuit_more() to temporarily set PL_bufend and
then restore it.
This was done as groundwork for the following commit, which will add
more uses of PEEKSPACE() to scan_ident() in order to fix some whitespace
and line number bugs, and PEEKSPACE() modifies PL_bufend directly
if it encounters a newline at the end of the buffer -- that last bit
being why changing intuit_more() to modify-and-restore PL_bufend is
safe, since the end of the buffer will always be a ']'
Brian Fraser [Sun, 1 Sep 2013 13:21:52 +0000 (10:21 -0300)]
[perl #119123] disallow literal control character variables
This introduces a deprecation warning on things like $^T, where ^T
is a literal control character in the source code.
Brian Fraser [Sun, 1 Sep 2013 13:51:31 +0000 (10:51 -0300)]
toke.c, S_scan_ident: Skip over newlines as well as spaces and tabs.
Not doing this lead to some bugs, such as 'eval "*{\nOIN}"'
returning a glob for "\nOIN" rather than just "OIN", or
'eval "\${\cT\n}"' giving spurious syntax errors.
Note however that this is not a full fix, since that latter
test case still fails outside of an eval.
Brian Fraser [Sun, 1 Sep 2013 12:41:20 +0000 (09:41 -0300)]
Reworked t/base/lex.t to use less hardcoded test numbers.
The later two thirds of the file now consistently use $test, which
should make future tweaks to the file simpler.
Nicholas Clark [Wed, 18 Sep 2013 08:00:45 +0000 (10:00 +0200)]
Merge the cast fixes that enable drand48 to work on 32-bit HP-UX builds.
Nicholas Clark [Tue, 17 Sep 2013 12:40:29 +0000 (14:40 +0200)]
Update the comment in util.h related to HAS_QUAD and U64TYPE.
Nicholas Clark [Tue, 17 Sep 2013 12:29:10 +0000 (14:29 +0200)]
Use U64_CONST() for the drand48 constants that are larger than 2**32.
Without this the constants are truncated to 32 bits on (at least) HP-UX
with 32 bit IVs, which results in some very wrong output from rand().
Nicholas Clark [Tue, 17 Sep 2013 12:10:28 +0000 (14:10 +0200)]
For PERL_CORE, don't undefine HAS_QUAD when building with 32 bit IVs.
When commit
de1c2614758a00c0 added support for $Config{d_quad} and HAS_QUAD
in Oct 1999, it also undefined HAS_QUAD. It's been that way ever since.
This isn't actually correct on a platform with a 64 bit type available, but
changing it now will cause too much breakage on CPAN for too little gain
(See RT #119753). However, we do need HAS_QUAD in the core for use by the
drand48 code, so make the undef conditional on PERL_CORE not being set.
Tony Cook [Wed, 18 Sep 2013 06:44:20 +0000 (16:44 +1000)]
[perl #119539] clarify that copy() doesn't create the directory
Tony Cook [Wed, 18 Sep 2013 06:33:41 +0000 (16:33 +1000)]
bump File::Copy's $VERSION
Tony Cook [Wed, 18 Sep 2013 06:29:17 +0000 (16:29 +1000)]
add Nathan Glenn to AUTHORS
Nathan Glenn [Sat, 31 Aug 2013 00:40:53 +0000 (17:40 -0700)]
Added note about directories not being created with File::Copy::copy
Father Chrysostomos [Wed, 18 Sep 2013 01:20:37 +0000 (18:20 -0700)]
perldelta for
acb340504/#119501
Father Chrysostomos [Wed, 18 Sep 2013 01:17:10 +0000 (18:17 -0700)]
Father Chrysostomos [Wed, 18 Sep 2013 01:14:46 +0000 (18:14 -0700)]
perldelta: Two to-dos that don’t deserve mention
9cefd268 [perl #118747] Allow in-place s///g when !!PL_sawampersand
c04e4151 op.c:leave_scope: use mg_free before sv_force_normal [#119295]
In both cases, it is a module relying on undocumented internals.
In the former, we do have an optimisation kicking in in cases where it
didn’t in 5.18 and earlier. But those cases are rare enough that an
entry in ‘Performance Enhancements’ would be misleading.
In the latter, the order in which clearing and ‘free’ magic are trig-
gered is more predictable, but I don’t believe that has ever been
guaranteed, and only creepy modules will be affected by it.
Tony Cook [Wed, 18 Sep 2013 01:31:02 +0000 (11:31 +1000)]
depend on the new win32 sys/errno2.h header
Steve Hay [Tue, 17 Sep 2013 22:14:05 +0000 (23:14 +0100)]
Sort the contents of %Modules in Porting/Maintainers.pl
(No changes other than the ordering of the entries.)
Steve Hay [Tue, 17 Sep 2013 22:06:21 +0000 (23:06 +0100)]
Steve Hay [Tue, 17 Sep 2013 21:59:52 +0000 (22:59 +0100)]
perldelta - Three module upgrades which haven't come from new CPAN releases
Steve Hay [Tue, 17 Sep 2013 21:59:02 +0000 (22:59 +0100)]
Use non-dev version number in IO::Socket
There was no need for
80d2c56d79 to use a dev number. The IO distribution's
modules all have differing version numbers, mostly all well ahead of the
last CPAN release (UPSTREAM is 'blead' anyway), and no others are dev
numbers.
Steve Hay [Tue, 17 Sep 2013 21:44:24 +0000 (22:44 +0100)]
Upgrade parent from version 0.227 to 0.228
Matthew Horsfall (alh) [Tue, 17 Sep 2013 12:49:35 +0000 (08:49 -0400)]
RT-119817 - Treat =back as the end of a warning description.
This keeps any trailing data in the file from showing up as part
of the last warning's description.
Steve Hay [Tue, 17 Sep 2013 17:15:00 +0000 (18:15 +0100)]
perldelta - Fill in one more item and add currently remaining possible TODOs
Father Chrysostomos [Tue, 17 Sep 2013 15:28:36 +0000 (08:28 -0700)]
perlhack: Make MAINTENANCE BRANCHES a link
Father Chrysostomos [Tue, 17 Sep 2013 15:26:12 +0000 (08:26 -0700)]
[perl #119501] \(1+2) always referencing the same sv
2484f8dbbb hid the fact that constant folding happens by making 1+2
fold to a PADTMP, an SV that is never used as an lvalue and gets cop-
ied if one tries to use it that way.
The PADTMP mechanism is what allows \"$x" to return a new value each
time, even though ops like "$x" actually reuse the same scalar repeat-
edly to return values.
Because \ copies PADTMPs, \(1+2) ends up folding 1+2 to 3 (marked
PADTMP), which is then copied by the \ at compile time. (Constant
folding works by evaluating certain ops at compile time and then
inlining their returned value.) The result is that we have a folded
\3 where 3 is *not* marked PADTMP (the reference is, instead); hence
\(1+2) gives the same scalar each time, producing a value that can be
modified, affecting future evaluations.
The solution is to skip folding \ if its argument is a PADTMP.
Father Chrysostomos [Mon, 16 Sep 2013 17:09:11 +0000 (10:09 -0700)]
Rewrap one perldiag entry for better splain output
The right margin is slightly less uneven this way (once the pod
formatting is stripped).
Steve Hay [Tue, 17 Sep 2013 15:26:22 +0000 (16:26 +0100)]
Upgrade CPAN from version 2.00 to 2.03-TRIAL
Nicholas Clark [Tue, 17 Sep 2013 11:58:04 +0000 (13:58 +0200)]
Merge fixes needed to build with 32 bit IVs and HAS_QUAD defined.
However, don't actually eliminate the #undef HAS_QUAD for 32 bit IVs, as
the costs (in CPAN breakage) look to be higher than the gains.
(See RT #119753 for more detailed analysis.)
Nicholas Clark [Wed, 11 Sep 2013 12:44:48 +0000 (13:44 +0100)]
Eliminate POPq, POPuq, TOPq, TOPuq, dPOPqv, dTOPqv, qPOPuqv, dTOPuqv.
These shortcut macros are unused in the core, and unused by any code on CPAN.
If any XS code we can't see *is* using them, it will now fail to compile, and
can easily be fixed by replacing the macros with their expansion.
Nicholas Clark [Wed, 11 Sep 2013 11:18:43 +0000 (12:18 +0100)]
The choice of 7 or 13 byte extended UTF-8 should be based on UVSIZE.
Previously it was based on HAS_QUAD, which is not (as) correct.
Nicholas Clark [Wed, 11 Sep 2013 11:12:25 +0000 (12:12 +0100)]
Use IVSIZE not HAS_QUAD to enable 'q' and 'Q' formats in pack.
Whilst the code for 'q' and 'Q' in pp_pack is itself well behaved if enabled
on a perl with 32 bit IVs (using SvNV instead of SvIV and SvUV), the
regression tests are not. Several tests use an eval of "pack 'q'" to
determine if 64 bit integer support is available (instead of
$Config{ivsize}), and t/op/pack.t fails many tests. While these could be
fixed (or skipped), unfortunately the approach of evaling "pack 'q'" is
fairly popular on CPAN, so the breakage isn't just in the perl core, and
might also be present in code we can't see or submit patches for.
Nicholas Clark [Wed, 11 Sep 2013 11:09:25 +0000 (12:09 +0100)]
Use IVSIZE not HAS_QUAD to enable "long long" formats in sv_vcatpvfn_flags().
Without this, enabling HAS_QUAD on 32 bit IV systems causes a lot of test
failures. Potentially those failures could be addressed, but it seems like
a lot more work for little gain.
H.Merijn Brand [Tue, 17 Sep 2013 08:36:38 +0000 (10:36 +0200)]
Atari FreeMiNT: MiNT does not have .exe for _exe
[perl #89502]: Add support for Atari FreeMiNT platform
Steve Hay [Tue, 17 Sep 2013 07:39:38 +0000 (08:39 +0100)]
perldelta - Correction to new Win32 entry in Platform-Specific Notes
Christian Millour [Tue, 17 Sep 2013 07:21:22 +0000 (09:21 +0200)]
perldelta for #119857
Christian Millour [Tue, 17 Sep 2013 06:17:19 +0000 (08:17 +0200)]
add reference to maintenance branches in perlpolicy
Steve Hay [Tue, 17 Sep 2013 07:35:58 +0000 (08:35 +0100)]
perldelta - Document two new test scripts
Tony Cook [Tue, 17 Sep 2013 07:03:25 +0000 (17:03 +1000)]
Tony Cook [Tue, 17 Sep 2013 06:57:37 +0000 (16:57 +1000)]
[perl #85228] stop $!=EINVAL; waitpid(0,0) from looping
Tony Cook [Tue, 17 Sep 2013 05:59:13 +0000 (15:59 +1000)]
[perl #119097] Test that print() is not returning EINTR
Victor [Mon, 12 Aug 2013 08:49:58 +0000 (12:49 +0400)]
Fixing eintr_print.t intermittent hang
1. Disable test for PERLIO=stdio
2. Remove binmode - it turns out it's useless
3. Copy OS blacklist from eintr.t ( RT #85842, RT #84688)
4. Add additional delay before child exit, just in case.
Victor [Fri, 2 Aug 2013 10:39:59 +0000 (14:39 +0400)]
Test that print() is not returning EINTR.
fails under 5.14.x ( see RT #119097 )
also fails under 5.8.x
Currently test enabled on linux/bsd/solaris/darwin
Tony Cook [Tue, 17 Sep 2013 05:45:43 +0000 (15:45 +1000)]
[perl #119857] make win32_link and win32_rename raise ENOSPC and EDQUOT when appropriate
Tony Cook [Tue, 17 Sep 2013 05:43:53 +0000 (15:43 +1000)]
add Christian Millour to AUTHORS
Christian Millour [Mon, 16 Sep 2013 23:52:26 +0000 (01:52 +0200)]
make win32_link and win32_rename raise ENOSPC and EDQUOT when appropriate
Karl Williamson [Tue, 17 Sep 2013 05:12:28 +0000 (23:12 -0600)]
perldelta for
3f40aba3 Merge branch 'ebcdic' into blead
Karl Williamson [Tue, 17 Sep 2013 04:42:09 +0000 (22:42 -0600)]
perlapi: Typos; clarify comment
Karl Williamson [Mon, 16 Sep 2013 03:16:10 +0000 (21:16 -0600)]
regexec.c: Reorder cases in switch to eliminate a goto
I imagine that the goto gets optimized out, but this is clearer anyway;
considered less harmful.
Craig A. Berry [Sat, 14 Sep 2013 23:25:58 +0000 (18:25 -0500)]
Earlier safety check for @INC entry in pp_require.
In order to make sure we find embedded NULs early enough, before
the directories in @INC have been through library calls that use C
strings, check each directory in pp_require before concatenating
a filename onto it.
François Perrad [Fri, 13 Sep 2013 07:34:58 +0000 (09:34 +0200)]
skip lib/ExtUtils/t/Embed.t when cross-compiling
The toolchain is not installed on the target when cross-compiling.
So, this test must be skipped, see patch below.
Steve Hay [Mon, 16 Sep 2013 21:43:45 +0000 (22:43 +0100)]
perldelta - A couple more corrections to recent additions
Steve Hay [Mon, 16 Sep 2013 20:59:12 +0000 (21:59 +0100)]
Note that the USE_64_BIT_INT build option is not supported with MSVC++ 6.0
Also change the Windows makefiles to disable this option if it is set with
this compiler selected. (Use !UNDEF (for nmake) and != (for dmake) to be
sure of disabling the option if it is defined on the command-line rather
than by editing the makefile.)
Steve Hay [Mon, 16 Sep 2013 19:57:31 +0000 (20:57 +0100)]
perldelta - Remove spurious New Errors entries
Spotted by Father C - These are not new errors. They are just listed with
explicit ‘exists’ and ‘delete’ now (and alphabetised under the same) rather
than %s.
Father Chrysostomos [Mon, 16 Sep 2013 02:24:29 +0000 (19:24 -0700)]
Tweak B::Concise tests following OPpCONST_FOLDED removal
Father Chrysostomos [Sun, 15 Sep 2013 22:55:19 +0000 (15:55 -0700)]
Fix minor flagging bug in op.c:fold_constants
Nothing is making use of this in at present in the affected cases.
What this code does is mark an op as being folded so that 1+2 can
be distinguished from 3. qq"foo" was erroneously being marked as
folded, unlike "foo". Constant folding is part of the way that
quote-like operators are implemented. Conceptually there is no
folding here, so we should pretend it did not happen. This was a
partial cause of bug #116086.
Father Chrysostomos [Sun, 15 Sep 2013 22:41:32 +0000 (15:41 -0700)]
Alphabetise perldiag
Father Chrysostomos [Sun, 15 Sep 2013 22:37:43 +0000 (15:37 -0700)]
Remove OPpCONST_FOLDED
Now that we have op->op_folded, we don’t need OPpCONST_FOLDED any
more. In removing it, I modified B::Concise to output op_folded the
way OPpCONST_FOLDED was output before, since it can be helpful to have
it when reading op dumps.
Father Chrysostomos [Sun, 15 Sep 2013 22:56:34 +0000 (15:56 -0700)]
Make B::OP::slabbed, folded, etc. actually work
B::OP::slabbed = 47
B::OP::savefree = 48
B::OP::static = 49
B::OP::folded = 50
Added by
3164fde474, these have never worked, because the alias index
check in B::OP::next was never updated.
Instead of hard-coding the number, use C_ARRAY_LENGTH(...).
Steve Hay [Mon, 16 Sep 2013 14:40:14 +0000 (15:40 +0100)]
Steve Hay [Mon, 16 Sep 2013 14:38:58 +0000 (15:38 +0100)]
Merge branch 'errno' into blead
Fix the handling of new Exxx constants in errno.h in VC++ 2010 onwards.
Handling these new constants was originally done in commits
b59e75b34c and
912c63ed00 but that design's redefinition of some constants can break XS
modules which link against other libraries in which the constants have not
been redefined in that way.
For example, the following program ought to print 106 in VC++ 2010 builds,
but instead prints 10053, causing some breakage in mod_perl (specifically,
APR::Status::is_ECONNABORTED() does not work):
#include <windows.h>
#include "EXTERN.h"
#include "perl.h"
#include "XSUB.h"
void main(void) {
printf("ECONNABORTED=%d\n", ECONNABORTED);
}
This could be worked around in mod_perl, but the same problem will also
occur in other XS extensions so it seems that a more general solution to
make perl friendlier in its dealings with errno.h constants would be better.
The solution here is to switch to the alternative design mentioned in the
first commit cited above, namely, assigning Exxx constants rather than
WSAExxx constants to $! where possible, and not redefining any Exxx
constants.
This causes some backwards-compatibility issues with a few WSAExxx error
codes having corresponding Exxx codes < 100, e.g. wherever WSAEINVAL (10022)
might previously have been assigned to $! we now assign EINVAL (22) instead,
but this breakage is unavoidable if we are to reach a sane state now.
The breakage is minimized by intercepting Perl-level assignments to $! to
convert WSAExxx codes into Exxx codes where possible, so that existing code
performing such assignments will not need to be updated. It is possible that
with some magic, tests for $! having WSAExxx values could be made to test
for Exxx values instead, but the difficulty of doing this was not deemed
worthwhile (yet, at least) for the small amount of breakage involved.
Thanks to Jan Dubois for assistance and advice on the general approach
taken in these changes.
Steve Hay [Sat, 7 Sep 2013 22:44:38 +0000 (23:44 +0100)]
Silence noise from Errno_pm.PL on Windows
A block of warnings like this appears when building Errno with VC++:
Number found where operator expected at (eval 633) line 1, near ")0xC0000093"
(Missing operator before 0xC0000093?)
This is caused by #defines like these in the Windows header files:
WinBase.h:#define EXCEPTION_FLT_UNDERFLOW STATUS_FLOAT_UNDERFLOW
WinNT.h:#define STATUS_FLOAT_UNDERFLOW ((DWORD )0xC0000093L)
and can be silenced by teaching Errno_pm.PL about the otherwise unexpected
whitespace immediately after "DWORD" in the cast.
Steve Hay [Sat, 7 Sep 2013 20:53:35 +0000 (21:53 +0100)]
Fix the #include of winsock2.h in win32/include/sys/errno2.h
For some reason (I didn't figure out why) cpan/Win32/Win32.xs did not
compile as of the previous commit, but in any case we cannot just blindly
include winsock2.h because (a) it is too late if winsock.h has already
been included and (b) early WinCE doesn't have it.
The definition of _WINSOCKAPI_ was also wrongly not updated to _WINSOCK2API_
but was mistaken in its comment "Don't drag in everything" anyway: it's
actually just an old-fashioned "include guard", not a means to include a
minimal API from the file. (The mistaken comment came from the original
ext/Errno/Errno_pm.PL before this round of $!-related changes. That file
also had a redundant "#include <winsock.h>" left in it which should have
been removed along with earlier work that removed the other one, so tidy
that up too.)
Steal a hopefully more correct incantation from win32/include/sys/socket.h
instead.
Now that we allow for the possibility of winsock.h being used again, we
must take care not to make use of WSAECANCELLED if it is not defined.
(That is the only WSAExxx constant which we use that was new in winsock2.h.)
Steve Hay [Sat, 7 Sep 2013 18:03:16 +0000 (19:03 +0100)]
We need winsock2.h for WSAECANCELLED in errno2.h
Steve Hay [Sat, 7 Sep 2013 18:02:38 +0000 (19:02 +0100)]
Clarify a comment in errno2.h
Steve Hay [Sat, 7 Sep 2013 16:38:49 +0000 (17:38 +0100)]
Add more errno.h constants to POSIX for exporting if they exist
This change adds the following constants, which are all found in errno.h in
MS VC++ 2010 and higher (the other new constants added to that version of
errno.h are already listed in POSIX.pm):
EBADMSG
ECANCELED
EIDRM
EILSEQ
ENODATA
ENOLINK
ENOMSG
ENOSR
ENOSTR
ENOTRECOVERABLE
ENOTSUP
EOTHER
EOVERFLOW
EOWNERDEAD
EPROTO
ETIME
Steve Hay [Sat, 7 Sep 2013 16:28:29 +0000 (17:28 +0100)]
Fix Windows build for compilers other than VC10+
The new errno2.h only ensures that those new Exxx constants which are
required by convert_wsa_error_to_errno() have definitions; the other new
Exxx constants will remain undefined on compilers other than VC10+, so we
must take care not to use them unless we know they are defined.
Steve Hay [Fri, 6 Sep 2013 09:17:54 +0000 (10:17 +0100)]
Simplify errno2.h slightly
There is no need for it to worry about ensuring that standard errno.h values
like EINTR are defined.
There is no point in it defining dummy constants like EINVALIDPROCTABLE which
do not exist in errno.h at all. They would only be given the corresponding
WSAExxx values anyway, so the win32sck.c function might just as well return
them directly.
Other constants like ESOCKTNOSUPPORT which do not exist in errno.h on Windows
but presumably exist elsewhere since POSIX tries to exports them are still
worth defining (where possible -- it isn't possible for EHOSTDOWN since there
is no corresponding WSAEHOSTDOWN); add comments for these for clarity.
Also comment which errno.h constants are new in VC10, and mention three of
the four that win32/include/sys/socket.h used to redefine because they are
used in the perl core (ENOTSOCK, EAFNOSUPPORT and ECONNABORTED) -- the other
one (ECONNRESET) does not appear to be used in the perl core as far as I can
tell.
Steve Hay [Thu, 5 Sep 2013 14:18:20 +0000 (15:18 +0100)]
Fix stringification of $! in VC10+ builds where errno > sys_nerr
VC++ 2010 and above define a "POSIX supplement" of errno values ranging from
EADDRINUSE (100) to EWOULDBLOCK (140), but sys_nerr is still 43 and strerror()
returns "Unknown error" for them. We already avoid using strerror() if errno
> sys_nerr, but we treat such values as Windows error codes (i.e. <winerror.h>
values) and look up the corresponding system messages for them. There is no
better plan for these POSIX supplement errno values, but they must be
converted from <errno.h> values to <winerror.h> values first otherwise we
will look up the wrong system message. In practice, we only expect to find
errno > sys_nerr in the case of Windows sockets errors (we used to assign
WSAGetLastError() to errno), so we simply convert Exxx values to WSAExxx
values where possible, and use a default <winerror.h> value otherwise (namely,
ERROR_INVALID_FUNCTION).
This change fixes code such as this:
perl -le "$!=107; print $!"
which now outputs the expected "No connection could be made because the
target machine actively refused it." rather than "The program stopped
because an alternate diskette was not inserted." (in VC10+ builds).
Also: Fix the spelling of ECANCELED.
Steve Hay [Wed, 4 Sep 2013 17:20:03 +0000 (18:20 +0100)]
Intercept assignment to $! to translate WSAExxx values to Exxx values on Windows
Since perl previously assigned WSAExxx values to $! on Windows it is quite
possible that (Perl-level) user code may also manually make similar assignments,
which will now cause breakage if the value put in $! is subsequently compared to
Errno/POSIX constants because the latter are now the corresponding Exxx values
where possible.
An example of this is in Net::Ping::tcp_connect(), which does the following to
fetch a socket-level error code:
unpack("i", getsockopt($self->{"fh"}, SOL_SOCKET, SO_ERROR))
and assigns the result (a WSAExxx value such as 10061) to $! and then goes wrong
in the subsequent test (in ping_tcp()) for $! == ECONNREFUSED (which is now 107
rather than 10061 if perl is built with VC10 or higher).
To avoid this we now intercept assignment to $! and convert any WSAExxx values
to Exxx values first. This causes a minor oddity in that this:
perl -le "$! = 10061; print 0+$!"
will now output 107 (for VC10+ perls) but this is surely preferable to the
alternative breakage described above.
Steve Hay [Tue, 3 Sep 2013 21:57:46 +0000 (22:57 +0100)]
Fix IO::Socket::connect() in the light of $! changes
We now assign Exxx values rather than WSAExxx values to $! in the event of a
sockets error if there is an Exxx value corresponding to the WSAExxx error.
This is indeed the case for several Exxx values < 100, amongst them EINVAL (22),
which corresponds to WSAEINVAL (10022), which IO::Socket::connect() has a test
for. Note that the test used a hard-coded value for WSAExxx since there were
(and still are) no constants in Perl for that. Users will now be able to test
for EINVAL instead, but no existing code will be using that because it has
always had the value 22.
This non-backwards-compatible breakage seems unavoidable if we are to get things
in a sane state for the future because it is really a problem with how things
have long been, namely that $! was being set to winsock2.h values, but if there
was a corresponding errno.h value and it was < 100 then *that* was used for the
Errno/POSIX constant, so authors had no good way of testing for such $! values.
The new scheme corrects that by consistently using the same values for $! and
for Errno/POSIX [except in the case of third-party code setting its own $!/errno
values, which will be addressed in the next commit].
(Ironically, Exxx values >= 100 (the ones which caused all this trouble in the
first place!) don't have this problem as long as code uses the Exxx constants
rather than the hard-coded 100XX numbers. For example, if code tests $! ==
ECONNABORTED, where $! and ECONNABORTED were both previously set to winsock2.h
values then the same test will still work, except that now $! and ECONNABORTED
are both set to errno.h values in VC10+ (or both left as winsock2.h values upto
VC9). It does still break if any code is directly testing $! == 10053, but
that's not such a problem: that's really broken code, which should be using the
constant instead.)
Steve Hay [Tue, 3 Sep 2013 21:57:29 +0000 (22:57 +0100)]
Fix a problem with mod_perl on Windows using VS2010+.
The problem is caused by the following two commits, which sought to deal with
new Exxx values (with values >= 100) in errno.h which Microsoft added in VS2010:
http://perl5.git.perl.org/perl.git/commit/
b59e75b34cef3fedd214c9b6ee744146cf8b3308
http://perl5.git.perl.org/perl.git/commit/
912c63ed00375338703043928cac3c740d00cc9d
The former commit was mostly a patch to Errno and POSIX, together with some
other cleaning up in win32/win32.h and win32/include/sys/socket.h; the latter
commit was a fixup to win32/include/sys/socket.h which restored (more
aggressively) the ENOTSOCK->WSAENOTSOCK redefinition and added similar
redefinitions for ECONNABORTED, ECONNRESET and EAFNOSUPPORT.
It is the latter commit which causes this little program to output
ECONNABORTED=10053 rather than ECONNABORTED=106 as it ought to do in VC10 (and
higher) builds of perl:
#include <windows.h>
#include "EXTERN.h"
#include "perl.h"
#include "XSUB.h"
void main(void) {
printf("ECONNABORTED=%d\n", ECONNABORTED);
}
That change is now causing problems with mod_perl, in which the (Perl level)
APR::Status::is_ECONNABORTED() and the (C level) APR_STATUS_IS_ECONNABORTED()
which it calls are failing to recognize an aborted connection (indicated by
error code ECONNABORTED set by Apache httpd.exe).
The APR_STATUS_IS_ECONNABORTED() macro is picked up by mod_perl from APR's
apr_errno.h:
#define APR_STATUS_IS_ECONNABORTED(s) ((s) == APR_ECONNABORTED \
|| (s) == APR_OS_START_SYSERR + WSAECONNABORTED)
where
#ifdef ECONNABORTED
#define APR_ECONNABORTED ECONNABORTED
#else
#define APR_ECONNABORTED (APR_OS_START_CANONERR + 18)
#endif
When this is compiled into httpd.exe ECONNABORTED is 106 (from errno.h) and
APR_STATUS_IS_ECONNABORTED(s) amounts to
#define APR_STATUS_IS_ECONNABORTED(s) ((s) == 106 || (s) == 730053)
but when compiled into APR/Status.dll ECONNABORTED is 10053 (redefined to
WSAECONNABORTED as above) so APR_STATUS_IS_ECONNABORTED(s) then amounts to
#define APR_STATUS_IS_ECONNABORTED(s) ((s) == 10053 || (s) == 730053)
which doesn't pick up an error code of 106 coming from httpd.exe.
This could be worked around in mod_perl by redefining ECONNABORTED and the other
three back to their original errno.h values to match what httpd.exe is using,
but that might just cause problems in the other direction, with those values no
longer matching the values which perl.exe is using.
Moreoever, this problem could affect other XS interfaces (not just mod_perl), so
it really needs to be fixed in perl.
This commit implements the alternative solution mentioned the commit message for
the first commit cited above. (As noted in that previous commit message, this
solution equally has potential problems, missing out on the advantages of the
original solution implemented, namely, better backwards compatibility with other
perl code having hard-coded numeric error codes, but this will be unavoidable if
we want to get to a sane state.)
Note that changing the $! values is an incompatible change, so should probably
not be done for 5.18.x. That's unfortunate for mod_perl (and anything else
similarly affected), but that will just have to either live with the breakage
(which was introduced in 5.14.0) until 5.20.0 or else try the workaround
mentioned above for 5.14.x, 5.16.x and 5.18.x.
The main change is to use a new function throughout win32/win32sck.c to convert
WSAGetLastError() values into errno.h constants before assigning to errno. Every
possible WSAExxx value (as documented by MSDN) is mapped to an Exxx value,
although it is still possible for other WSAxxx values to get assigned to errno
unchanged (but that has always been the case, and any non-existent Exxx values
get mapped back to WSAExxx values anyway...).
We must then ensure that all of those Exxx values are defined, which is now done
(in the new file, win32/include/sys/errno2.h) in a friendlier manner, being
careful not to redefine any (specifically the new ones in VC++ 2010 and above)
which already exist. The rest are defined as the WSAExxx values, as mentioned
above.
Finally, we need the Errno module to know about these values, which is done by
having it include that same new header file to ensure that it gets the same
definitions, rather than having to play its own games. The new header is also
used in POSIX, which similarly wants definitions of as many as possible of its
hard-coded list of Exxx values.
H.Merijn Brand [Mon, 16 Sep 2013 12:47:08 +0000 (14:47 +0200)]
Regenerate Configure and chainsaw voidflags
Nicholas Clark [Mon, 16 Sep 2013 10:00:42 +0000 (12:00 +0200)]
Merge changes which eliminate VOIDFLAGS and VOIDUSED from active code.
The remaining uses are in Configure and its related generated files, along
with canned config.sh files and configure.com. The next regeneration of
Configure from metaconfg should eliminate nearly all of these, at which point
the remainder can be removed by hand.
Brian Fraser [Wed, 11 Sep 2013 10:22:45 +0000 (12:22 +0200)]
Eliminate the use of voidflags from Configure's pointer size determination.
This should permit metaconfig to drop the entire voidflags logic from
Configure and the config.sh related files.
Nicholas Clark [Wed, 11 Sep 2013 09:54:42 +0000 (11:54 +0200)]
Eliminate the only use of VOIDFLAGS, as part of STMT_START in perl.h
STMT_START has used VOIDFLAGS as part of its conditional compilation since it
was added by commit
728e280303ae3d3b (March 1996). The code originally read:
/* Now which other defined()s do we need here ??? */
Since then it has been amended to avoid entering that definition for GCC
on Solaris. Given that all current Solaris compilers are C89 conformant,
VOIDFLAGS will always be true (actually 15), so the test is redundant.
Even back in 1996, it's possible that VOIDFLAGS was always non-zero on
SunOS, rendering the test obsolete from the start.
Spotted by Brian Fraser, and extracted from a larger patch of his.
Nicholas Clark [Wed, 11 Sep 2013 07:14:36 +0000 (09:14 +0200)]
Stop using VOIDUSED in the headers.
VOIDUSED was provided by metaconfig to work around pre-ANSI compilers that
didn't support all uses of the void type. We've been mandating a proper ANSI
C89 compiler for at least a decade, so we don't need this any more.