Jess Robinson [Sat, 29 Dec 2012 21:26:27 +0000 (21:26 +0000)]
Set appropriate flags for linux or linux-like systems.
This means that osnames that look like *linux*, like linux-androideabi
or linux-gnueabihf are treated like linux systems.
Father Chrysostomos [Fri, 17 Jan 2014 14:30:57 +0000 (06:30 -0800)]
sv_buf_to_rw can be static
sv_buf_to_ro needs to be non-static because op.c uses it, but
sv_buf_to_rw is only called from sv.c.
Father Chrysostomos [Fri, 17 Jan 2014 14:11:02 +0000 (06:11 -0800)]
Always define sTHX as 0 when not using mem debug header
I broke the threaded build in
b001a0d149 by assuming that sTHX was
always set to 0 when memory debug headers were not in use. That is
an easy assumption to make, so this commit makes that assumption true.
(And there is no reason it cannot be made true. There is no need for
sTHX to be set under PERL_IMPLICIT_CONTEXT but not PERL_TRACK_MEMPOOL,
where sTHX was not even used before
b001a0d149.)
Father Chrysostomos [Fri, 17 Jan 2014 02:03:35 +0000 (18:03 -0800)]
[Merge] Cow Tools
This branch provides a PERL_DEBUG_READONLY_COW mode that turns COW
violations into crashes. It also add documentation to perlguts
explaining how to handle COW strings in XS code.
There are two ‘known’ failures in IO::Compress’ tests. A patch has
been sent upstream.
(Apologies to Gary Larson.)
Father Chrysostomos [Sun, 12 Jan 2014 00:45:18 +0000 (16:45 -0800)]
util.c: A couple of defines to simplify #ifdef maze
No real simplification in terms of the number of #ifdefs, but more
readability (hopefully).
Father Chrysostomos [Fri, 3 Jan 2014 13:58:22 +0000 (05:58 -0800)]
Skip t/re/uniprops.t under PERL_DEBUG_READONLY_COW
Under Linux, using mmap for all memory allocation stops working when we
get up to 70,000 lines of code. So just skip it.
Father Chrysostomos [Thu, 2 Jan 2014 20:39:05 +0000 (12:39 -0800)]
SvGROW should un-cow under PERL_OLD_COPY_ON_WRITE
Otherwise pp_uc (and presumably other pieces of code) will end up mod-
ifying shared buffers.
Brought to you by PERL_DEBUG_READONLY_COW.
Father Chrysostomos [Mon, 30 Dec 2013 00:47:43 +0000 (16:47 -0800)]
perldiag for new mprotect warnings
Father Chrysostomos [Mon, 30 Dec 2013 00:39:36 +0000 (16:39 -0800)]
Teach podcheck about mmap(2)
Father Chrysostomos [Sat, 28 Dec 2013 14:06:27 +0000 (06:06 -0800)]
COW documentation
plus read-only documentation, since hysterically the two are
intertwined.
Father Chrysostomos [Thu, 12 Dec 2013 23:26:01 +0000 (15:26 -0800)]
XS::APItest: Flatten src for utf16_to_utf8_reversed
utf16_to_utf8_reversed (the C function) modifies its input, so the XS
function of the same name should flatten the input to avoid modifying
shared buffers.
Brought to you by PERL_DEBUG_READONLY_COW.
Father Chrysostomos [Mon, 6 Jan 2014 00:20:58 +0000 (16:20 -0800)]
threads.t: Increase watchdog timeout
PERL_DEBUG_READONLY_COW is too slow for a 60s timeout.
Father Chrysostomos [Mon, 6 Jan 2014 00:09:35 +0000 (16:09 -0800)]
Make XS::APItest::establish_cleanup protect existing stacks
It causes pp_entersub to be called in odd places, which can cause the
context stack to be reallocated when an outer function call (like
pp_leavesub) has a pointer into the context stack in a C auto.
cleanup.t was failing for me under PERL_DEBUG_READONLY_COW +
STRESS_REALLOC + threads, because the context stack was reallocated
and the old address then freed and reused for something else, being
zeroed in the mean time. So pp_leavesub returned NULL (trying to
read retop from the context stack), causing the program to exit.
During global destruction, subs that had not be exited properly were
undefined, leading to:
1..3
Can't undef active subroutine during global destruction.
Father Chrysostomos [Wed, 4 Dec 2013 12:39:14 +0000 (04:39 -0800)]
PERL_DEBUG_READONLY_COW
Make perls compiled with -Accflags=-DPERL_DEBUG_READONLY_COW to turn
COW buffer violations into crashes.
We do this using mmap to allocate memory and then mprotect to mark
memory as read-only when buffers are shared.
We have to do this at the safesysmalloc level, because some code does
SvPV_set with buffers it allocates on its own via safemalloc().
Unfortunately this means many things are allocated using mmap that
will never be marked read-only, slowing things down considerably, but
I see no other way.
Because munmap and mprotect need to know the length, we use the
existing sTHX/perl_memory_debug_header mechanism used already by
PERL_TRACK_MEMPOOL and store the size there (as PERL_POISON already
does when PERL_TRACK_MEMPOOL is enabled). perl_memory_debug_header is
a struct positioned at the beginning of every allocated buffer, for
tracking things.
Chris 'BinGOs' Williams [Thu, 16 Jan 2014 20:06:49 +0000 (20:06 +0000)]
Update Module-Load-Conditional to CPAN version 0.60
[DELTA]
0.60 Thu Jan 16 12:28:24 GMT 2014
* Added autoload option to can_load()
Karl Williamson [Thu, 16 Jan 2014 17:26:16 +0000 (10:26 -0700)]
regcomp.c: Remove unnecessary variable
This variable is set and tested just once in all but one instance,
before being discarded. In that one instance, it's tested twice. The
underlying expression is just extracting a field from a structure.
There's no efficiency gain to using the variable, so might as well not
have it.
Karl Williamson [Thu, 16 Jan 2014 17:23:31 +0000 (10:23 -0700)]
Turn on read-only flag for some unchangeable inversion lists
These lists are read-only. Turning on the flag may allow some
optimisations to be done, including some that may be added in the
future.
Chris 'BinGOs' Williams [Thu, 16 Jan 2014 16:20:08 +0000 (16:20 +0000)]
Update known POD issues
Chris 'BinGOs' Williams [Thu, 16 Jan 2014 15:49:51 +0000 (15:49 +0000)]
Update Scalar-List-utils to CPAN version 1.36
[DELTA]
1.36 -- 2014/01/16 15:40:47
[CHANGES]
* Added Scalar::Util::unweaken()
* Various documentation changes/updates
[BUGFIXES]
* Correct uses of overload operators in unit tests (RT91969)
Tony Cook [Thu, 16 Jan 2014 04:37:27 +0000 (15:37 +1100)]
[perl #89502] improved FreeMINT support
Tony Cook [Thu, 16 Jan 2014 04:11:50 +0000 (15:11 +1100)]
describe the level of FreeMINT support
Tony Cook [Fri, 15 Nov 2013 00:56:46 +0000 (11:56 +1100)]
add Alan Hourihane to AUTHORS
Alan Hourihane [Fri, 15 Nov 2013 00:53:16 +0000 (11:53 +1100)]
new freemint hints file
Alan Hourihane [Fri, 15 Nov 2013 00:37:52 +0000 (11:37 +1100)]
freemint doesn't need the extra -lm -lposix either
Alan Hourihane [Fri, 15 Nov 2013 00:30:50 +0000 (11:30 +1100)]
fix some linuxisms and RETVAL warnings from xsubpp
Original patch by Alan, modified by Tony Cook to remove unused return
types.
Tony Cook [Mon, 11 Nov 2013 00:45:40 +0000 (11:45 +1100)]
[perl #89502] support for dld on Atari FreeMINT
Tony Cook [Thu, 16 Jan 2014 04:35:45 +0000 (15:35 +1100)]
fix a POD error
Ricardo Signes [Thu, 16 Jan 2014 04:26:07 +0000 (23:26 -0500)]
correct release schedule: next dev series is 5.21
Ricardo Signes [Thu, 16 Jan 2014 04:24:42 +0000 (23:24 -0500)]
update the release schedule
Tony Cook [Thu, 16 Jan 2014 02:38:21 +0000 (13:38 +1100)]
someone forgot to test_porting and it was me
add another strange email address to checkAUTHORS.pl
Tony Cook [Tue, 10 Dec 2013 22:48:15 +0000 (09:48 +1100)]
[perl #120670] make perl headers C++11 compatible
Tony Cook [Wed, 15 Jan 2014 22:48:10 +0000 (09:48 +1100)]
Tony Cook [Wed, 15 Jan 2014 22:40:43 +0000 (09:40 +1100)]
[perl #121010] Fix base.pm nonexistent module check with open files
Tony Cook [Wed, 15 Jan 2014 22:34:34 +0000 (09:34 +1100)]
bump $base::VERSION
Dagfinn Ilmari Mannsåker [Tue, 14 Jan 2014 14:18:52 +0000 (14:18 +0000)]
Fix base.pm nonexistent module check with open files
Tony Cook: update MANIFEST
Chris 'BinGOs' Williams [Wed, 15 Jan 2014 19:56:19 +0000 (19:56 +0000)]
Update Perl-OSType to CPAN version 1.007
[DELTA]
1.007 2014-01-15 09:45:41-05:00 America/New_York
[ADDED]
- Added 'android' as a Unix-type OS (Piotr Roszatycki)
Ricardo Signes [Wed, 15 Jan 2014 14:49:04 +0000 (09:49 -0500)]
fix a typo in a comment
Spotted by Matthew Horsfall
Brian Fraser [Wed, 15 Jan 2014 12:14:46 +0000 (09:14 -0300)]
Configure: default incpth to usrinc
A previous commit made findhdr use incpth instead of just usrinc;
however, at the moment, incpth is not being filled for non-gcc
compatible compilers, and so broke the build for those.
This commit makes incpth default to usrinc, which means that
if for whatever reason incpth is not filled, findhdr will still
function as it did before.
Tony Cook [Wed, 15 Jan 2014 03:35:25 +0000 (14:35 +1100)]
[perl #120933] static extensions now depend on nonxs extensions
This means that changes to static extensions like Win32CORE actually
result in them being rebuilt.
Tony Cook [Tue, 14 Jan 2014 22:58:10 +0000 (09:58 +1100)]
Tony Cook [Mon, 13 Jan 2014 05:20:00 +0000 (16:20 +1100)]
[perl #118843] work around recv() behaviour on cygwin
cygwin inherits recv behaviour from the Win32 sockets API which doesn't
modify the namebuf or it's associated size when you recv() from a
connected socket, handle this the same way Win32 does by zeroing the
length if it's the same as before calling recv().
Also adds some basic socket function tests to the core tests.
Tony Cook [Tue, 14 Jan 2014 21:58:03 +0000 (08:58 +1100)]
Tony Cook [Mon, 13 Jan 2014 22:24:40 +0000 (09:24 +1100)]
[perl #118817] avoid using 2 handles to write to the debug output
Previously the tests were run with the following config:
NonStop=0 TTY=db.out LineInfo=db.out
This meant that the debugger would write the prologue, command prompts
and their results and the epilogue to one handle, and any line trace
information to the second handle. Since those handles didn't share a
file position, the line trace info would overwrite the prologue, and
the epilogue would overwrite part of the line trace info.
When TTY=vt100 on Redhat systems this made the epilogue just long
enough to overwrite the line trace data that a test matched against,
causing the test to fail.
To fix this, I avoided setting LineInfo:
NonStop=0 TTY=db.out
and since LineInfo defaults to using the TTY handle, both types of
content are written to db.out *without* overwriting each other.
Unfortunately this broke some other tests, since the command prompts
which were overwritten by line trace information are now mixed in with
the line traces - I've modified the tests that failed to account for
the included command lines.
Ricardo Signes [Tue, 14 Jan 2014 14:33:14 +0000 (09:33 -0500)]
Merge branch 'experimental-autoderef' into blead
Ricardo Signes [Tue, 14 Jan 2014 14:13:53 +0000 (09:13 -0500)]
perldelta for experimental::autoderef
Ricardo Signes [Tue, 14 Jan 2014 13:34:19 +0000 (08:34 -0500)]
rename aggref warnings to autoderef
Ricardo Signes [Tue, 14 Jan 2014 13:30:46 +0000 (08:30 -0500)]
avoid a keys-on-scalar warning in a test
Father Chrysostomos [Sat, 9 Nov 2013 14:56:42 +0000 (06:56 -0800)]
More test tweaks
Father Chrysostomos [Sat, 9 Nov 2013 14:31:58 +0000 (06:31 -0800)]
Fix copy & paste error in tests
Father Chrysostomos [Sat, 9 Nov 2013 14:30:32 +0000 (06:30 -0800)]
Increase $warnings::VERSION to 1.21
Father Chrysostomos [Sat, 9 Nov 2013 14:30:03 +0000 (06:30 -0800)]
Make key/push $scalar experimental
We need a better name for the experimental category, but I have not
thought of one, even after sleeping on it.
Father Chrysostomos [Tue, 14 Jan 2014 14:30:23 +0000 (06:30 -0800)]
Get t/io/utf8.t working under PERL_UNICODE
Father Chrysostomos [Tue, 14 Jan 2014 06:34:21 +0000 (22:34 -0800)]
Use NOT_REACHED in one spot in hv.c
This reduces the size of hv.o by 32 bytes under clang.
Tony Cook [Tue, 14 Jan 2014 05:54:29 +0000 (16:54 +1100)]
use the more portable File::Spec->devnull for the stderr redirect
this allows pending-author.t to work on MSWin32
Tony Cook [Tue, 14 Jan 2014 05:53:40 +0000 (16:53 +1100)]
[perl #116971] avoid feeding manisort a bad @INC
the C<< use TestInit qw(T) >> in manifest.t both changes to the
build tree *and* initializes $ENV{PERL5LIB} to "lib", so there's
no need fo runperl() to supply (an incorrect) -I.. when running
manisort.
Tony Cook [Tue, 14 Jan 2014 04:46:27 +0000 (15:46 +1100)]
[perl #116971] avoid feeding utils run by utils.t a bad @INC
the C<< use TestInit qw(T) >> both changes to the build tree root
*and* initializes $ENV{PERL5LIB} to "lib", so there's no need for
runperl() to supply -I.. when testing each tool.
Tony Cook [Tue, 14 Jan 2014 04:14:42 +0000 (15:14 +1100)]
Shlomi Fish [Mon, 13 Jan 2014 12:16:29 +0000 (14:16 +0200)]
Add examples for IO::Socket::UNIX.
In the SYNOPSIS.
See
http://www.nntp.perl.org/group/perl.perl5.porters/2014/01/msg211274.html:
<QUOTE>
From: Shlomi Fish
Subject: Patch for perldoc IO::Socket::UNIX
I noticed http://perldoc.perl.org/IO/Socket/UNIX.html does not contain any
examples (including not in the synopsis). If acceptable, I can a prepare a
patch that will include a simple server and a client for UNIX domain sockets,
see: http://www.mail-archive.com/beginners%40perl.org/msg116942.html
</QUOTE>
TC: modified the version number modification
Shlomi Fish [Mon, 13 Jan 2014 11:54:00 +0000 (13:54 +0200)]
Remove an old note about autoflush from the POD.
For IO::Socket::UNIX. See:
http://www.nntp.perl.org/group/perl.perl5.porters/2014/01/msg211275.html
<QUOTE>
And while you're at it, you might as well get rid of the autoflush note,
given IO::Socket 1.18 being fairly ancient (it shipped with 5.6.0).
</QUOTE>
Brian Fraser [Tue, 14 Jan 2014 03:19:08 +0000 (00:19 -0300)]
perldelta for -Dsysroot
Brian Fraser [Fri, 3 Jan 2014 04:32:09 +0000 (01:32 -0300)]
Configure, sysroot: failing to guess usrinc needn't be fatal
Previously, if we failed to guess usrinc, incpth, or libpth and were
cross-compiling, Configure would've bailed out immediately. This commit
makes it more lenient if using -Dsysroot; in case of failure it will now
warn, but Configure will continue as usual; this is because the defaults
for those variables will now use sysroot, so they have a higher chance
of being accurate.
Brian Fraser [Thu, 2 Jan 2014 03:17:17 +0000 (00:17 -0300)]
Configure, sysroot: Update defaults to use $sysroot
Brian Fraser [Mon, 30 Dec 2013 05:33:05 +0000 (02:33 -0300)]
Configure, sysroot: add --sysroot to cppflags
Jess Robinson [Fri, 19 Oct 2012 18:05:25 +0000 (19:05 +0100)]
Introduce sysroot variable to Configure
This is borrowed from gcc and allows us to indicate the logical root
directory for headers and libraries, under which all -I and -L are
searched for. This patch adjusts Configure to search under $sysroot
(if supplied) for headers and libraries, instead of /.
--sysroot is added to ccflags and friends so that make in
ExtUtils::MakeMaker, and other extensions, will use it.
Currently this is only done if compiling with some variant of gcc
or g++.
Brian Fraser [Sat, 11 Jan 2014 05:50:47 +0000 (02:50 -0300)]
Configure: use incpth, not usrinc, to search for headers.
incpth is all the paths that will eventually be used by the compiler to
search for headers; usrinc is the first of those paths to have all three
of errno.h, stdio.h, and time.h inside.
Brian Fraser [Tue, 31 Dec 2013 16:33:42 +0000 (13:33 -0300)]
Configure: Change a use of cc to $cc
Brian Fraser [Wed, 1 Jan 2014 22:55:33 +0000 (19:55 -0300)]
Configure: Add $cppflags to cppstdin
Brian Fraser [Tue, 31 Dec 2013 15:30:41 +0000 (12:30 -0300)]
Configure: Better guesses for usrinc and friends on some compilers
It turns out that we had some much more accurate way of getting
usrinc, libpth and incpth -- we can just ask the compiler. This
was stashed away and only used when cross-compiling; what this
commit does is move things around so that it's also used in the
more common case.
Father Chrysostomos [Mon, 13 Jan 2014 00:47:31 +0000 (16:47 -0800)]
Make getc unset the utf8 flag on its retval
The same scalar (the target) is used to return the value each time the
same getc operator is called. It was turning on the utf8 flag when
necessary, but not turning it off.
(I suspect recv has the same bug. I haven’t confirmed.)
Tony Cook [Mon, 13 Jan 2014 02:01:14 +0000 (13:01 +1100)]
Ricardo Signes [Mon, 13 Jan 2014 00:18:40 +0000 (19:18 -0500)]
correct a Pod warning in the new perldelta template
(cherry picked from commit
1332f10946560bb419517a581b7f56982fcc659f)
Daniel Dragan [Sun, 12 Jan 2014 08:39:50 +0000 (03:39 -0500)]
reduce size of const static tables in time64.c
The data in the const static arrays can be expressed with smaller
datatypes. This reduces binary size slightly.
Father Chrysostomos [Sun, 12 Jan 2014 15:55:30 +0000 (07:55 -0800)]
Use NOT_REACHED in one more place in op.c
Under clang, this reduces the size of op.o by 16 bytes.
$ clang -v
Apple LLVM version 4.2 (clang-425.0.27) (based on LLVM 3.2svn)
Target: x86_64-apple-darwin12.5.0
Thread model: posix
Father Chrysostomos [Fri, 10 Jan 2014 13:53:19 +0000 (05:53 -0800)]
pp_ctl.c:pp_require: Remove redundant VMS null checks
SvPVX(newSVpv("",...)) should never return null.
Father Chrysostomos [Sat, 11 Jan 2014 15:00:06 +0000 (07:00 -0800)]
pp_sys.c:S_doform: remove redundant null check
This static function is called from exactly two places that both die
aforehand if cv is null. Further, the cv parameter is already marked
as non-null.
Father Chrysostomos [Sat, 11 Jan 2014 06:05:35 +0000 (22:05 -0800)]
t/re/pat.t: Keep minitest passing
Father Chrysostomos [Sat, 11 Jan 2014 06:01:43 +0000 (22:01 -0800)]
Add Russel O'Connor to AUTHORS
Apparently he was the original author of commit
b0d9ce3858 (Oct ’99).
Chris 'BinGOs' Williams [Sat, 11 Jan 2014 17:39:02 +0000 (17:39 +0000)]
Update Module-Build to CPAN version 0.4204
[DELTA]
0.4204 - Fri Jan 10 00:29:31 CET 2014
[BUG FIXES]
- Map conflicts back to runtime [Leon Timmermans]
- Use mod2fname whenever it's available [Leon Timmermans, Brian Fraser]
- Accept custom entries in meta_merge 1.4
Father Chrysostomos [Fri, 10 Jan 2014 14:17:24 +0000 (06:17 -0800)]
perldiag: Rewrap an entry for better splain output
Father Chrysostomos [Fri, 10 Jan 2014 14:10:43 +0000 (06:10 -0800)]
perldiag: Use (F) consistently
and explain in the prose when there is an error. One entry already
explained this.
Father Chrysostomos [Fri, 10 Jan 2014 13:59:39 +0000 (05:59 -0800)]
Fix require’s get-magic handling for @INC elements
It was only calling get-magic before checking whether the argument was
a reference if the array was tied, which is not the only thing that
can cause an @INC element to have get-magic. It should have been
checking for get-magic on the element itself (which is a faster
check, too).
And then there were too many FETCH calls.
I do not know whether we should be calling get-magic exactly once
when the ‘Can’t locate’ error occurs. At least this commit reduces
the number of FETCHes.
Brian Gottreu [Fri, 10 Jan 2014 19:36:32 +0000 (13:36 -0600)]
Correct capitalization of Devel::ptkdb in perldebug
Craig A. Berry [Fri, 10 Jan 2014 13:50:36 +0000 (07:50 -0600)]
Don't multiply define _new_invlist_C_array.
It's now an external function in regcomp.c, so it doesn't need to
also be defined in re_comp.c. And in fact it mustn't be as the VMS
linker warns with:
%ILINK-W-MULDEF, symbol PERL__NEW_INVLIST_C_ARRAY multiply defined
module: RE_COMP
which becomes a run-time error.
Karl Williamson [Fri, 10 Jan 2014 05:06:24 +0000 (22:06 -0700)]
regexec.c: Don't skip trying POSIX class if [] has individ char
commit
b99851e1941e002dd4816ee6c76fd49bbee1d7f3
Author: Karl Williamson <public@khwilliamson.com>
Date: Wed Dec 4 10:06:04 2013 -0700
PATCH: [perl #120675] Unexpected tainting via regex using locale
introduced a bug in which under /li matching a bracketed character class
that contains both characters and POSIX classes would skip matching
against any POSIX class. The example in the added test illustrates the
problem:
"h" =~ /[g\w]/li; # Failed! in C locale
Daniel Dragan [Thu, 9 Jan 2014 04:44:16 +0000 (23:44 -0500)]
fix missing _rotl64 symbol on Visual C 2003
Due to a bug in the CRT (msvcr71.dll), these 2 functions are not defined
in any lib Perl can use (static link CRTs dont apply, Perl only uses DLL
CRTs), but they are available as intrinsics. This solves a link error
about missing symbol __rotl64 in hv.obj, from usage in hv_func.h, on 32 bit
USE_64_BIT_INT VC 2003 builds. _rotr64 is included for completeness. This
fix is filed as [perl #120925].
Karl Williamson [Thu, 9 Jan 2014 21:08:07 +0000 (14:08 -0700)]
Merge branch into blead to compile into C all POSIX classes...
This series of commits creates C compile time inversion lists for the
remaining POSIX classes that didn't have them, plus IDStart and IDCont
which are used with the 'utf8' pragma, plus one that is used for
compiling regular expressions.
This should allow faster start up times for Perl programs that use these
on code points 256 and above, while having negligible effect on programs
that don't. It also leads to some simplification of core code, and
removal of two arrays of smaller inversion lists.
Karl Williamson [Tue, 7 Jan 2014 17:39:19 +0000 (10:39 -0700)]
regexec.c: Use compiled-in POSIX definitions
This changes the regex engine to not go out to the disk when needing to
find the definitions of POSIX classes, but to use the compiled-in ones
introduced earlier in this commit series.
Karl Williamson [Mon, 6 Jan 2014 20:41:46 +0000 (13:41 -0700)]
IDStart and IDCont no longer go out to disk
These are the base names for various macros used in parsing identifiers.
Prior to this patch, parsing a code point above Latin1 caused loading
disk files. This patch causes all the information to be compiled into
the Perl binary.
Karl Williamson [Mon, 6 Jan 2014 19:22:02 +0000 (12:22 -0700)]
isWORDCHAR_uni(), isDIGIT_utf8() etc no longer go out to disk
Previous commits in this series have caused all the POSIX classes to be
completely specified at C compile time. This allows us to revise the
base function used by all these macros to use these definitions,
avoiding reading them in from disk.
Karl Williamson [Mon, 6 Jan 2014 19:14:31 +0000 (12:14 -0700)]
Move initialization of PL_XPosix_ptrs[] to perl.c
This was performed unconditionally in regcomp.c. However, future
commits will use this from other code. Almost all (but not completely
all) Perl code uses regular expressions, so only rarely will this small
amount of initialization be performed when it currently isn't.
Karl Williamson [Mon, 6 Jan 2014 18:57:53 +0000 (11:57 -0700)]
regen/mk_invlists.pl: White-space only
This outdents a block to be in line with adjacent lines.
Karl Williamson [Mon, 6 Jan 2014 18:55:17 +0000 (11:55 -0700)]
regcomp.c: Reword expression for clarity
I believe the new version is clearer as to what is meant, and it brings
it in line with the same expression in nearby uses.
Karl Williamson [Mon, 6 Jan 2014 18:52:21 +0000 (11:52 -0700)]
Rmv PL_Posix_ptrs
Previous commits in this series have removed all uses of this global
array. This completely removes it.
Since it is a global, consideration need be given to possible uses of it
outside the core. It has never been externally documented, and is an
opaque structure whose internals have changed with every release. The
functions used to access it are almost all static to regcomp.c; those
few that aren't have been hidden from all but the few .c files that need
to have access to them, via #if's.
Karl Williamson [Mon, 6 Jan 2014 18:41:53 +0000 (11:41 -0700)]
regcomp.c: Rmv remaining uses of PL_Posix_ptrs
Previous commits have removed all but a few uses of PL_Posix_ptrs. This
removes the rest. ASCII is the same whether over all code points, or
just the ASCII range, so we can substitute the version for all code
points. There is an extra intersection introduced by this commit during
the construction of a synthetic start class under /a and /aa, but the
performance hit should be negligible.
Karl Williamson [Mon, 6 Jan 2014 18:36:49 +0000 (11:36 -0700)]
regcomp.c: Collapse two branches.
Previous commits in this series have removed the need to special case
[:ascii:]. This commit removes the special casing. There is a slight
performance penalty if this is the only POSIX class in the bracketed
class, and is being compiled under /a or /aa: An extra intersection will
be performed. Since this is regex compilation, this should be
unnoticeable.
Karl Williamson [Sun, 5 Jan 2014 06:13:56 +0000 (23:13 -0700)]
regcomp.c: Trade a little time for simplicity
Perl currently has two sets of inversion lists for the POSIX classes
built in. One set is for the entire Unicode range; the other for just
the ASCII range. This latter set could be derived from the larger one
at run time by doing an intersection with ASCII. This commit starts the
process of entirely removing the second set, thus avoiding its
bookkeeping. This commit avoids one use of the ASCII set, instead
adding an intersection with ASCII after all the POSIX classes in a
bracketed character set have been combined. Thus the penalty is one
intersection per compilation of each bracketed character classes that
contain POSIX classes.
Karl Williamson [Sun, 5 Jan 2014 04:12:07 +0000 (21:12 -0700)]
regcomp.c: Collapse two code branches
Previous commits have simplified things so these two if-then-else
branches can be collapsed into one.
Karl Williamson [Sat, 4 Jan 2014 22:31:36 +0000 (15:31 -0700)]
Keep separate list of posix classes and complements
In building up the list of code points that are matched by a bracketed
character class, there can be both posix classes and complemented posix
classes, like [\s\W]. By keeping each type in a separate list, we can
simplify code in later commits.
Karl Williamson [Sat, 4 Jan 2014 21:45:54 +0000 (14:45 -0700)]
regcomp.c: White-space, comments only
This outdents code due to the removal of a block in the previous commit.
And it clarifies some comments about it.
Karl Williamson [Sat, 4 Jan 2014 21:43:16 +0000 (14:43 -0700)]
regcomp.c: Collapse two code branches
The previous commit has enabled us to collapse the branches for dealing
with, e.g., \w and \W into the same branch, using a flag to indicate to
complement or not.