Join GitHub today
GitHub is home to over 50 million developers working together to host and review code, manage projects, and build software together.
Sign uppack doesn't work in new gcc 4.9.1 compiler #14554
Comments
From @khwilliamsonThis is a bug report for perl from khw@khw.(none), See http://nntp.perl.org/group/perl.perl5.porters/225619 Tony Cook says "This looks like another "rubbish in the packing" issue." # Failed test 4311 - at op/pack.t line 1127 Flags: Site configuration information for perl 5.21.10: Configured by khw at Mon Mar 2 08:04:33 MST 2015. Summary of my perl5 (revision 5 version 21 subversion 10) configuration: Locally applied patches: @INC for perl 5.21.10: /home/khw/blead/lib/perl5/site_perl/5.21.10/x86_64-linux-thread-multi-ld Environment for perl 5.21.10: PATH=/home/khw/bin:/home/khw/perl5/perlbrew/bin:/home/khw/print/bin:/bin:/usr/local/sbin:/usr/local/bin:/usr/sbin:/usr/bin:/sbin:/usr/games:/usr/local/games:/home/khw/cxoffice/bin |
From @jhiStrangely, cannot repro in OS X with g++ 4.9.2 (*), with x86_64 and "Configure -des -Dusedevel". Something about Ubuntu in particular? It would be good to change t/op/pack.t:1127 to be is($p1, $p2, "t=$t, c=$t"); There are a few spots in pp_pack.c where the pack() code intentionally calls Zero() for (*) g++-mp-4.9 (mp = MacPorts), -v saying "gcc version 4.9.2 (MacPorts gcc49 4.9.2_1)" |
From [Unknown Contact. See original ticket]Strangely, cannot repro in OS X with g++ 4.9.2 (*), with x86_64 and "Configure -des -Dusedevel". Something about Ubuntu in particular? It would be good to change t/op/pack.t:1127 to be is($p1, $p2, "t=$t, c=$t"); There are a few spots in pp_pack.c where the pack() code intentionally calls Zero() for (*) g++-mp-4.9 (mp = MacPorts), -v saying "gcc version 4.9.2 (MacPorts gcc49 4.9.2_1)" |
From @tonycozOn Tue Mar 03 04:04:28 2015, jhi wrote:
You need at least -Dcc=g++ (it doesn't happen with gcc) and -Duselongdouble (the only floating point type with packing filler). I tested that with gcc 5.0 today, but I reproduced it with 4.9.1 a few weeks ago. Here's the relevant generated codecode: /home/tony/dev/perl/git/perl/pp_pack.c:2732 ; Zero() ; the strange part, we already stored the result above, here we store it again Tony |
|
The RT System itself - Status changed from 'new' to 'open' |
From @jhiI did use -Dcc=g++-mp-4.9, but not -Duselongdouble. (I had this mental glitch of long double packing working with just HAS_LONG_DOUBLE, without USE_LONG_DOUBLE, caused by prolonged exposure to printf() code which does exactly that...) But nope, still no fail from op/pack.t in OS X, with g++ 4.9.2 and -Duselongdouble. ./miniperl -Ilib -V Characteristics of this binary (from libperl): |
From [Unknown Contact. See original ticket]I did use -Dcc=g++-mp-4.9, but not -Duselongdouble. (I had this mental glitch of long double packing working with just HAS_LONG_DOUBLE, without USE_LONG_DOUBLE, caused by prolonged exposure to printf() code which does exactly that...) But nope, still no fail from op/pack.t in OS X, with g++ 4.9.2 and -Duselongdouble. ./miniperl -Ilib -V Characteristics of this binary (from libperl): |
From @jhiI tried poking around to get to see the same disassembly but it seems that with OS X no can do. The g++ -S does some, there's otool(1) which is kind of objdump, but it's not the same. But I *think* by visual fuzzy grep this is the same spot as seen by (a) g++ -S -fverbose-asm (b) otool -tvVj: (a) L2017: (b) 000000000000b5c8 488b842498000000 movq 0x98(%rsp), %rax |
From [Unknown Contact. See original ticket]I tried poking around to get to see the same disassembly but it seems that with OS X no can do. The g++ -S does some, there's otool(1) which is kind of objdump, but it's not the same. But I *think* by visual fuzzy grep this is the same spot as seen by (a) g++ -S -fverbose-asm (b) otool -tvVj: (a) L2017: (b) 000000000000b5c8 488b842498000000 movq 0x98(%rsp), %rax |
From @khwilliamsonIf this doesn't get fixed for v5.22, the branch at |
From [Unknown Contact. See original ticket]If this doesn't get fixed for v5.22, the branch at |
From @tonycozOn Thu Mar 26 20:59:53 2015, khw wrote:
Please try the attached patch. Tony |
From @tonycoz0001-perl-123971-fix-long-double-pack-padding-on-newer-GC.patchFrom a2db6c92fc2880ba1f288b6c4599a779997dd4e7 Mon Sep 17 00:00:00 2001
From: Tony Cook <tony@develop-help.com>
Date: Thu, 9 Apr 2015 15:45:16 +1000
Subject: [PATCH] [perl #123971] fix long double pack padding on newer GCC
---
pp_pack.c | 11 +++++++++++
1 file changed, 11 insertions(+)
diff --git a/pp_pack.c b/pp_pack.c
index 06adade..1d732a8 100644
--- a/pp_pack.c
+++ b/pp_pack.c
@@ -2709,6 +2709,12 @@ S_pack_rec(pTHX_ SV *cat, tempsym_t* symptr, SV **beglist, SV **endlist )
#ifdef __GNUC__
/* to work round a gcc/x86 bug; don't use SvNV */
anv.nv = sv_2nv(fromstr);
+# if defined(LONGDOUBLE_X86_80_BIT) && defined(USE_LONG_DOUBLE) \
+ && LONG_DOUBLESIZE > 10
+ /* GCC sometimes overwrites the padding in the
+ assignment above */
+ Zero(anv.bytes+10, sizeof(anv.bytes) - 10, U8);
+# endif
#else
anv.nv = SvNV(fromstr);
#endif
@@ -2726,6 +2732,11 @@ S_pack_rec(pTHX_ SV *cat, tempsym_t* symptr, SV **beglist, SV **endlist )
# ifdef __GNUC__
/* to work round a gcc/x86 bug; don't use SvNV */
aldouble.ld = (long double)sv_2nv(fromstr);
+# if defined(LONGDOUBLE_X86_80_BIT) && LONG_DOUBLESIZE > 10
+ /* GCC sometimes overwrites the padding in the
+ assignment above */
+ Zero(aldouble.bytes+10, sizeof(aldouble.bytes) - 10, U8);
+# endif
# else
aldouble.ld = (long double)SvNV(fromstr);
# endif
--
1.7.10.4
|
From @khwilliamsonOn 04/08/2015 11:46 PM, Tony Cook via RT wrote:
This patch causes the previously failing tests to now pass. |
|
@tonycoz - Status changed from 'open' to 'resolved' |
Migrated from rt.perl.org#123971 (status was 'resolved')
Searchable as RT123971$