--- Additional Comments From stefan dot becker at nokia dot com 2009-03-03
08:00 ---
Yes, I know. The problem should go away once we deploy a GCC version that uses
gold as linker and recompile the problematic library.
--
http://sourceware.org/bugzilla/show_bug.cgi?id=7091
--- Yo
--- Additional Comments From ian at airs dot com 2009-03-03 07:52 ---
I'm not sure there is much I can do without a test case, or at least some
in-depth analysis which you would have to do.
--
http://sourceware.org/bugzilla/show_bug.cgi?id=7091
--- You are receiving this mail bec
--- Additional Comments From stefan dot becker at nokia dot com 2009-03-02
13:45 ---
Unfortunately the fix didn't help with the original file. I still get the same
assert message.
--
http://sourceware.org/bugzilla/show_bug.cgi?id=7091
--- You are receiving this mail because: ---
--- Additional Comments From ian at airs dot com 2009-02-28 00:18 ---
I hand-edited a file to recreate the crash, and committed a fix to avoid the
crash.
--
What|Removed |Added
--
--- Additional Comments From cvs-commit at gcc dot gnu dot org 2009-02-28
00:12 ---
Subject: Bug 7091
CVSROOT:/cvs/src
Module name:src
Changes by: i...@sourceware.org 2009-02-28 00:12:26
Modified files:
gold : ChangeLog object.cc output.cc output.h
--- Additional Comments From ian at airs dot com 2008-12-15 18:33 ---
If I understand the error correctly, then the object file appears to be
erroneous, and it's unlikely that gold will ever handle it. However, it is of
course hard to be sure of the issues without having access to the ac
--- Additional Comments From stefan dot becker at nokia dot com 2008-12-13
15:39 ---
So our only option is to make sure that the pre-compiled library is recompiled
and relinked with a newer gnuld linker or gold?
--
http://sourceware.org/bugzilla/show_bug.cgi?id=7091
--- You are
--- Additional Comments From ian at airs dot com 2008-12-13 02:25 ---
That suggests that there is a local symbol defined in an SHT_REL section. That
does not make sense, and should not be possible. However, if it did somehow
happen, then it does seem plausible that gold would crash in t
--- Additional Comments From stefan dot becker at nokia dot com 2008-12-12
18:31 ---
Created an attachment (id=3104)
--> (http://sourceware.org/bugzilla/attachment.cgi?id=3104&action=view)
Obfuscated output from readelf -WS --full-section-name (production library)
Attached is the reque
--- Additional Comments From ian at airs dot com 2008-12-12 16:44 ---
This crash means that there is a section symbol attached to an input merge
section, but the input merge section was apparently not mapped to any output
section. That means that we do not know what value to give to the
--- Additional Comments From stefan dot becker at nokia dot com 2008-12-12
10:07 ---
Created an attachment (id=3103)
--> (http://sourceware.org/bugzilla/attachment.cgi?id=3103&action=view)
Obfuscated output from readelf -Wr (production library)
Thanks for the fix. While the simple test
--- Additional Comments From ian at airs dot com 2008-12-12 02:44 ---
Thanks. It looks like GNU ld is generated R_386_NONE relocations for some
reason, and that is confusing gold. I committed a patch which should cause gold
-r to discard R_xxx_NONE relocations. I don't have a test case
--- Additional Comments From stefan dot becker at nokia dot com 2008-12-11
17:50 ---
Created an attachment (id=3100)
--> (http://sourceware.org/bugzilla/attachment.cgi?id=3100&action=view)
Obfuscated output from readelf -Wr
Here is the requested data. I hope it doesn't matter that I ob
--- Additional Comments From ian at airs dot com 2008-12-11 16:05 ---
It looks like gold is getting confused by an absolute relocation. Can you
attach the output of readelf -r on the precompiled_gnuld.a file? Thanks.
--
http://sourceware.org/bugzilla/show_bug.cgi?id=7091
--- Yo
--- Additional Comments From stefan dot becker at nokia dot com 2008-12-11
09:37 ---
Dooh, no wonder I couldn't make heads or tails out of the backtrace. Should have
thought about that myself :-)
Here is the backtrace generated from a binary compiled with -O0:
(gdb) bt
#0 gold::do_gol
--- Additional Comments From ian at airs dot com 2008-12-10 16:45 ---
The first thing I need to see is a backtrace from a linker which was compiled
without optimization. Your backtrace is truncated relative to the source code
because it doesn't show the inlined functions. I can't figure
16 matches
Mail list logo
