http://sourceware.org/bugzilla/show_bug.cgi?id=12261
--- Comment #3 from Ian Lance Taylor ian at airs dot com 2010-11-24 03:28:08
UTC ---
Thanks for the patch. I've been troubled by this for a while, because defining
a symbol in the base version doesn't reliably cause the glibc dynamic linker
http://sourceware.org/bugzilla/show_bug.cgi?id=11423
Ian Lance Taylor ian at airs dot com changed:
What|Removed |Added
Status|ASSIGNED|WAITING
http://sourceware.org/bugzilla/show_bug.cgi?id=12001
Ian Lance Taylor ian at airs dot com changed:
What|Removed |Added
CC||ian at airs dot
--
What|Removed |Added
CC||ian at airs dot com
http://sourceware.org/bugzilla/show_bug.cgi?id=11997
--- You are receiving this mail
--- Additional Comments From ian at airs dot com 2010-09-18 04:47 ---
The memory test failure is fixed.
However, there does still seem to be a relevant difference between gold and GNU
ld. I used -z max-page-size=0x1000 to eliminate differences due to the page
size. With that, gold
--- Additional Comments From ian at airs dot com 2010-09-13 04:18 ---
This is presumably complaining about STB_GNU_UNIQUE. That is supported in
current versions of gold, though it was not supported in the 2.20 release.
--
What|Removed |Added
--
What|Removed |Added
CC||nickc at redhat dot com
http://sourceware.org/bugzilla/show_bug.cgi?id=11997
--- You are receiving this
--- Additional Comments From ian at airs dot com 2010-09-08 23:55 ---
Already fixed.
--
What|Removed |Added
Status|NEW |RESOLVED
--- Additional Comments From ian at airs dot com 2010-09-07 13:19 ---
BSD ar does have a symbol table. All versions of ar since about 1977 have a
symbol table.
It would be possible to have gold scan an entire archive if there is no symbol
table. The problem here
--
What|Removed |Added
AssignedTo|ian at airs dot com |unassigned at sources dot
||redhat dot com
http
--- Additional Comments From ian at airs dot com 2010-09-07 04:46 ---
An archive is expected to have a symbol map, which the linker uses to determine
which objects to include from the archive. If your objects don't have regular
symbols, then the ar program is not going to be build
--- Additional Comments From ian at airs dot com 2010-08-23 04:50 ---
I would use parentheses and line up the ?: under the first part of the
expression.
Patch is OK.
Thanks.
--
http://sourceware.org/bugzilla/show_bug.cgi?id=11899
--- You are receiving this mail because
/Makefile.am.diff?cvsroot=srcr1=1.140r2=1.141
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gold/testsuite/Makefile.in.diff?cvsroot=srcr1=1.148r2=1.149
--- Additional Comments From ian at airs dot com 2010-08-19 22:54 ---
Fixed for i386 and x86_64. If you find any problems, please open a new bug
--- Additional Comments From ian at airs dot com 2010-08-19 23:00 ---
The patch is not correct in general, because the paddr_ field of segments is
not
set in general at the point where it is tested. It will be set when a linker
script is used, but not by default. The paddr_ field
--- Additional Comments From ian at airs dot com 2010-08-04 08:58 ---
This is definitely an STT_GNU_IFUNC issue.
*** This bug has been marked as a duplicate of 10893 ***
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-08-04 08:58 ---
*** Bug 11694 has been marked as a duplicate of this bug. ***
--
What|Removed |Added
CC
--
What|Removed |Added
AssignedTo|ian at airs dot com |dougkwan at google dot com
Status|NEW |ASSIGNED
http://sourceware.org
--- Additional Comments From ian at airs dot com 2010-08-03 14:26 ---
The issue with the -plugin option is simply that you need to use
--enable-plugins
when you configure gold.
I don't know what the duplicate definition of version 0 problem is. I would
need
to be able to recreate
/gold/layout.cc.diff?cvsroot=srcr1=1.173r2=1.174
--- Additional Comments From ian at airs dot com 2010-08-03 15:07 ---
Thanks for the bug report. This is now fixed.
With gold, a better and more reliable way to do what I think you are trying to
do
is to use the -Ttext option. That sets
--- Additional Comments From ian at airs dot com 2010-08-03 21:03 ---
There have been many many bug fixes in gold since the binutils 2.20.1 release.
Would it be possible for you to try it with the current development version of
gold?
--
What|Removed
/elf.c.diff?cvsroot=srcr1=1.513r2=1.514
--- Additional Comments From ian at airs dot com 2010-08-02 10:35 ---
What is good.so and how was it created? I don't see how gold could create
segment headers that look like that.
I do see a problem when I run strip on a shared library generated
ian at airs dot com 2010-08-02 12:02 ---
I committed a patch which fixes the problem that I saw. Let me know if you
still
see a problem.
--
What|Removed |Added
/symtab.h.diff?cvsroot=srcr1=1.108r2=1.109
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gold/yyscript.y.diff?cvsroot=srcr1=1.24r2=1.25
--- Additional Comments From ian at airs dot com 2010-08-02 13:37 ---
Fixed.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-06-23 04:00 ---
Thanks for the bug report. This has alread ybeen fixed in the development
sources. The patch is:
2010-03-05 Ian Lance Taylor i...@google.com
* incremental.cc: Include libiberty.h.
http://sourceware.org
--- Additional Comments From ian at airs dot com 2010-06-11 13:10 ---
gold does not yet support STT_GNU_IFUNC symbols at all.
--
What|Removed |Added
Status|NEW
--
What|Removed |Added
CC||ccoutant at google dot com
http://sourceware.org/bugzilla/show_bug.cgi?id=11683
--- You are receiving
--- Additional Comments From ian at airs dot com 2010-04-08 00:38 ---
This may have been fixed by a checkin earlier today. Can you give it another
try? Thanks.
--
http://sourceware.org/bugzilla/show_bug.cgi?id=11466
--- You are receiving this mail because: ---
You
--- Additional Comments From ian at airs dot com 2010-04-02 17:40 ---
Which version of gold? Which version of the Linux kernel? Which target?
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-04-01 01:44 ---
As I said in comment #17, please open a new bug report about any specific
issues.
Please do not reopen this bug report. Thanks.
All options which take values do accept =VALUE. I expect that you are using
old
--- Additional Comments From ian at airs dot com 2010-03-29 21:02 ---
I expect that I or somebody will fix this fairly soon but I don't know exactly
when.
Obviously if anybody wants to send a patch, that would be great.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-03-24 13:11 ---
Created an attachment (id=4676)
-- (http://sourceware.org/bugzilla/attachment.cgi?id=4676action=view)
Patch
I think a more reliable patch would be something like this, which changes \1 to
\\1.
--
http
--- Additional Comments From ian at airs dot com 2010-02-04 15:07 ---
Thanks for the bug report. Can you list the errors you get when building gold?
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-02-04 03:45 ---
Bernhard, are you still seeing the problem?
--
http://sourceware.org/bugzilla/show_bug.cgi?id=10708
--- You are receiving this mail because: ---
You are on the CC list for the bug, or are watching someone
--- Additional Comments From ian at airs dot com 2010-01-12 07:14 ---
As far as I can tell, everything works fine with the current development
version
of gold. If you find more problems, please reopen with more details, such as
the
parts of the {gcc,g++}.log file for one
--- Additional Comments From ian at airs dot com 2010-01-12 07:20 ---
With the current development version of gold, I don't see any differences like
this. I see symbols like omp_destroy_lock@@OMP_3.0, which is slightly
different
from the reported omp_destroy_l...@omp_3.0. I'm
--- Additional Comments From ian at airs dot com 2010-01-08 19:43 ---
Fixed.
--
What|Removed |Added
Status|ASSIGNED|RESOLVED
--- Additional Comments From ian at airs dot com 2010-01-08 19:45 ---
Thanks for pointing that out. Fixed.
--
What|Removed |Added
Status|NEW
--- Additional Comments From ian at airs dot com 2010-01-08 21:09 ---
I have committed a patch which I think will fix this problem. Let me know if
it
doesn't work.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-01-08 21:10 ---
Thanks for the bug report. This has been fixed, and gold now supports -G.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-01-08 21:26 ---
Thanks for filing the bug report. This is now fixed in the development version
of gold.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-01-08 23:21 ---
The problem here is that the GNU linker does not warn when it sees a reference
from a debugging section to a symbol defined in a section which is discarded by
the linker script. I don't know whether this should
--- Additional Comments From ian at airs dot com 2010-01-09 00:27 ---
I've committed a patch which should fix this problem. At least, the kernel
does
build now with gold, though I did not test whether it ran correctly after
building.
--
What|Removed
--- Additional Comments From ian at airs dot com 2010-01-07 18:33 ---
I believe that all the command line options mentioned here are now implemented,
or are erroneous.
Thanks for the comprehensive report.
Please open a new bug report for any new missing options, or for any problems
--- Additional Comments From ian at airs dot com 2010-01-07 18:43 ---
Thanks for the bug report. I committed a variant of your patch which also
instantiates Xindex::read_symtab_xindex.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-01-07 18:56 ---
These two programs just look different, and it's hard for me to believe that
this
is due to a difference between GNU ld and gold. For example, the gold binary
has
calls to fgets where the GNU ld binary does
--- Additional Comments From ian at airs dot com 2010-01-07 19:34 ---
Thanks for the bug report and the test case. I committed a patch which fixes
the
problem.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2010-01-07 21:12 ---
Thanks for the bug report. Fixed.
--
What|Removed |Added
Status|NEW
--- Additional Comments From ian at airs dot com 2010-01-07 21:17 ---
What do you mean by always combines dynamic relocations? It's true that gold
always generates only a single .rel.dyn or .rela.dyn output section. What the
-z
combreloc option controls is whether the relocations
--- Additional Comments From ian at airs dot com 2010-01-08 02:54 ---
gold will never change to use separate sections for dynamic relocations. That
makes no sense in gold's architecture. Different sections can not possibly be
required for IFUNC or GNU2 TLS, because the dynamic linker
--- Additional Comments From ian at airs dot com 2010-01-06 14:06 ---
At first glance I'm not sure I agree. __start_section is special already
because it only works when the section has a special name. Can you sketch out a
case where a program uses __start_section but still wants
--- Additional Comments From ian at airs dot com 2010-01-06 05:40 ---
c++filt is a think wrapper around libiberty/cplus-dem.c and friends. The master
source for libiberty is the gcc repository. So, this bug should be filed for
gcc at http://gcc.gnu.org/bugzilla/. Sorry
--- Additional Comments From ian at airs dot com 2010-01-04 23:16 ---
I attached the patch to this bug.
I opened issue 359 for google-breakpad.
--
http://sourceware.org/bugzilla/show_bug.cgi?id=10338
--- You are receiving this mail because: ---
You are on the CC list
--- Additional Comments From ian at airs dot com 2010-01-05 05:52 ---
2 ld: def: unknown -z option
* bobcat
* gnugk
There is no -z def option for the GNU linker. However, as noted earlier, the
GNU
linker does not complain about unknown -z options. These two packages
presumably
--- Additional Comments From ian at airs dot com 2009-12-30 08:33 ---
I think I've fixed all the problems except one where elflint seems wrong to me.
The remaining problem that I see is
loadable segment [6] is writable but contains no writable sections
This is about a PT_TLS segment
--- Additional Comments From ian at airs dot com 2009-12-30 19:31 ---
Thanks for pointing that out. I don't think I would have written it that way,
but clearly gold should follow the standard. I committed a patch to fix it.
elflint is now clean in my test.
--
What
--- Additional Comments From ian at airs dot com 2009-12-30 20:13 ---
I took a look at this, but frankly I've exhausted my patience with trying to
build rpm5. There are too many dependencies that I don't have, and even once I
got it to configure it doesn't build, complaining about
--- Additional Comments From ian at airs dot com 2009-12-30 20:25 ---
I finally got back to this bug report. I added support for -x/--discard-all.
Thanks for reporting it.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2009-12-30 22:43 ---
Thanks for the bug report. The root problem here is that fuse_chan_new appears
twice in the version script. I can't think of any reasonable meaning for that,
so I changed it to be an error. gold now gives
--- Additional Comments From ian at airs dot com 2009-12-30 22:44 ---
H.J., can you add a note here about what else gold needs to support IFUNC?
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2009-12-31 00:28 ---
Thanks for the bug report. I can't help but notice that though you mention --
exclude-libs in the description, you aren't actually using it in the link
command
line. Still I will look to see if I can spot
--- Additional Comments From ian at airs dot com 2009-12-31 01:17 ---
I think I've spotted the problem. I committed a patch for it. Please let me
know if you still have problems with this.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2009-12-31 01:59 ---
Thanks for the bug report. I've added the --sort-common option to gold.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2009-12-31 05:15 ---
I was able to find what I assume is the same problem in a simpler test case.
With luck, this is now fixed.
*** This bug has been marked as a duplicate of 10979 ***
--
What|Removed
--- Additional Comments From ian at airs dot com 2009-12-31 05:15 ---
*** Bug 10669 has been marked as a duplicate of this bug. ***
--
What|Removed |Added
CC
--- Additional Comments From ian at airs dot com 2009-12-31 05:14 ---
The problem with idjc is a complicated case which arises because it defines a
common symbol named shutdown, then links against a shared library which
refers
to shutdown with a glibc version, and then sees from libc
--- Additional Comments From ian at airs dot com 2009-12-31 05:46 ---
The klibc problem is due to -Ttext. gold was trying to put the file header and
segment header in the text segment, but it was in the wrong place due to
-Ttext.
I've committed a patch.
I believe that fixes all
--- Additional Comments From ian at airs dot com 2009-12-31 06:07 ---
I think that this is the current set of missing options:
--add-needed
--allow-multiple-definition
--cref
--no-undefined-version
--section-start
--warn-multiple-gp
--warn-shared-textrel
--warn-unresolved-symbols
-G
--- Additional Comments From ian at airs dot com 2009-12-30 01:34 ---
I finally took another look at this. The problem here is simply that dump_syms
is incorrect. It only works if the stabs strings have been optimized into a
single compilation unit. The GNU linker does
--- Additional Comments From ian at airs dot com 2009-12-30 01:36 ---
Created an attachment (id=4487)
-- (http://sourceware.org/bugzilla/attachment.cgi?id=4487action=view)
google-breakpad patch
This patch seems to fix google breakpad, at least for this test case.
--
http
--- Additional Comments From ian at airs dot com 2009-12-22 14:11 ---
Oh wait, sorry, this is the kernel itself. It must be something else.
--
http://sourceware.org/bugzilla/show_bug.cgi?id=11108
--- You are receiving this mail because: ---
You are on the CC list
--- Additional Comments From ian at airs dot com 2009-12-03 05:35 ---
Thanks for the bug report. This has already been fixed on the binutils 2.20
branch and on mainline. The fix will be in any future release.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2009-11-11 06:48 ---
That part is never going to work with gold. It is asking the linker for the
default linker script, editing it, and passing it back to the linker. gold does
not have a default linker script to edit. Building glibc
--- Additional Comments From ian at airs dot com 2009-11-05 20:01 ---
I already committed a patch to remove the erroneous #if 0.
Bug is fixed.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2009-11-04 16:41 ---
Thanks for the bug report. I've patched the linker to avoid the crash.
However, the core problem is that the linker script expects that the dynamic
reloc sections will have specific names, and gold does not use
--- Additional Comments From ian at airs dot com 2009-11-03 15:27 ---
Fixed.
--
What|Removed |Added
Status|NEW |RESOLVED
--- Additional Comments From ian at airs dot com 2009-11-04 01:30 ---
Thanks for the bug report. Fixed.
--
What|Removed |Added
Status|NEW
--- Additional Comments From ian at airs dot com 2009-10-29 04:20 ---
It would be very helpful if you could pull together a self-contained test case.
Thanks.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2009-10-20 04:10 ---
Thanks for the bug report. This was an error on my part when I copied the gold
sources to the binutils 2.20 release branch. This was never a problem in the
main development sources, and has now been fixed
--- Additional Comments From ian at airs dot com 2009-10-16 06:00 ---
I have added support for these options.
--
What|Removed |Added
Status|NEW
--- Additional Comments From ian at airs dot com 2009-10-12 22:35 ---
Comment #7 does not necessarily indicate that there are a lot of packages which
provide a union-of-defined-symbols interface. What is indicates is that a lot
of people think that linking against the KDE or GNOME
--- Additional Comments From ian at airs dot com 2009-09-30 13:20 ---
Can you confirm that this was with the development version of gold? There were
some bugs in this area fixed back February.
Otherwise, as far as I can see, this can only happen if open returns -1 with
errno set
--- Additional Comments From ian at airs dot com 2009-09-09 03:16 ---
If SHF_LINK_ORDER is set in the flags field, then the sh_link field is permitted
to contain the special values SHN_BEFORE (65280) and SHN_AFTER (65281).
SHN_BEFORE means that this section should come before all other
--- Additional Comments From ian at airs dot com 2009-08-12 09:36 ---
Thanks for the bug report. This has already been fixed in the development
sources.
http://sourceware.org/cgi-bin/cvsweb.cgi/src/gold/reduced_debug_output.cc.diff?r1=1.1r2=1.2cvsroot=srcf=h
--
What
--- Additional Comments From ian at airs dot com 2009-08-12 09:41 ---
The -z options were introduced to GNU ld in emulation of the Solaris linker.
The Solaris linker is documented to accept -z defs, not -z -defs. The GNU
linker man page also documents -z defs, not -z -defs. When using
--- Additional Comments From ian at airs dot com 2009-08-12 10:22 ---
gold, unlike GNU ld, does not attempt to locate libraries which are referenced
by DT_NEEDED entries (see PR 10238 for some more background). Therefore, when
gold sees a shared library which has DT_NEEDED entries
--- Additional Comments From ian at airs dot com 2009-07-21 05:26 ---
Thanks for noticing this. I committed a patch to fix the problem.
--
What|Removed |Added
--- Additional Comments From ian at airs dot com 2009-07-17 01:28 ---
Patch committed.
--
What|Removed |Added
Status|ASSIGNED|RESOLVED
--- Additional Comments From ian at airs dot com 2009-07-16 00:08 ---
I don't see any problem in the .eh_frame section. It appears to me that
.eh_frame information for discarded sections is indeed being discarded.
I do see duplication in the .debug_frame section. Note that readelf -wf
--- Additional Comments From ian at airs dot com 2009-07-16 05:32 ---
Created an attachment (id=4055)
-- (http://sourceware.org/bugzilla/attachment.cgi?id=4055action=view)
Patch
The difference between gold and GNU ld seems to be that when GNU ld sees a
comdat section with a different
--
What|Removed |Added
AssignedTo|ian at airs dot com |ccoutant at google dot com
http://sourceware.org/bugzilla/show_bug.cgi?id=10287
--- You are receiving
--
What|Removed |Added
CC||ian at airs dot com
http://sourceware.org/bugzilla/show_bug.cgi?id=10287
--- You are receiving this mail
--- Additional Comments From ian at airs dot com 2009-06-24 13:17 ---
Offhand, I don't think this is a bug. I don't see why -r should necessarily
change the search path used for archives. You can easily change the search path
by using the -Bstatic option.
--
http://sourceware.org
--- Additional Comments From ian at airs dot com 2009-06-24 14:39 ---
Your object file is unusual in having both an empty .text section and an empty
.rel.text section. gold was failing an assertion while processing the .rel.text
section, in that it expected to have some data to relocate
--- Additional Comments From ian at airs dot com 2009-06-24 19:52 ---
Thanks for the bug report. x86_64 Linux 2.6.27 now builds with gold. I haven't
tried actually running it, though. All the problems were related to linker
script control of .note sections.
--
What
--- Additional Comments From ian at airs dot com 2009-06-24 20:42 ---
The core issue is that valgrind runs the linker with --verbose to extract the
linker script. It then seds the linker script to adjust the start address of
the text segment. This works OK with the GNU linker
--- Additional Comments From ian at airs dot com 2009-06-24 21:00 ---
Thanks for attaching the library.
Unfortunately, I can't recreate the problem. I get lots of bad e_ehsize (36 !=
52) errors. However, I don't get the internal error.
I'm using the current development sources. It's
--- Additional Comments From ian at airs dot com 2009-06-24 21:05 ---
I downloaded alsa-lib 1.0.20 and built it with the current development version
of gold without any trouble.
So, I guess I do need you to give me a test case. It is also conceivable that
the problem was fixed
--
What|Removed |Added
Component|binutils|gold
http://sourceware.org/bugzilla/show_bug.cgi?id=10287
--- You are receiving this mail because:
--- Additional Comments From ian at airs dot com 2009-06-23 06:12 ---
The last empty segment is a result of the empty .data section. The GNU linker
discards empty sections; gold does not. I think gold's behaviour is reasonable,
and that does not seem to be your problem in any case
--- Additional Comments From ian at airs dot com 2009-06-23 06:16 ---
This appears to be fixed in the current development sources. Please let me know
if it is still broken for you.
--
What|Removed |Added
201 - 300 of 375 matches
Mail list logo