http://sourceware.org/bugzilla/show_bug.cgi?id=14007
--- Comment #2 from Markus Trippelsdorf
2012-04-22 07:38:26 UTC ---
Created attachment 6359
--> http://sourceware.org/bugzilla/attachment.cgi?id=6359
testcase
markus@x4 /tmp % clang++ -O4 test.ii
/usr/lib64/gcc/x86_64-pc-linux-gnu/4.8.0/../
http://sourceware.org/bugzilla/show_bug.cgi?id=14007
--- Comment #3 from Ian Lance Taylor 2012-04-22 17:04:54
UTC ---
Thanks. Since I don't have clang++, I need to see the invocation of the
linker, not the invocation of clang++. Probably clang++ has a -v option which
will show precisely how th
http://sourceware.org/bugzilla/show_bug.cgi?id=14007
--- Comment #4 from Markus Trippelsdorf
2012-04-22 17:14:05 UTC ---
markus@x4 tmp % gdb ld
Reading symbols from /usr/bin/ld...Reading symbols from
/var/debug/usr/x86_64-pc-linux-gnu/binutils-bin/git/ld.gold.debug...done.
done.
(gdb) run -m elf
http://sourceware.org/bugzilla/show_bug.cgi?id=14007
--- Comment #5 from Markus Trippelsdorf
2012-04-22 17:22:25 UTC ---
BTW the LLVMgold.so plugin from Clang-3.0 doesn't cause the crash.
If you don't have clang installed, I could attach the buggy version of
LLVMgold.so and libLTO.so from Clang
http://sourceware.org/bugzilla/show_bug.cgi?id=14007
--- Comment #6 from Markus Trippelsdorf
2012-04-22 17:32:06 UTC ---
Created attachment 6360
--> http://sourceware.org/bugzilla/attachment.cgi?id=6360
testcase clang libs + object file
--
Configure bugmail: http://sourceware.org/bugzilla/us
http://sourceware.org/bugzilla/show_bug.cgi?id=14009
Bug #: 14009
Summary: gcc 4.7 conversion to init_array breaks users needing
control of C++ static initialization order
Product: binutils
Version: unspecified
Status: N
http://sourceware.org/bugzilla/show_bug.cgi?id=14007
--- Comment #7 from Ian Lance Taylor 2012-04-22 21:20:32
UTC ---
Ideally I need whatever is required to recreate the problem with just gold and
absolutely nothing else. If I have to get into building clang, experience
tells me that I will alm
http://sourceware.org/bugzilla/show_bug.cgi?id=14009
Ian Lance Taylor changed:
What|Removed |Added
CC||ian at airs dot com
--- Comment #1
http://sourceware.org/bugzilla/show_bug.cgi?id=14007
--- Comment #8 from Markus Trippelsdorf
2012-04-22 21:36:35 UTC ---
(In reply to comment #7)
>
> That said, I actually looked at the backtrace, and I see this:
>
>
> #3 0x00571ee1 in gold::Symbol_table::add_from_pluginobj<64, false
http://sourceware.org/bugzilla/show_bug.cgi?id=14009
--- Comment #2 from Ivan Godard 2012-04-22
21:47:00 UTC ---
I'm likely missing something here but as far as I can tell this won't work.
First, the governing order for execution under init_array is the order of
pointers in the init_array itsel
http://sourceware.org/bugzilla/show_bug.cgi?id=14009
Mike Frysinger changed:
What|Removed |Added
CC||vapier at gentoo dot org
--- Comment
http://sourceware.org/bugzilla/show_bug.cgi?id=14009
--- Comment #4 from Ivan Godard 2012-04-22
22:23:11 UTC ---
Constructor priority is not usable in large projects incorporating third-party
binaries and usinging mixed platform tools. Even without 3rd party code,
priorities require a global vie
http://sourceware.org/bugzilla/show_bug.cgi?id=14007
--- Comment #9 from Ian Lance Taylor 2012-04-22 22:28:29
UTC ---
My apologies. I misinterpreted the bug report.
I don't object to having the linker try to validate the data that the plugin
passes to it, but it's not something I plan to work
http://sourceware.org/bugzilla/show_bug.cgi?id=14009
--- Comment #5 from Ian Lance Taylor 2012-04-22 22:30:15
UTC ---
You said that you use a linker script to order the .ctors sections.
How do you do that?
--
Configure bugmail: http://sourceware.org/bugzilla/userprefs.cgi?tab=email
--- Yo
http://sourceware.org/bugzilla/show_bug.cgi?id=14009
--- Comment #6 from Ivan Godard 2012-04-22
22:50:21 UTC ---
We manually maintain, for each project (== library in our environment) a list
of intra-project dependencies (file a must be initialized before file b) and
inter-project dependencies (
http://sourceware.org/bugzilla/show_bug.cgi?id=14009
--- Comment #7 from Ian Lance Taylor 2012-04-23 03:27:43
UTC ---
As far as I can see you can use the exact same strategy for .init_array, only
you sort the sections the other way around.
You may think that the .ctors sections contain the cons
http://sourceware.org/bugzilla/show_bug.cgi?id=14009
--- Comment #8 from Ivan Godard 2012-04-23
03:43:32 UTC ---
Ah - I see. Thank you.
I was misled by old linkers (pre-C++) that put sections physically adjacent so
execution would fall from one to another if you set it up right.
--
Configure
http://sourceware.org/bugzilla/show_bug.cgi?id=14007
Markus Trippelsdorf changed:
What|Removed |Added
Priority|P2 |P3
Summary|gold crash
18 matches
Mail list logo
