[PATCH 2/2] unwind: relax the pass condition of strace-k.test test case

[Dmitry V. Levin]

On Mon, Jun 16, 2014 at 12:07:32PM +0900, Masatake YAMATO wrote:
> How about x86_64?
> 
> I guess testing on x86_64 is failed, too.

No, on x86_64 the test always succeeds.

On armv7hl it is unreliable: sometimes it succeeds
(e.g. https://koji.fedoraproject.org/koji/taskinfo?taskID=7043723),
sometimes it fails
(e.g. https://koji.fedoraproject.org/koji/taskinfo?taskID=7049221),
and on my local armv7hl host it always fails with "too many stack frames".

> It seems that the libunwind on fedora is not linked to liblzma. 
> As the result it cannot decode minidebug info.
> 
>    https://bugzilla.redhat.com/show_bug.cgi?id=972737

This is not necessarily the root of all problems.

> As far as trying with privately built libunwind f1, f0 and main are
> decoded expectedly.
> 
> BTW, I'm afraid stack-fcall used in the test is more optimized than I
> expected.

No, stack-fcall is not overoptimized.  I've split stack-fcall.c into several
compilation units just to be sure that intermediate function calls are not
optimized out (see https://sourceforge.net/p/strace/code/ci/ldv/unwind/~/log/)
and made another test build:
https://koji.fedoraproject.org/koji/taskinfo?taskID=7049219
You can see that on x86_64 the test succeeded, and on i686 and armv7hl it
failed.


-- 
ldv
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 181 bytes
Desc: not available
URL: <http://lists.strace.io/pipermail/strace-devel/attachments/20140617/380660d8/attachment.bin>