[Libunwind-devel] Re: Ltest-nomalloc failure

[Zach Welch]

On 10/26/2010 09:50 AM, Arun Sharma wrote:
> On Mon, Oct 25, 2010 at 9:08 PM, Zach Welch <address@hidden> wrote:
> 
>> Setting a breakpoint on malloc when running Ltest-nomalloc shows that
>> fopen() indirectly allocates memory. I am not sure how to prevent that,
>> but here's a backtrace for our mutual amusement and consideration.
>>
> 
> x86 defaults to:  --enable-debug-frame=no, so it shouldn't be an issue.
> There are other code paths that result in calls to malloc when this option
> is enabled.

Disabling that option did allow this test to pass. Thanks!

Which platforms require --enable-debug-frame by default? I have a stack
of patches pending that clean up the build system, and I could add one
that sets the correct default for this option. Thus, users would only
need to specify --{enable,disable}-debug-frame when they want to use a
non-standard configuration. Does this sound like a desirable change?

Also, it would seem vaguely logical to skip the nomalloc test when
libunwind is configured with --enable-debug-frame=yes. If rewriting with
open() should be considered the proper solution, this test should at
least be marked XFAIL when that option has been enabled. Which of these
would be considered most acceptable? More generally, is there a better
option for getting this test to work correctly for more use cases?

-- 
Zach Welch
CodeSourcery
address@hidden
(650) 331-3385 x743