This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: ubsan, asan testing is broken due to coloring


On Wed, Nov 12, 2014 at 1:25 AM, Yury Gribov <y.gribov@samsung.com> wrote:
> [CC-ing sanitizer team.]
>
> On 11/12/2014 08:02 AM, Andrew Pinski wrote:
>>
>> With some configurations (looks like out of tree testing more than in
>> tree testing), all of ubsan and asan tests fail due to the
>> libsanitizer using coloring and that confuses the dejagnu pattern
>> matching.
>
>
> Right, we fix new errors like this every now and then but they keep popping
> up.
>
>> I don't have time to look fully into how to fix this issue
>> and I don't care much to coloring anyways so I disabled in the source
>> for my own use so the tests now pass.
>
>
> First, we could run with ASAN_OPTIONS=color=0.  I think we once tracked this
> error to QEMU incorrectly returning 1 to ASan's isatty() but never bothered
> to fix because fixing tests is so easier.

I was running not under qemu but rather running using a newer glibc
and running using the new ld.so.  And expect (dejagnu) uses psedu ttys
as it wants not to have different behavior as an user would do.  Why
it is different between in tree testing and installed testing, I don't
know.

Thanks,
Andrew Pinski


>
> -Y


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]