Le 16/10/2015 10:08, Christophe Lyon a écrit :
Hi,
We have noticed a few random failures in gfortran tests in our validations.
Maxim investigated some of them and noticed a possible race condition
in the streamio tests, for which he'll post a patch.
I looked for other similar cases (checking which files are unlinked
several times during 'make check'), and noticed that a few other cases
might read/write/delete the same file concurrently.
The proposed fix is to use different names for the different testcases.
I ran make check -jX on various ARM/AArch64 configuration with no regression.
OK?
Looks good to me. Thanks.
Mikael
