Stefano Lattarini wrote:
> Hi Jim.
>
> On 08/01/2012 04:52 PM, Jim Meyering wrote:
>> I ran coreutils "make check" tests 60 times (on Fedora 17, x86_64),
>> recording the results of each run like this:
>>
>> for i in $(seq 100); do make -j25 check
>>
> So, parallel make ...
>
>> -C tests VERBOSE=yes \
>> RUN_EXPENSIVE_TESTS=no >& makerr-$i
>>
> ... and output redirection. Bad combination; take a look at:
>
> <http://debbugs.gnu.org/cgi/bugreport.cgi?bug=11413#8>
>
> So, if this is to be considered a bug, it is a make bug, not an
> Automake one IMHO. To verify that this is the case you might want
> to re-try your 100 runs above with this slight change:
>
> for i in $(seq 100); do
> + : > makerr-$i
> make -j25 check -C tests VERBOSE=yes \
> RUN_EXPENSIVE_TESTS=no >& makerr-$i \
> && t=. || t=X
> printf $t
> done
>
> and verify that in this case all the 'makerr-*' files are indeed equal.
Hi Stefano,
Thanks for the pointer. I knew that rang a bell.
For the record, to avoid the problem one must append to the existing log file:
for i in $(seq 100); do
+ : > makerr-$i
make -j25 check -C tests VERBOSE=yes \
- RUN_EXPENSIVE_TESTS=no >& makerr-$i \
+ RUN_EXPENSIVE_TESTS=no >>& makerr-$i \
&& t=. || t=X
printf $t
done
