Re: hdf5 fortran build failure with gcc 12
Dave Love wrote on 2022/02/01 19:20:
Orion Poplawski <orion(a)nwra.com> writes:
> On 1/29/22 19:40, Mamoru TASAKA wrote:
>> Looks like "minusone" variable looking like constant value defined
>> in
>> fortran/test/tH5A_1_8.F90 needs PARAMETER attribute. Tried this diff:
>> ```
>> diff --git a/hdf5.spec b/hdf5.spec
>> index 3b1d27b..2bac44d 100644
>> --- a/hdf5.spec
>> +++ b/hdf5.spec
>> @@ -187,6 +187,8 @@ autoreconf -f -i
>> # Modify low optimization level for gnu compilers
>> sed -e 's|-O -finline-functions|-O3 -finline-functions|g' -i
>> config/gnu-flags
>> +sed -i fortran/test/tH5A_1_8.F90 -e 's|INTEGER :: minusone|INTEGER,
>> PARAMETER :: minusone|'
>> +
>> %build
>> #Do out of tree builds
>> %global _configure ../configure
>> ```
>> https://koji.fedoraproject.org/koji/taskinfo?taskID=82124997
>> Unfortunately ppc64le %check fails, looking like long double
>> related,
>> which I believe is related to recent gcc default change to
>> -mabi=ieeelongdouble on ppc64le,
>> however I don't know how to fix this.
>
> Thank you! I've passed that on to upstream.
I'm no longer a language lawyer, but it looks to me as if it should be
reported as a GCC bug. Is there something in the standard which says
the code is invalid?
Well, the problem here is that the code itself now compiles, but the result after
execution of the binary seems wrong, so we have to narrow the code.
Then, out of curiosity, I tried debugging this, and indeed something seems wrong
with gcc here. Reporeted:
https://bugzilla.redhat.com/show_bug.cgi?id=2050569
Mamoru