[Debian-med-packaging] Bug#1021394: src:libcereal: FTBFS on arm, ppc64le, s390x

[Scott Talbert]

On Fri, 7 Oct 2022, Andreas Tille wrote:

> Control: tags -1 help
> Control: tags -1 upstream
> Control: tags -1 forwarded https://github.com/USCiLab/cereal/issues/744
>
> Am Fri, Oct 07, 2022 at 09:37:03AM -0400 schrieb Scott Talbert:
>> Justification: fails to build from source (but built successfully in the past)
>
> I guess it never has built in the past but the package was formerly
> Architecture: all and thus it was somehow available on the said
> architectures without really working there which was left unnoticed.

It seems, though, that it *has* been built in the past, e.g., on arm64:
https://buildd.debian.org/status/logs.php?pkg=libcereal&arch=arm64

It looks like this package was not always Architecture: all, see:
https://salsa.debian.org/med-team/libcereal/-/commit/3e3f4a3b4e9bb068a87fd6fd118df01571b88448

>> libcereal 1.3.2+dfsg-3 FTBFS on arm, ppc64le, s390x.  This seems to be
>> the relevant error:
>>
>> [ 22%] Building CXX object unittests/CMakeFiles/test_map.dir/map.cpp.o
>> cd /<<PKGBUILDDIR>>/obj-aarch64-linux-gnu/unittests && /usr/bin/c++  -I/<<PKGBUILDDIR>>/include -g -O2 -ffile-prefix-map=/<<PKGBUILDDIR>>=. -fstack-protector-strong -Wformat -Werror=format-security -Wdate-time -D_FORTIFY_SOURCE=2 -Wall -Wextra -pedantic -Wshadow -Wold-style-cast -Werror -std=gnu++11 -MD -MT unittests/CMakeFiles/test_map.dir/map.cpp.o -MF CMakeFiles/test_map.dir/map.cpp.o.d -o CMakeFiles/test_map.dir/map.cpp.o -c /<<PKGBUILDDIR>>/unittests/map.cpp
>> In file included from /<<PKGBUILDDIR>>/unittests/map.cpp:28:
>> /<<PKGBUILDDIR>>/unittests/map.hpp: In instantiation of ‘void test_map() [with IArchive = cereal::BinaryInputArchive; OArchive = cereal::BinaryOutputArchive]’:
>> /<<PKGBUILDDIR>>/unittests/map.cpp:34:68:   required from here
>> /<<PKGBUILDDIR>>/unittests/map.hpp:65:43: error: narrowing conversion of ‘random_value<char>(gen)’ from ‘std::enable_if<true, char>::type’ {aka ‘char’} to ‘signed char’ [-Werror=narrowing]
>>    65 |       o_esplmap.insert({random_value<char>(gen),  { random_value<int>(gen), random_value<int>(gen) }});
>>       |                         ~~~~~~~~~~~~~~~~~~^~~~~
>
> This was reported by a Fedora user[1] - please see the build log
> in the upstream issue.
>
>> See buildd logs for more information.
>
> I'm tempted to reduce the severity of this bug to important since the
> binary packages of the said architectures never were built and thus the
> criterion for "serious" is not really fulfilled.  What do you think?
>
> Another suggestion would be to exclude the two affected tests for the
> said architectures for the moment - no idea how harmful this might be
> in the end.

I'm not sure that we're seeing the same error as Fedora.  It looks like 
we're seeing a compile error while compiling the tests, rather than a test 
failure.

Regards,
Scott