Build failure on mips with gcc 5.3 and botan 2.18.2

[ffontaine]

We have the following build failure on one mips toolchain since version 2.18.2 and 219329f:

src/lib/hash/sha3/sha3.cpp: In function 'std::tuple<long long unsigned int, long long unsigned int, long long unsigned int, long long unsigned int, long long unsigned int> Botan::{anonymous}::xor_CNs(const uint64_t*)':
src/lib/hash/sha3/sha3.cpp:41:42: error: converting to 'std::tuple<long long unsigned int, long long unsigned int, long long unsigned int, long long unsigned int, long long unsigned int>' from initializer list would use explicit constructor 'constexpr std::tuple< <template-parameter-1-1> >::tuple(_UElements&& ...) [with _UElements = {long long unsigned int, long long unsigned int, long long unsigned int, long long unsigned int, long long unsigned int}; <template-parameter-2-2> = void; _Elements = {long long unsigned int, long long unsigned int, long long unsigned int, long long unsigned int, long long unsigned int}]'
       A[4] ^ A[9] ^ A[14] ^ A[19] ^ A[24]};
                                          ^

Build is fixed if I revert 219329f. Do you have some advise? Should we add a dependency on gcc 6 or above?

Full build log: http://autobuild.buildroot.org/results/1aa/1aafc2bfdb82c6424327062e17f5745c15417526/build-end.log

[randombit]

Sorry about that. I did not intend to increase the minimum compiler version. I thought we had a GCC 4.4 CI build to ensure this did not happen, but possibly that was lost when we converted from Travis to GitHub Actions for CI.

As long as you are building only with GCC, reverting 219329f is just fine. I'll work on something to fix this for the next release.

[ffontaine]

Thanks for your feedback, I'll revert this commit on buildroot. It should be noted that 'm not 100% sure that this is only linked to the gcc version. This could also be linked to this specific and very old toolchain.

[reneme]

GCC 4.4 seems to choke on the implicit call to the std::tuple<> constructor. By specifically adding it to the return statement, it seems to be happy: https://godbolt.org/z/5YnvvKTfT (see line 28).

[reneme]

See here for a patch.

[ffontaine]

I confirm that #2849 fix the build failure on mips with gcc 5.3