[PATCH] libstdc++: Test for unsupported engine range for 128bits floating points [PR119739]

[Jonathan Wakely]

On Wed, 13 May 2026 at 18:22, Rainer Orth <ro@cebitec.uni-bielefeld.de> wrote:
>
> Hi Jonathan,
>
> >> libstdc++-v3/ChangeLog:
> >>
> >>         * testsuite/26_numerics/random/uniform_real_distribution/operators/gencanon_eng_neg.cc:
> >>         New test.
> >> ---
> >> I was planing to add the test before christmas break, but didn't managed to.
> >> I have some ideas, how to handle it, without requiring 128bit integers,
> >> but not sure how important this is. The failures will seem to be very arbitral
> >> form user-perspective, but custom engines are I assume uncommon.
> >>
> >> Tested new test on x86_64-linux locally with all supported standard modes,
> >> and with/without -m32. OK for trunk?
> >
> > OK given Nathan's review.
>
> the test FAILs on 32 and 64-bit sparc-sun-solaris2.11:
>
> +FAIL: 26_numerics/random/uniform_real_distribution/operators/gencanon_eng_neg.cc  -std=gnu++20  (test for errors, line 58)
> +FAIL: 26_numerics/random/uniform_real_distribution/operators/gencanon_eng_neg.cc  -std=gnu++20  (test for errors, line 65)
> +FAIL: 26_numerics/random/uniform_real_distribution/operators/gencanon_eng_neg.cc  -std=gnu++20  (test for errors, line 66)
> +FAIL: 26_numerics/random/uniform_real_distribution/operators/gencanon_eng_neg.cc  -std=gnu++20  (test for errors, line 73)
> +FAIL: 26_numerics/random/uniform_real_distribution/operators/gencanon_eng_neg.cc  -std=gnu++20  (test for errors, line 74)
> +FAIL: 26_numerics/random/uniform_real_distribution/operators/gencanon_eng_neg.cc  -std=gnu++20  (test for errors, line 81)
>
> __SIZEOF_FLOAT128__ is undefined, yet the test expects errors on those
> lines.

Right, you can't put { dg-error } inside #ifdef, because dejagnu
doesn't care about the preprocessor, it looks at all lines.