https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114065
--- Comment #54 from hainque at adacore dot com <hainque at adacore dot com> --- > > We're still validating v18 in-house. Our biggest testsuite still shows > about 15 test failures, which so far we believe should all be ok this > still needs to be confirmed. Some of the tests contain user code that makes > invalid assumptions on internal gnat units, for example. We're adjusting > accordingly and retest. > Ok, so we have confirmed that most of the 15 failures were in legacy tests that relied on invalid assumptions regarding runtime APIs. This is progress. One problem remains: we can't drop support for negative Durations entirely, as they are explicitly allowed by the RM and the GNAT.Calendar interface currently accepts them, even if the code doesn't handle actual negative values adequately. We are working on fixes for this in the current implementation and we'll be happy to share the corrected version together with the corresponding tests we use as soon as we have them ready. We'd appreciate if you could figure out a way to retrofit that code/functionality in system.C_time, to verify how the merged version behaves wrt the issues you were willing to address initially. Olivier
