Hi David, > On 2 Sep 2021, at 15:47, David Malcolm <dmalc...@redhat.com> wrote: > > On Thu, 2021-08-19 at 19:59 +0100, Iain Sandoe wrote: >>
>> tested on i686, x86_64-darwin, x86_64,powerpc64-linux, > > Which versions of DejaGnu, BTW? framework 1.5, 1.5.1, 1.6.2 expect 5.45 / 5.45.4 / 5.45r2(darwin) (various local patches from distros) tcl 8.5 and 8.6[.9,11] I do see problems with stability of the summary counts with 1.5.1*** on Power machines (for my usual work I have a local 1.6.2 build) >> OK for master? > > Did you try this with RUN_UNDER_VALGRIND set? Assuming that that still > works, yes, looks good to me. This turned into a small rabbit hole, (not sure if that relates to the depth or the size of the rabbit). — there are a lot of ERROR outputs that make their way into the summary file — it seems that these are XFAILed, but the noise makes it hard to check equivalence .. anyway I tried a whole bunch of build permutations to see if forcing DWARF-4 fixed the issues.. it did not. The end result is that for both x86_64 and powerpc64le*** I see 2 progressions in the valgrind results … and AFAICT no regressions - if I do a side-by-side compare of the .sum files, the results seem reasonable. Of couse, it’s very hard to be 100% sure since the error output includes process numbers which makes diffs less helpful. anyway, I’ve applied this, and think I can declare x86_64-darwin as supporting Jit (actually, so does i686 and powerpc, but there are problems with the threaded test - 99% sure that’s testsuite-related and the two toy examples use a system routine that’s not available in darwin9). thanks Iain *** I’d recommend using framework 1.6.2 for stability of summary counts, the actual .sum seems ok with 1.5.1 - but not the listed results - maybe some manifestation of https://gcc.gnu.org/bugzilla/show_bug.cgi?id=64983 (although that was not supposed to apply to 1.5.1 at least on Darwin).
