Re: CI changes and test failures on ppc64le-linux-gnu

[Anthony Green <green@moxielogic.com>]

I found ppc64le hardware in the GCC compile farm and everything passed
(with new and old host compilers).  So I suppose this is a problem with
qemu.  I'll remove that platform from testing on travis-ci.

AG



On Wed, Oct 16, 2019 at 4:22 PM Anthony Green <green@moxielogic.com> wrote:

> I've been working on adding additional architecture support to our
> travis-ci tests and, to that end, I just committed a patch to test on 32-
> and 64-bit ARM and ppc64le Linux systems as well as using foreign fedora
> container images emulated with qemu.
>
>
> https://github.com/libffi/libffi/commit/825b2a359468af8bf3570a5a0695e8b805c5446f
>
> The builder images are defined in a new project here:
>
> https://github.com/libffi/ci-build-containers
>
> One of the tricks to get this working was to split the testing of
> different optimization levels into different build jobs (so as to not to
> exceed the build timeout limits on travis).  So now you can set
> LIBFFI_TEST_OPTIMIZATIONS="-O2" and 'make check' will only test that
> optimization level.  Travis runs multiple build jobs in parallel, so this
> seems like a good strategy.
>
> Test results look good, with the exception of some ppc64le failures on the
> struct-related tests.  Unfortunately, I'm unable to run gdb in this kind of
> emulated environment, so I'd appreciate it if somebody could have a  look.
>
> I also just noticed that Nov 12 will mark the 5th anniversary of the last
> libffi release.  It would be nice to release a new one on that day this
> year.   There are still a number of patches in the queue that I'll try to
> dig into again, but I would appreciate help!
>
> Thanks!
>
> AG
>
>