[Date Prev][Date Next] [Thread Prev][Thread Next] [Date Index] [Thread Index]

Re: Bug#815495: zeromq3: FTBFS on kFreeBSD: 'xucred cred' has incomplete type


László Böszörményi (GCS) wrote:
> On Mon, Feb 22, 2016 at 4:30 AM, Steven Chamberlain <steven@pyro.eu.org> wrote:
> > It was just a missing #include.  Please find patch attached
> > (sys_ucred_h.patch).
>  Thanks, this looks sane. Would you submit it to upstream or should I do it?

Maybe it is a good time that I finally get a GitHub account, then I can
try to submit it to upstream that way.

> > There were some other issues on kfreebsd at run-time:
> >
> > | FAIL: test_system
> > | =================
> > |
> > | Function not implemented (src/thread.cpp:135)
> > | FAIL test_system (exit status: 134)
> >
> > most tests were failing that way.  GNU/kFreeBSD doesn't implement
> > pthread_setschedparam yet, so I think we should gracefully ignore ENOSYS
> > instead of aborting.
>  Do you think that something similar may happen on Linux machines as well?

Maybe ENOSYS is GNU/kFreeBSD-specific;  there is a ENOTSUP specified by
POSIX.1 that other architectures may use:
"This value is never returned on Linux (but portable and future-proof applications should nevertheless handle this error return value)."

It would be appropriate to check for the latter on Linux in case someday
the thread implementation is changed or a new architecture is created
without support for this.  Perhaps it should print a warning somewhere
if it happens.

> > With that fixed, another error was.
> >
> > | FAIL: test_shutdown_stress
> > | ==========================
> > |
> > | Connection reset by peer (src/stream_engine.cpp:162)
> > | FAIL test_shutdown_stress (exit status: 134)
> >
> > The other patch (kfreebsd-support.patch) fixes all these things, and all
> > tests pass for me now on kfreebsd-amd64.
>  How did you test it? Simple chroot or pbuilder building? The Linux
> self-test failures, aborts happen on all archtectures. However I
> couldn't reproduce those in a dirty chroot to a clean pbuilder build
> process, neither on amd64 nor on i386. But on buildds it happens all
> the time. :(

That is odd, usually test suites work fine on Linux then we see issues
on GNU/kFreeBSD where nobody has tried it before.

I used a messy sid chroot, maybe with some outdated or extra packages in
there.  I'll try again in a clean sid chroot.

I think you need to `cat ./test-suite.log` or export TEST_VERBOSE=1
somewhere in order to see the actual error mesages in the build
logs.  It will surely be a different error happening on Linux.

Steven Chamberlain

Attachment: signature.asc
Description: Digital signature

Reply to: