Bug#880669: irony-mode: tests do not reliably pass for several archs

[Nicholas D Steeves]

On Fri, Nov 10, 2017 at 04:53:50PM -0500, Aaron M. Ucko wrote:
> Nicholas D Steeves  writes:
> 
> > I've uploaded irony-mode_1.2.0-2~exp1_amd64.  I'm giving LLVM+Clang a
> > try with the new package, as I've read that MacOS users must use Clang
> > rather than GCC...and I suspect that switching to LLVM+Clang might
> > close this bug (hypothesis: error on mips64 and arm64 only exists when
> > building with GCC).  Also, I suspect that upstream intends
> > irony-server to be compiled with clang ;-)
> 
> Interesting idea.
> 
> This change does seem to have helped for arm64, but the jury's still out
> on mips64el.  However, I see new failures on amd64 and s390x, and builds
> for many architectures (albeit all non-release ones) can no longer even
> start because clang is unavailable there.  As such, I'd recommend
> building with (and build-depending on) clang only on those architectures
> on which GCC builds hit errors.

Unfortunately the transition to libclang-5.0-dev (libclang-dev is a
necessary dependency for irony-mode) requested by the Debian
LLVM/Clang will leave non-release archs in limbo even with GCC.  I
believe the two outcomes are 1) libclang-5.0-dev gains support for
these arches 2) the porter downgrades the libclang-5.0-dev dependency
to the latest version supported by the arch.

I like your recommendation.  If mip64el still fails with the upstream
patch then I'll special-case it to build with Clang.

> On the Mac front, Clang has indeed been the only compiler Apple's
> official Development Tools package supports for years (briefly preceded
> by LLVM GCC), but GCC does still support the platform and is available
> through third-party package managers such as Fink.

Ahhh.  I wonder if a Fink-packaged-GCC-compiled on a modern MacOS
would use symbols from both libgcc (eg: libgcc_s.so.1) and llvm+clang
(I can't check at the moment)?  Depends on the libclang-5.0.so.1
dynamic library is certain, in any case, and I wonder if a
Fink-packaged libclang-5.0.so.1 that links with libgcc_s.so.1
instead of (the llvm+clang runtime lib) would be required?

> At any rate, thanks for looking into the errors!

You're welcome :-)  I've prepared an upload for unstable and will
upload as soon as mips64el and mipsel builds & self-test report
success for ~exp2 (compiled with GCC instead of Clang).  I hope these
archs will not need to be special-cased!

https://buildd.debian.org/status/package.php?p=irony-mode=experimental

Cheers,
Nicholas


signature.asc
Description: PGP signature

Bug#880669: irony-mode: tests do not reliably pass for several archs

[Nicholas D Steeves]

Please keep an eye on:
https://buildd.debian.org/status/package.php?p=irony-mode=experimental

for up-to-date info on if this is fixed in experimental.  2~exp2 is
the release I hope will do the trick

Cheers,
Nicholas


signature.asc
Description: PGP signature

Bug#880669: irony-mode: tests do not reliably pass for several archs

[Nicholas D Steeves]

Update:

My LLVM-3.8+Clang-5 experiment revealed this sometimes also affects
amd64:
https://buildd.debian.org/status/fetch.php?pkg=irony-mode=amd64=1.2.0-2~exp1=1510272887=0

Never on my amd64 development box, and only on the buildd.

Also, good news!  Upstream has committed a fix that I believe will
probably do the trick.

Cheers,
Nicholas


signature.asc
Description: PGP signature