Buliding on arm broke in 7.4.1

[nomeata]

As you can see here:  https://buildd.debian.org/status/logs.php?pkg=ghc&arch=armel the GHC build fails on armel on Debian, and the change was introduced between the last release candidate and the final release. This is not a good time to make such changes, as it makes testing the release candidates somewhat obsolete. *slightgrumble*

The problematic change was most likely [50c07f743330edc193ee406beb85c4bb6b5e20ee] but I can’t offer any productive advice, sorry.

[nomeata]

That should have been [5a984f4388ef85d5c3af973b21a12c12b36c1ed4], sorry.

[simonmar]

Sorry about that. Can you test a fix and send it to us, and we'll get it into 7.4.2?

[nomeata]

Yes, I can test it, but not come up with one...

[igloo]

Replying to nomeata:

As you can see here:  https://buildd.debian.org/status/logs.php?pkg=ghc&arch=armel the GHC build fails on armel on Debian, and the change was introduced between the last release candidate and the final release. This is not a good time to make such changes, as it makes testing the release candidates somewhat obsolete. *slightgrumble*

Sorry. Other Arm users had reported that the compiler was broken without these patches. As the patches could only affect the Arm platform (as the changes are all inside CPP conditionals), and as the platform was apparently broken without them, it seemed safe to put them in.

We could have put out another RC, but that would have meant more man hours, at least another week's delay, and there are diminishing returns on RCs - the more we put out, the fewer people are able to find time to test them.

Incidentally, 4647278919ce724717db93af605741b4a419b3b2 is also involved.

[nomeata]

Don’t worry, it is just a slight grumble, as long as a fix is found.

Would it be helpful if we (Debian) set up a GHC builder on an armel machine? Is the builder software capable of handling builds that take longer than 24h, and can it be configured to run less often, e.g. once per week?

[igloo]

See also #5824.

[nomeata]

First responses from Debian admins about providing a build service to upstream are negative ( http://lists.debian.org/debian-haskell/2012/02/msg00029.html), but point to  http://gcc.gnu.org/wiki/CompileFarm for a service that offers access to machines of various architectures to Free Software projects. Maybe the GHC team wants to use these for more build bots?

[igloo]

Unfortunately, we don't have the resources to support all platforms ourselves. While we will help where we can, we really rely on the community to support other platforms, both in terms of running buildbots, and in fixing any platform-specific issues that they reveal.

[nomeata]

Replying to igloo:

Unfortunately, we don't have the resources to support all platforms ourselves. While we will help where we can, we really rely on the community to support other platforms, both in terms of running buildbots, and in fixing any platform-specific issues that they reveal.

Fair enough. Not sure how much I can help with fixing, but I’ll contact the GCC guys and see if I can manage a few buildbots there.

[simonmar]

Now that #5824 is fixed, I believe we can close this ticket, right?

Note #5824 is still in the merge state, the fixes will be merged before the 7.4.2 release.

[nomeata]

Replying to simonmar:

Now that #5824 is fixed, I believe we can close this ticket, right?

I’m risking to answer a rhetorical question here, but: yes :-)

Note #5824 is still in the merge state, the fixes will be merged before the 7.4.2 release.

Thanks, that is great!

[guest]

Well, we just merged the two patches in #5824 for Ubuntu/7.4.1 and uploaded it. The result built on armel, but not armhf, with errors like this

"inplace/bin/ghc-stage1" -prof -H32m -O -lffi -optl-pthread -optc-mlong-calls -package-name ghc-prim-0.2.0.0 -hide-all-packages -i -ilibraries/ghc-prim/. -ilibraries/ghc-prim/dist-install/build -ilibraries/ghc-prim/dist-install/build/autogen -Ilibraries/ghc-prim/dist-install/build -Ilibraries/ghc-prim/dist-install/build/autogen -Ilibraries/ghc-prim/. -optP-include -optPlibraries/ghc-prim/dist-install/build/autogen/cabal_macros.h -package rts-1.0 -package-name ghc-prim -XHaskell98 -XCPP -XMagicHash -XForeignFunctionInterface -XUnliftedFFITypes -XUnboxedTuples -XEmptyDataDecls -XNoImplicitPrelude -O2 -no-user-package-conf -rtsopts -odir libraries/ghc-prim/dist-install/build -hidir libraries/ghc-prim/dist-install/build -stubdir libraries/ghc-prim/dist-install/build -hisuf p_hi -osuf p_o -hcsuf p_hc -c libraries/ghc-prim/./GHC/Magic.hs -o libraries/ghc-prim/dist-install/build/GHC/Magic.p_o /usr/bin/ld: error: /tmp/ghc21839_0/ghc21839_0.p_o uses VFP register arguments, libraries/ghc-prim/dist-install/build/GHC/Magic.p_o does not /usr/bin/ld: failed to merge target specific data of file /tmp/ghc21839_0/ghc21839_0.p_o

7.0.4 was able to be built on this architecture. Do you expect that this is fixed in HEAD? I haven't tried a build.

Full log:  https://launchpadlibrarian.net/94834705/buildlog_ubuntu-precise-armhf.ghc_7.4.1-1ubuntu1_FAILEDTOBUILD.txt.gz

[guest]

Actually I have been informed that we were perhaps always misbuilding for armhf, and the toolchain only recently got the ability to notice an error out. #5914 has got to the bottom of the problem; not passing necessary options to llc.