Merge LLVM 4.0 Darwin stdenv

[copumpkin]

This issue is to help me and @LnL7 (and anyone else who wants to help) track which packages the LLVM 4 upgrade breaks. I'm harvesting it from here.

• atlasWithLapack.x86_64-darwin
• atomicparsley.x86_64-darwin
• bedtools.x86_64-darwin
• buildbot.x86_64-darwin
• caf.x86_64-darwin
• capnproto.x86_64-darwin
• flann.x86_64-darwin
• grib-api.x86_64-darwin
• haskellPackages.foldl-statistics.x86_64-darwin
• haskellPackages.pipes-concurrency.x86_64-darwin
• iperf2.x86_64-darwin
• lean2.x86_64-darwin
• nix.x86_64-darwin
• nixStable.x86_64-darwin
• nodejs-4_x.x86_64-darwin
• nodejs-slim-4_x.x86_64-darwin
• podofo.x86_64-darwin
• qt4.x86_64-darwin
• qt48.x86_64-darwin
• qt48Full.x86_64-darwin
• qt56.qtbase.x86_64-darwin
• qt57.qtbase.x86_64-darwin
• root.x86_64-darwin
• rtags.x86_64-darwin
• rustc.x86_64-darwin
• swftools.x86_64-darwin
• vtk.x86_64-darwin

[copumpkin]

For capnproto, see capnproto/capnproto#433
For nix/nixStable, see NixOS/nix#1288 and @LnL7's patch to libc++ 4

[LnL7]

I think the qt57 and qt56 failures are just because it was fixed in master recently.

[LnL7]

Looks like nodejs-4_x works with #24329

[copumpkin]

Buildbot appears to be spurious

[copumpkin]

iperf2 fixed in c46600d

[copumpkin]

root is caused by https://sft.its.cern.ch/jira/browse/ROOT-8728. I don't see a clean/easy fix. Might need to use a different version of LLVM to build it.

Edit: they almost have a working patch in that ticket; very responsive team!

[copumpkin]

For rustc, see rust-lang/rust#37609. For now I'll point rustc at LLVM 3.9, since that appears to be working fine (build still hasn't finished but looks like it's moving as expected).

[copumpkin]

@LnL7 fixed rustc in 01d8d1b

[copumpkin]

qt4 and friends seem to be a signed pointer comparison, so should be an easy fix (if the build ever finishes)

[copumpkin]

I have a working qt4 build that I'll push later. Also, caf should be fixed by adding this patch: actor-framework/actor-framework@c5a3ee2 (see actor-framework/actor-framework#545 for more)

[copumpkin]

qt4 and friends fixed in 409fe12

[copumpkin]

caf fixed in 8636ce2

[copumpkin]

rtags in 3a9c217

[copumpkin]

This is now merged into staging

[vcunat]

Now it's in master, but note the two failures in the unstable job. EDIT: I guess nix-repl only needs an update.

[copumpkin]

Thanks! I'll look into the bootstrap tools issue. Nix is currently broken against LLVM 4 but the next 1.11 release will fix it. If that takes too long to appear I'll apply the patch in our expression but I'd prefer not to

…

On Sat, Apr 8, 2017 at 05:05 Vladimír Čunát ***@***.***> wrote: Now it's in master, but note the two failures in the unstable job <http://hydra.nixos.org/build/51317581#tabs-constituents>. — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#24294 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAKP57yDcleNNTmy0N9Vv3WTWzKKnGbks5rt03ZgaJpZM4Mocuu> .

[copumpkin]

@vcunat figured out the stdenvBootstrapTools issue, just running a (long) test build and will push a fix later. Do you think I should just apply an ad-hoc fix to Nix?

[vcunat]

If it's optional stdenv.isDarwin, I think nothing will be harmed.

[copumpkin]

@vcunat it should be pretty harmless on other platforms too, and I don't think Nix causes many rebuilds, right? The patch just makes it conform more strictly to C++ rules (which LLVM 4 now cares about)

[copumpkin]

Okay, both failures should be fixed now 😄

[vcunat]

The bootstrapping is still failing on Hydra: http://hydra.nixos.org/build/51386512 (which is after 72d9016)

[copumpkin]

Yeah there's another issue that I'm about to fix, unrelated to LLVM 4 and caused by our recent purchase of more Macs and an impurity

…

On Sun, Apr 9, 2017 at 03:58 Vladimír Čunát ***@***.***> wrote: The bootstrapping is still failing on Hydra: http://hydra.nixos.org/build/51386512 (which is after 72d9016 <72d9016> ) — You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub <#24294 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AAAKPzU8W44saXRaLddlon71KyiXbVBvks5ruI-sgaJpZM4Mocuu> .

[copumpkin]

So the issue causing http://hydra.nixos.org/build/51386512/nixlog/1 should be fixed in d93f2c3