bug#30282: package julia build error

[ludo@gnu.org (Ludovic Courtès)]

Hey Pjotr,

Pjotr Prins <pjotr.public12@thebird.nl> skribis:

> ls deps/scratch/ 
>
>     arpack-ng-3.3.0
>     cmake-3.7.1-Linux-x86_64
>     curl-7.53.1
>     dsfmt-2.2.3
>     fftw-3.3.6-pl1-double
>     fftw-3.3.6-pl1-single
>     gmp-6.1.2
>     libgit2-2fcb8705e584ca61f6c4657525c9d2713f6a39d2
>     libssh2-30e9c1347e3b8baa2951db612f05e6d87fc8e2f2
>     libunwind-1.1-julia2
>     libuv-52d72a52cc7ccd570929990f010ed16e2ec604c8
>     llvm-3.9.1
>     mbedtls-2.3.0-gpl
>     mpfr-3.1.5
>     openblas-85636ff1a015d04d3a8f960bc644b85ee5157135
>     openlibm-1581174c85f7b645b15ba1ac1c3a98fb601f0fe7
>     openspecfun-39699a1c1824bf88410cabb8a7438af91ea98f4c
>     patchelf-0.9
>     pcre2-10.23
>     SuiteSparse-4.4.5
>     utf8proc-40e605959eb5cb90b2587fa88e3b661558fbc55a
>
> ls deps/patches/
>
>     arpack-tests-blasint.patch         llvm-3.7.1.patch                        llvm-D28759-loopclearance.patch
>     compiler-rt-3.7.1.patch            llvm-3.7.1_symlinks.patch               llvm-D28786-callclearance.patch
>     dSFMT.c.patch                      llvm-3.8.0_bindir.patch                 llvm-D9168_argument_alignment.patch
>     dSFMT.h.patch                      llvm-3.8.0_ppc64_SUBFC8.patch           llvm-nodllalias.patch
>     gmp-exception.patch                llvm-3.8.0_threads.patch                llvm-PR22923.patch
>     instcombine-llvm-3.3.patch         llvm-3.8.0_winshlib.patch               llvm-PR26180.patch
>     int128-vector.llvm-3.3.patch       llvm-3.8.1-version.patch                llvm-PR27046.patch
>     libgit2-agent-nonfatal.patch       llvm-3.9.0_cygwin.patch                 llvm-PR276266.patch
>     libgit2-free-config.patch          llvm-3.9.0_D27296-libssp.patch          llvm-PR277939.patch
>     libgit2-gitconfig-symlink.patch    llvm-3.9.0_threads.patch                llvm-PR278088.patch
>     libgit2-mbedtls.patch              llvm-3.9.0_win64-reloc-dwarf.patch      llvm-PR278321.patch
>     libgit2-mbedtls-verify.patch       llvm-arm-fix-prel31.patch               llvm-PR278923.patch
>     libgit2-mbedtls-writer-fix.patch   llvm-D14260.patch                       llvm-PR29010-i386-xmm.patch
>     libgit2-remote-push-NULL.patch     llvm-D17165-D18583.patch                llvm-r282182.patch
>     libgit2-ssh.patch                  llvm-D17326_unpack_load.patch           llvm-rL279933-ppc-atomicrmw-lowering.patch
>     libssh2-encryptedpem.patch         llvm-D17712.patch                       llvm-rL293230-icc17-cmake.patch
>     libssh2-netinet-in.patch           llvm-D21271-instcombine-tbaa-3.7.patch  llvm-win64-reloc-dwarf.patch
>     libunwind-arm-dyn.patch            llvm-D21271-instcombine-tbaa-3.8.patch  mbedtls-ssl.h.patch
>     libunwind-arm-pc-offset.patch      llvm-D23597_sdag_names.patch            openblas-clangasmbug.patch
>     libunwind-dwarf-ver.patch          llvm-D24300_ptx_intrinsics.patch        openblas-cross-compile.patch
>     libunwind-freebsd-mapper.patch     llvm-D25865-cmakeshlib.patch            openblas-freebsd.patch
>     libunwind-prefer-extbl.patch       llvm-D27389.patch                       openblas-power-assembly-fixes.patch
>     lldb-3.7.1.patch                   llvm-D27397.patch                       osx-10.10.llvm-3.3.patch
>     llvm-3.3.patch                     llvm-D27609-AArch64-UABS_G3.patch       pcre-mingw.patch
>     llvm-3.7.0.patch                   llvm-D27629-AArch64-large_model.patch   SuiteSparse-winclang.patch
>     llvm-3.7.1_2.patch                 llvm-D28009.patch                       win64-int128.llvm-3.3.patch
>     llvm-3.7.1_3.patch                 llvm-D28215_FreeBSD_shlib.patch
>     llvm-3.7.1_destsharedlibdir.patch  llvm-D28221-avx512.patch
>
> I mean, seriously!

Yeah.  :-/

> As I wrote before, the test of the current Guix build probably fails due to the
> openblas linkage Guix uses (openblas output does change and depends on build
> choices). Not sure which switch is causing it, but it should be fixable. There
> are more tests failing, however, that need fixing. 

Indeed there’s always the possibility that test failures are due to a
missing patch.

> For version 7 Julia promises no longer to patch LLVM, see 
>
>   https://discourse.julialang.org/t/why-is-source-build-julia-required-to-use-gpu-packages/9747/2
>
> So, unless someone feels incensed to fix things, I propose to hold out until
> Julia7 and maybe have another go again. 

That sounds reasonable.

It’s a bit weird that 0.6.0 tests passed at some point in Guix, and
eventually started failing.  Not sure what happened.

Ludo’.