GNU bug report logs - #38454
[PATCH] gnu: lapack: Update to 3.9.0.

Previous Next

Full log

Message #16 received at 38454-done <at> debbugs.gnu.org (full text, mbox):

From: Ludovic Courtès <ludo <at> gnu.org>
To: Konrad Hinsen <konrad.hinsen <at> fastmail.net>
Cc: Danny Milosavljevic <dannym <at> scratchpost.org>,
 Mathieu Othacehe <m.othacehe <at> gmail.com>, 38454-done <at> debbugs.gnu.org
Subject: Re: [bug#38454] [PATCH] gnu: lapack: Update to 3.9.0.
Date: Tue, 07 Jan 2020 11:20:12 +0100

Hi Konrad,

(Mathieu, Danny: see below for qemu-binfmt concerns.  :-))

Ludovic Courtès <ludovic.courtes <at> inria.fr> skribis:

> Konrad Hinsen <konrad.hinsen <at> fastmail.net> skribis:
>
>> * gnu/packages/maths.scm (lapack): Update to 3.9.0.
>
> Unfortunately this fails to build on armhf-linux (tested via
> qemu-binfmt):
>
> starting phase `configure'
> source directory: "/tmp/guix-build-lapack-3.9.0.drv-0/lapack-3.9.0" (relative from build: "../lapack-3.9.0")
> build directory: "/tmp/guix-build-lapack-3.9.0.drv-0/build"
> running 'cmake' with arguments ("../lapack-3.9.0" "-DCMAKE_BUILD_TYPE=RelWithDebInfo" "-DCMAKE_INSTALL_PREFIX=/gnu/store/ynakjwrsq06j7q75ysqfrmhp8mhhdxrb-lapack-3.9.0" "-DCMAKE_INSTALL_LIBDIR=lib" "-DCMAKE_INSTALL_RPATH_USE_LINK_PATH=TRUE" "-DCMAKE_INSTALL_RPATH=/gnu/store/ynakjwrsq06j7q75ysqfrmhp8mhhdxrb-lapack-3.9.0/lib" "-DCMAKE_VERBOSE_MAKEFILE=ON" "-DBUILD_SHARED_LIBS:BOOL=YES" "-DLAPACKE=ON" "-DLAPACKE_WITH_TMG=ON" "-DBUILD_TESTING=ON")
> -- The Fortran compiler identification is unknown
> CMake Error at /gnu/store/lxj8wisj7ylwc67x1rshiw24fb5digx5-cmake-minimal-3.15.1/share/cmake-3.15/Modules/CMakeCompilerIdDetection.cmake:26 (list):
>   list sub-command REMOVE_ITEM requires two or more arguments.
> Call Stack (most recent call first):
>   /gnu/store/lxj8wisj7ylwc67x1rshiw24fb5digx5-cmake-minimal-3.15.1/share/cmake-3.15/Modules/CMakeDetermineCompilerId.cmake:211 (compiler_id_detection)
>   /gnu/store/lxj8wisj7ylwc67x1rshiw24fb5digx5-cmake-minimal-3.15.1/share/cmake-3.15/Modules/CMakeDetermineCompilerId.cmake:230 (CMAKE_DETERMINE_COMPILER_ID_WRITE)
>   /gnu/store/lxj8wisj7ylwc67x1rshiw24fb5digx5-cmake-minimal-3.15.1/share/cmake-3.15/Modules/CMakeDetermineCompilerId.cmake:32 (CMAKE_DETERMINE_COMPILER_ID_BUILD)
>   /gnu/store/lxj8wisj7ylwc67x1rshiw24fb5digx5-cmake-minimal-3.15.1/share/cmake-3.15/Modules/CMakeDetermineCCompiler.cmake:116 (CMAKE_DETERMINE_COMPILER_ID)
>   CMakeLists.txt:3 (project)
>
> [… repeated several times…]

This turned out to be a side effect of building through qemu-binfmt;
building on real hardware succeeds.  So, patch committed, at last—sorry
for the delay!

--8<---------------cut here---------------start------------->8---
starting phase `configure'
source directory: "/tmp/guix-build-lapack-3.9.0.drv-0/lapack-3.9.0" (relative from build: "../lapack-3.9.0")
build directory: "/tmp/guix-build-lapack-3.9.0.drv-0/build"
running 'cmake' with arguments ("../lapack-3.9.0" "-DCMAKE_BUILD_TYPE=RelWithDebInfo" "-DCMAKE_INSTALL_PREFIX=/gnu/stor
e/ynakjwrsq06j7q75ysqfrmhp8mhhdxrb-lapack-3.9.0" "-DCMAKE_INSTALL_LIBDIR=lib" "-DCMAKE_INSTALL_RPATH_USE_LINK_PATH=TRUE
" "-DCMAKE_INSTALL_RPATH=/gnu/store/ynakjwrsq06j7q75ysqfrmhp8mhhdxrb-lapack-3.9.0/lib" "-DCMAKE_VERBOSE_MAKEFILE=ON" "-
DBUILD_SHARED_LIBS:BOOL=YES" "-DLAPACKE=ON" "-DLAPACKE_WITH_TMG=ON" "-DBUILD_TESTING=ON")
-- The Fortran compiler identification is GNU 7.4.0
-- The C compiler identification is GNU 7.4.0
-- Check for working Fortran compiler: /gnu/store/91iyya22d6s055iwlwja3vvlpc4i8sih-gfortran-7.4.0/bin/gfortran
-- Check for working Fortran compiler: /gnu/store/91iyya22d6s055iwlwja3vvlpc4i8sih-gfortran-7.4.0/bin/gfortran  -- works
-- Detecting Fortran compiler ABI info
-- Detecting Fortran compiler ABI info - done
-- Checking whether /gnu/store/91iyya22d6s055iwlwja3vvlpc4i8sih-gfortran-7.4.0/bin/gfortran supports Fortran 90
-- Checking whether /gnu/store/91iyya22d6s055iwlwja3vvlpc4i8sih-gfortran-7.4.0/bin/gfortran supports Fortran 90 -- yes
-- Check for working C compiler: /gnu/store/pkx3kcvj1mbkkgzxi1w57c1vfvhwzxh7-gcc-7.4.0/bin/gcc
-- Check for working C compiler: /gnu/store/pkx3kcvj1mbkkgzxi1w57c1vfvhwzxh7-gcc-7.4.0/bin/gcc -- works
-- Detecting C compiler ABI info
-- Detecting C compiler ABI info - done
--8<---------------cut here---------------end--------------->8---


So, what’s going on under qemu-binfmt?

When doing a qemu-binfmt build, ‘CMakeOutput.log’ contains several
instances of:

--8<---------------cut here---------------start------------->8---
The system is: Linux - 5.4.8-gnu - armv7l
Compiling the Fortran compiler identification source file "CMakeFortranCompilerId.F" succeeded.
Compiler: /gnu/store/91iyya22d6s055iwlwja3vvlpc4i8sih-gfortran-7.4.0/bin/gfortran 
Build flags: 
Id flags: -v 
--8<---------------cut here---------------end--------------->8---

Yet ‘CMakeError.log’ reads:

--8<---------------cut here---------------start------------->8---
Compilation of the Fortran compiler identification source "CMakeFortranCompilerId.F" did not produce an executable in "/tmp/guix-build-lapack-3.9.0.drv-0/build/CMakeFiles/3.15.1/CompilerIdFortran".
--8<---------------cut here---------------end--------------->8---

I do see an ‘a.out’ file in there, though.

The code for that is in ‘CMakeDetermineCompilerId.cmake’:

--8<---------------cut here---------------start------------->8---
    # Find the executable produced by the compiler, try all files in the
    # binary dir.
    string(REGEX REPLACE "([][])" "[\\1]" _glob_id_dir "${CMAKE_${lang}_COMPILER_ID_DIR}")
    file(GLOB files
      RELATIVE ${CMAKE_${lang}_COMPILER_ID_DIR}

      # normal case
      ${_glob_id_dir}/*

      # com.apple.package-type.bundle.unit-test
      ${_glob_id_dir}/*.xctest/*

      # com.apple.product-type.framework
      ${_glob_id_dir}/*.framework/*
      )
    list(REMOVE_ITEM files "${src}")
    set(COMPILER_${lang}_PRODUCED_FILES "")
    foreach(file ${files})
      if(NOT IS_DIRECTORY ${CMAKE_${lang}_COMPILER_ID_DIR}/${file})
        list(APPEND COMPILER_${lang}_PRODUCED_FILES ${file})
        file(APPEND ${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeOutput.log
          "Compilation of the ${lang} compiler identification source \""
          "${src}\" produced \"${file}\"\n\n")
      endif()
    endforeach()

    if(NOT COMPILER_${lang}_PRODUCED_FILES)
      # No executable was found.
      file(APPEND ${CMAKE_BINARY_DIR}${CMAKE_FILES_DIRECTORY}/CMakeError.log
        "Compilation of the ${lang} compiler identification source \""
        "${src}\" did not produce an executable in \""
        "${CMAKE_${lang}_COMPILER_ID_DIR}\".\n\n")
    endif()
--8<---------------cut here---------------end--------------->8---

Why would that fail?  Mystery…

There’s some code in CMake that looks at /proc/cpuinfo, so that could
somehow make the wrong decision, but I don’t see how.  Also, why would
it affect emulation of ARMv7 but not emulation of AArch64, as seems to
be the case?

Ideas, people?

Thanks,
Ludo’.

Previous Next

GNU bug tracking system
Copyright (C) 1999 Darren O. Benham, 1997,2003 nCipher Corporation Ltd, 1994-97 Ian Jackson.