Version 1.14.5 fortran lib fails to build on arch i386 - Error: There is no specific subroutine for the generic ‘verify’ at (1)

[pini-gh]

Describe the bug
While building on current Debian unstable i386 with gfortran 14.2.0 the build fails with:

+ gfortran -I. -I../../../../fortran/test -I../../src -I../../fortran/src -std=f2008 -Waliasing -Wall -Wcharacter-truncation -Wextra -Wimplicit-interface -Wsurprising -Wunderflow -pedantic -Wintrinsics-std -Wimplicit-procedure -Wreal-q-constant -Wfunction-elimination -Wrealloc-lhs -Wrealloc-lhs-all -Wno-c-binding-type -Winteger-division -Wfrontend-loop-interchange -fdiagnostics-urls=never -fno-diagnostics-color -s -Wno-unused-dummy-argument -Wno-array-temporaries -O3 -I../../fortran/src -I../../fortran/src -g -O2 -ffile-prefix-map=/<<PKGBUILDDIR>>=. -fstack-protector-strong -c -o tH5T_F03.o ../../../../fortran/test/tH5T_F03.F90
../../../../fortran/test/tH5T_F03.F90:3009:17:

 3009 |         IF(.NOT.(orig_data(i,j).EQ.orig_data(i,j))) CYCLE  ! skip IF value is NaN
      |                 1
Warning: Equality comparison for REAL(4) at (1) [-Wcompare-reals]
../../../../fortran/test/tH5T_F03.F90:1184:81:

 1184 |        CALL verify("h5kind_to_type",dset_data_r31(i),data_out_r31(i),total_error)
      |                                                                                 1
Error: There is no specific subroutine for the generic ‘verify’ at (1)
make[4]: *** [Makefile:1563: tH5T_F03.o] Error 1

Full build log: https://buildd.debian.org/status/fetch.php?pkg=hdf5&arch=i386&ver=1.14.5%2Brepack-1%7Eexp1&stamp=1730037856&raw=0

Expected behavior
Successful build.

Platform (please complete the following information)

• HDF5 version 1.14.5
• OS and version: Debian unstable i386
• Compiler and version: GCC 14.2.0
• Build system (e.g. CMake, Autotools) and version: Autotools
  • autoconf 2.72
  • automake 1.16.5
• Any configure options you specified
  cd debian/build-serial && ../../configure --build=i686-linux-gnu --prefix=/usr --includedir=\${prefix}/include --mandir=\${prefix}/share/man --infodir=\${prefix}/share/info --sysconfdir=/etc --localstatedir=/var --disable-option-checking --disable-silent-rules --libdir=\${prefix}/lib/i386-linux-gnu --libexecdir=\${prefix}/lib/i386-linux-gnu --disable-maintainer-mode --disable-dependency-tracking --prefix=/usr --host=i686-linux-gnu --build=i686-linux-gnu --libdir=\${prefix}/lib/i386-linux-gnu --mandir=\${prefix}/share/man --with-pthread --enable-linux-lfs --enable-shared --enable-build-mode=production --disable-sharedlib-rpath --with-zlib --with-default-api-version=v114 --with-szlib --enable-fortran --enable-fortran2003 --enable-hltools --enable-ros3-vfd --includedir=\${prefix}/include/hdf5/serial --with-flavor=serial --with-default-plugindir=/usr/lib/i386-linux-gnu/hdf5/serial/plugins --enable-unsupported --enable-threadsafe --enable-cxx --enable-java
• MPI library and version (parallel HDF5) : not relevant (serial build)

Additional context
This error doesn't occur when building for arch amd64 (x86_64).

[hyoklee]

@pini-gh , thank you for your report!
FYI, this problem occurs on i386 + Alpine + CMake as well.

[pini-gh]

Any fix available?

[pini-gh]

May I suggest a higher priority? This issue is a blocker for transitioning HDF5 1.14 into the next Debian stable release (trixie).

[derobins]

Hi @pini-gh !

We've been slowly moving away from supporting i386 since, well, nobody really supports it anymore. We are planning to get some i386 CI up, though, if only to see if we're making bad type size assumptions that we'll see in the warning output.

Since we do want Debian to deploy recent versions of HDF5, I'll have our Fortran people look at this ASAP (they say it's not going to be that hard to fix). As for a fix, we are retiring the 1.14 maintenance line, though we might have a 1.14.6 release next year (probably in the late spring, after 2.0 goes out). Can Debian work with a source patch until then? I think the fix will be fairly small.

[pini-gh]

Can Debian work with a source patch until then? I think the fix will be fairly small.

Sure. Thanks!

[brtnfld]

@pini-gh, can you try it with a cmake build? For our Alpine i386 CI testing, I could not reproduce the error about an invalid KIND. The only Fortran test that fails is with h5rget_obj_name_f with optional parameters and a blank string (or a smaller string) being passed in, which is strange because, with the optional parameters, it does not use the string. It seg faults when returning from the Fortran call; the C call is successful. If I specify a specific string length to pass to the API, I can get it to pass. It is probably a compiler issue; the gfortran version was 13.

[pini-gh]

@brtnfld
Unfortunately the Debian package does an autotools build. Il would be difficult for me to adapt it for a cmake build. But I could try and provide you with a docker image to reproduce the error into a Debian unstable i386 environment. What do you think?

[brtnfld]

Yes, a docker image would be helpful.

[pini-gh]

First, create this Dockerfile:

$ cat Dockerfile-hdf5-debian-i386 
FROM debian:unstable-20241111

ENV DEBIAN_FRONTEND=noninteractive
RUN echo "deb-src http://ftp.debian.org/debian/ experimental main" >/etc/apt/sources.list.d/experimental-src.list
RUN apt update && apt install -y build-essential git && apt -y build-dep hdf5

RUN adduser --quiet hdf5
USER hdf5
RUN cd && git clone -b hdf5_1.14.5 https://github.com/HDFGroup/hdf5
WORKDIR /home/hdf5/hdf5

Then build the image:

$ docker build -t hdf5-debian-i386 --platform 386 -f Dockerfile-hdf5-debian-i386 .

Run the image:

$ docker run -it --rm hdf5-debian-i386

Finally configure and build into the container:

$ ./configure --disable-silent-rules --enable-build-mode=production --enable-fortran --enable-fortran2003 --disable-hltools
$ make

The build should end with:

+ gfortran -I. -I../../src -I../../fortran/src -I../../fortran/src -I../../fortran/src -c -o tH5T_F03.o tH5T_F03.F90
tH5T_F03.F90:1184:81:

 1184 |        CALL verify("h5kind_to_type",dset_data_r31(i),data_out_r31(i),total_error)
      |                                                                                 1
Error: There is no specific subroutine for the generic 'verify' at (1)
make[2]: *** [Makefile:1557: tH5T_F03.o] Error 1
make[2]: Leaving directory '/home/hdf5/hdf5/fortran/test'
make[1]: *** [Makefile:915: all-recursive] Error 1
make[1]: Leaving directory '/home/hdf5/hdf5/fortran'
make: *** [Makefile:732: all-recursive] Error 1

[pini-gh]

First bad commit after git bisect: 9b5d968

commit 9b5d9680af8401528bb8c3b6d2b4c3cf30ccec5b
Author: Dana Robinson <43805+derobins@users.noreply.github.com>
Date:   Mon Jul 15 05:38:13 2024 -0700

    Clean up Fortran __float128 configure-time checks (#4649)
    
    * Always use DECIMAL_DIG instead of LDBL_DIG. This was controlled by
      an ifdef that is always true in C99 or greater
    
    It's confusing to use float.h C constants as variable names in
    configure.ac and the PAC_FC_LDBL_DIG macro.
    
    * Directly compare MY_FLT128_DIG and MY_LDBL_DIG
    
    * Make uniform across CMake and Autotools
    * Don't export quadmath.h variables to H5pubconf.h

 config/cmake/ConfigureChecks.cmake | 102 ++++++++++++++++++++++---------------
 config/cmake/H5pubconf.h.in        |  11 +---
 config/cmake/HDF5UseFortran.cmake  |  23 ++++-----
 config/cmake/HDFTests.c            |  12 +++++
 configure.ac                       |  88 ++++++++++++++++++++++++--------
 fortran/test/tH5T_F03.F90          |   4 --
 m4/aclocal_fc.m4                   |  38 +++++++-------
 7 files changed, 171 insertions(+), 107 deletions(-)

[pini-gh]

Actually this issue is a duplicate of #4916 and the fix is 0af668f.