Error building with gfortran DESPITE using -fallow-argument-mismatch -fallow-invalid-boz

[GrigorisLionis]

Requirements

• Reviewed ESMF User's Guide
• Searched GitHub Discussions

Affiliation(s)

No response

ESMF Version

v.8.6.1.

Issue

I am trying to build latest stable version from source (a.k.a. 8.6.1.)
I have encountered a problem compiling the sources, due to a type mismatch error
The output of the error is as following:
gfortran -c -fallow-argument-mismatch -fallow-invalid-boz -O -fPIC -pthread -ffree-line-length-none -fopenmp -I/home/user/esmf/esmf-8.6.1/src/Infrastructure/Util/src -I/home/user/esmf/esmf-8.6.1/build_config/Linux.gfortran.default -I/home/user/esmf/esmf-8.6.1/src/Infrastructure -I/home/user/esmf/esmf-8.6.1/src/Superstructure -I/home/user/esmf/esmf-8.6.1/mod/modO/Linux.gfortran.32.mpiuni.default -I/home/user/esmf/esmf-8.6.1/src/include -DESMF_NO_INTEGER_1_BYTE -DESMF_NO_INTEGER_2_BYTE -DESMF_MPIUNI -I/home/user/esmf/esmf-8.6.1/src/Infrastructure/stubs/mpiuni -DESMF_LOWERCASE_SINGLEUNDERSCORE -DESMF_MOAB=1 -DESMF_LAPACK=1 -DESMF_LAPACK_INTERNAL=1 -DESMF_NO_ACC_SOFTWARE_STACK=1 -DESMF_YAMLCPP=1 -DESMF_YAML=1 -DESMF_NO_OPENACC -DESMF_BOPT_O -DESMF_TESTCOMPTUNNEL -DS32=1 -DESMF_OS_Linux=1 -DESMF_COMM=mpiuni -DESMF_DIR=/home/user/esmf/esmf-8.6.1 /home/user/esmf/esmf-8.6.1/src/Infrastructure/Util/src/ESMF_FortranWordsize.F90 -o /home/user/esmf/esmf-8.6.1/obj/objO/Linux.gfortran.32.mpiuni.default/ESMF_FortranWordsize.o
/home/user/esmf/esmf-8.6.1/src/Infrastructure/Util/src/ESMF_FortranWordsize.F90:425:77:

425 | call ESMF_PointerDifference(C_SIZEOF (c_ptrvar), varTK(1), varTK(2), wsize)
| 1
Error: Type mismatch in argument ‘n’ at (1); passed INTEGER(4) to INTEGER(8)
/home/user/esmf/esmf-8.6.1/src/Infrastructure/Util/src/ESMF_FortranWordsize.F90:380:77:

380 | call ESMF_PointerDifference(C_SIZEOF (c_ptrvar), varTK(1), varTK(2), wsize)
| 1
Error: Type mismatch in argument ‘n’ at (1); passed INTEGER(4) to INTEGER(8)
/home/user/esmf/esmf-8.6.1/src/Infrastructure/Util/src/ESMF_FortranWordsize.F90:335:77:

335 | call ESMF_PointerDifference(C_SIZEOF (c_ptrvar), varTK(1), varTK(2), wsize)
| 1
Error: Type mismatch in argument ‘n’ at (1); passed INTEGER(4) to INTEGER(8)
/home/user/esmf/esmf-8.6.1/src/Infrastructure/Util/src/ESMF_FortranWordsize.F90:290:77:

290 | call ESMF_PointerDifference(C_SIZEOF (c_ptrvar), varTK(1), varTK(2), wsize)
| 1
Error: Type mismatch in argument ‘n’ at (1); passed INTEGER(4) to INTEGER(8)
/home/user/esmf/esmf-8.6.1/src/Infrastructure/Util/src/ESMF_FortranWordsize.F90:113:79:

113 | call ESMF_PointerDifference(C_SIZEOF (c_ptrvar), udt_ptr1, udt_ptr2, psize)
| 1
Error: Type mismatch in argument ‘n’ at (1); passed INTEGER(4) to INTEGER(8)
/home/user/esmf/esmf-8.6.1/src/Infrastructure/Util/src/ESMF_FortranWordsize.F90:442:7:

442 | use ESMF_FortranWordsizeMod
| 1
Fatal Error: Cannot open module file ‘esmf_fortranwordsizemod.mod’ for reading at (1): Δεν υπάρχει τέτοιο αρχείο ή κατάλογος
compilation terminated.

I have used a number of different fortran compilers with the same result.

Autotag

@oehmke

[danrosen25]

@GrigorisLionis
What version of gfortran is being used? Can you confirm that you're setting the build comm layer to 'mpiuni' and the build option to 'O'?

[GrigorisLionis]

Hello, thanks for the reply.

Yes, I am building the O version, with ESMF_COMM set to mpiuni.
I am using gfortran-13.3.0, but the same compile error pops up in many different gfortran used
( 1 /usr/bin/gcc-11/bin/gfortran 80 manual mode
2 /usr/bin/gfortran-4.4 50 manual mode
3 /usr/bin/gfortran-4.6 60 manual mode
4 /usr/bin/gfortran-4.9 40 manual mode
5 /usr/bin/gfortran-5 10 manual mode
6 /usr/bin/gfortran-9 20 manual mode
7 /usr/gcc-trunk/bin/gfortran 30 manual mode
8 /usr/local/gcc-13.3/bin/gfortran-13.3 50 manual mode)

[danrosen25]

These compilers have been tested for building ESMF v8.6.1 so I don't think this issue is related to the compiler.

• gfortran 13.2.0
• gfortran 12.2.0
• gfortran 11.2.0
• gfortran 8.3.0
• gfortran 7.4.0

I spoke to the core team and there are two potential system related reasons

• one of the settings provided to the ESMF build system is interfering with the build
• the gfortran built-in spec is interfering with the build by overriding the ESMF build settings

Can you run 'make info' and let me know what it says?

[GrigorisLionis]

I am sending you the output of the "make info"
make.info.txt
The problems seems to be a mixup of pointer sizes, between c and fortran code, in various parts of the autogenerated code.
The 32bit OS certainly does not help.

[danrosen25]

@theurich
Are we testing on any 32-bit ABI systems and do we have access to test on any 32-bit ABI systems?

[theurich]

We have not run nightly tests on a 32-bit ABI system in many years. Interacting via this discussion with @GrigorisLionis to get the ESMF internal details for this configuration correct is probably our best bet to move forward here. Let me take a look at the provided make.info.txt...

[theurich]

@danrosen25 - I think I see where the 32-bit ABI support in ESMF was broken (back in 2019). This discussion should probably migrate over as an issue to fix the 32-bit ABI support.

[danrosen25]

@GrigorisLionis
If this is the case then it might be some time before we can resolve support for 32 bit ABI. We've opened a ticket to follow up on 32 bit support
#254

[GrigorisLionis]

Guys, thank you very much for your answers. I have to say that I stumbled across ESMF looking for a easy way to regrid a NetCDF file using Python in a legacy machine, and ESMPy looked an excellent choice. It is much easier for my to change to 64 bit... (actually, -layman's opinion- in retrospect it is somewhat peculiar for you to have , for us to seek, 32 compatibility in this kind of code...)

[theurich]

Well, 32-bit ABI support was supposed to work out-of-the-box, so it was a surprise when you reported that it wasn't working. If you could give the branch suggested under #254 a quick test on your 32-bit system, that would help us out greatly. Thanks!

[danrosen25]

@GrigorisLionis
We also have a Docker image with ESMF and ESMPy preinstalled. We've created native images for both arm64 and amd64. I don't know if Docker can emulate amd64 on an i686 machine well but if you have either an arm64 or amd64 machine then you can use the Docker image for small projects.
https://hub.docker.com/r/esmf/esmf-build-release/tags
Once we've resolve the 32 bit ABI issue I can consider building a native image for i686 too.

[GrigorisLionis]

Hello, happy to help.

git clone https://github.com/esmf-org/esmf.git --branch v8.6.2b00 --depth 1
I have used this branch

make.info.txt
I am sending you the make info output

Same result as 8.6.1
The compilation stopped on the same file

I tried to "hack away" the easy problem (INT4->INT8) in that file (a function had a hardcoded 64bit int, and i used a INT8 to cast)
But another problem arose in a subsequent file

(INT8->INT4)
In that file, the function is defined with an C_INT argument (which presumably is a 32bit in my system) but it is called with an 64bit ESMF_KIND_I8

[GrigorisLionis]

CORRECTION
Sorry, have used different 8.6.2

when using
user@PC078 ~/esmf/esmf $ git clone https://github.com/esmf-org/esmf.git --branch patch/8.6.2
make.info.txt

the compilation is stuck at the second file

[theurich]

@GrigorisLionis - yes --branch patch/8.6.2 is what you want! Good to know about the next issue in ESMF_Info when you test this on your 32-bit system. I will take a look at that one next and let you know.
I suggest that we move to #254 as we iterate on the details of this fix.