Error building with gfortran on 32 bit ABI machine

[danrosen25]

Discussed in https://github.com/orgs/esmf-org/discussions/251

^{Originally posted by GrigorisLionis May 28, 2024}

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.

[theurich]

@GrigorisLionis at your convenience please give https://github.com/esmf-org/esmf/tree/patch/8.6.2 a try on your 32-bit system. Let us know how it goes. Thanks!

[theurich]

@GrigorisLionis - I just pushed another set of changes to https://github.com/esmf-org/esmf/tree/patch/8.6.2. Please test under your 32-bit system at your convenience. Thank you!

[GrigorisLionis]

This version compiles ok!!

make test resulted in some errors

I am sending you the results of the test
Linux.gfortran.32.mpiuni.default.tar.gz

[theurich]

Well, we are making progress. Thank you for testing! Two requests: (1) Please set ESMF_BOPT=g in your build environment, (2) unset ESMF_F90COMPILEOPTS (to confirm it is no longer needed). Then rebuild from a fresh clone and run the tests. I hope we get more info in the back traces with debug enabled. Thanks.

[GrigorisLionis]

We sure are. Testing the least I can do!
-compiled ok with seettings with fresh copy from git)
make.info.txt
-I am sending you all the results of the test
testg.tar.gz

[theurich]

@GrigorisLionis - Thanks again. I looked over the more complete back traces from your debug run. It is going to be touch to figure out what exactly is causing those crashes especially not having a 32-bit environment myself. You might be better off using a 64-bit system after all for now.
Alternatively you could try using this build in your ESMPy project and see if it holds up. A lot seems to function correctly, except something mostly around XGrid, which might not be an issue in your case. But of course it's a bit shaky not knowing exactly what those crashes are due.

@danrosen25 and @uturuncoglu - is it possible to set up a 32-bit runner for our own testing of ESMF? We probably need that anyway if we want to reliably support this build option again forward going.

[danrosen25]

@theurich
GitHub doesn't provide 32 bit runners. We can try to build a i386 image for Docker and emulate it through the Docker VM. I think you can force emulation with docker run --rm -it --platform linux/386 esmf/esmf-build-stack bash -l then we can maybe run this image through the GitHub actions to try to build ESMF.