Ctest failure on ARM64

[ccassidy-vaisala]

Trying to compile on ARM64 and ctest is throwing an error on a single test with v2.0.0, while x86 works fine. Is than an expected outcome for ARM64 tests? Thanks!

ARM64 log:

24: Test command: /NCEPLIBS-g2c-2.0.0/build/tests/tst_io
24: Working Directory: /NCEPLIBS-g2c-2.0.0/build/tests
24: Test timeout computed to be: 1500
24: Testing g2c I/O functions.
24: Testing simple calls with 4-byte ints...ok!
24: Testing template calls with 4-byte ints...ok!
24: Testing template calls with 1-byte ints...
24/42 Test #24: tst_io ...........................***Failed    0.00 sec

x86 log:

24: Test command: /NCEPLIBS-g2c-2.0.0/build/tests/tst_io
24: Working Directory: /NCEPLIBS-g2c-2.0.0/build/tests
24: Test timeout computed to be: 1500
24: Testing g2c I/O functions.
24: Testing simple calls with 4-byte ints...ok!
24: Testing template calls with 4-byte ints...ok!
24: Testing template calls with 1-byte ints...ok!
24: Testing template calls with 2-byte ints...ok!
24: Testing template calls with 8-byte ints...uval_in 1844674407370955161
24: val_in -922337203685477580
24: ok!
24: SUCCESS!
24/42 Test #24: tst_io ...........................   Passed    0.00 sec

[edwardhartnett]

OK, we need to add arm runners to our CI to test for this. There's an example here:
https://learn.arm.com/learning-paths/cross-platform/github-arm-runners/actions/#:~:text=Arm%2Dhosted%20runners-,Introduction,with%20your%20Personal%20Access%20Token.

[EricEngle-NOAA]

We have experience this same failure with the g2c builds for conda-forge (https://github.com/conda-forge/g2clib-
feedstock/pull/42). The culprit is the reading of the 2nd value for 1-byte ints

NCEPLIBS-g2c/tests/tst_io.c

Line 169 in 5f79509

if ((ret = g2c_file_io_byte(f, G2C_FILE_READ, &val_in)))

A value of -120 is written to file as char, but when read from the file into a char, I am seeing a value of -8 in my testing on my local Ubuntu Linux Arm VM.

[opoplawski]

I'm seeing the same for the Fedora builds.