0.5.0 broken on AArch64 with GFortran

[sheepforce]

The latest 0.5.0 release broke the builds on AArch64 machines with GFortran, I assume somehow by the changes from #21 . Here is an attached build log with compiler and library versions as well as the error itself:
testdrive.txt

The important part:

test-drive-aarch64-unknown-linux-gnu> FAILED: libtest-drive.so.0.5.0.p/src_testdrive.F90.o libtest-drive.so.0.5.0.p/testdrive.mod
test-drive-aarch64-unknown-linux-gnu> aarch64-unknown-linux-gnu-gfortran -Ilibtest-drive.so.0.5.0.p -I. -I.. -fdiagnostics-color=always -D_FILE_OFFSET_BITS=64 -Wall -ffree-line-length-none -fbacktrace -DWITH_QP=1 -DWITH_XDP=1 -fPIC -Jlibtest-drive.so.0.5.0.p -o libtest-drive.so.0.5.0.p/src_testdrive.F90.o -c ../src/testdrive.F90
test-drive-aarch64-unknown-linux-gnu> ../src/testdrive.F90:1257:42:
test-drive-aarch64-unknown-linux-gnu>  1257 |   subroutine check_complex_exceptional_xdp(error, actual, message, more)
test-drive-aarch64-unknown-linux-gnu>       |                                          1
test-drive-aarch64-unknown-linux-gnu> ......
test-drive-aarch64-unknown-linux-gnu>  1354 |   subroutine check_complex_exceptional_qp(error, actual, message, more)
test-drive-aarch64-unknown-linux-gnu>       |                                         2
test-drive-aarch64-unknown-linux-gnu> Error: Ambiguous interfaces in generic interface 'check' for 'check_complex_exceptional_xdp' at (1) and 'check_complex_exceptional_qp' at (2)
test-drive-aarch64-unknown-linux-gnu> ../src/testdrive.F90:1965:2:
test-drive-aarch64-unknown-linux-gnu>  1965 |   elemental function is_nan_xdp(val) result(is_nan)
test-drive-aarch64-unknown-linux-gnu>       |  1
test-drive-aarch64-unknown-linux-gnu> ......
test-drive-aarch64-unknown-linux-gnu>  1977 |   elemental function is_nan_qp(val) result(is_nan)
test-drive-aarch64-unknown-linux-gnu>       |  2
test-drive-aarch64-unknown-linux-gnu> Error: Ambiguous interfaces in generic interface 'is_nan' for 'is_nan_xdp' at (1) and 'is_nan_qp' at (2)
test-drive-aarch64-unknown-linux-gnu> ../src/testdrive.F90:1886:2:
test-drive-aarch64-unknown-linux-gnu>  1886 |   pure function complex_xdp_to_string(val) result(string)
test-drive-aarch64-unknown-linux-gnu>       |  1
test-drive-aarch64-unknown-linux-gnu> ......
test-drive-aarch64-unknown-linux-gnu>  1897 |   pure function complex_qp_to_string(val) result(string)
test-drive-aarch64-unknown-linux-gnu>       |  2
test-drive-aarch64-unknown-linux-gnu> Error: Ambiguous interfaces in generic interface 'to_string' for 'complex_xdp_to_string' at (1) and 'complex_qp_to_string' at (2)
test-drive-aarch64-unknown-linux-gnu> ninja: build stopped: subcommand failed.

[awvwgk]

This seems to be an issue with using extended double and quadruple precision detection on aarch64. Can you check whether building with -Dxdp=disabled in meson or -DWITH_XDP=off in CMake the build is passing?

[sheepforce]

Yes, -Dxdp=disabled indeed fixes the problem on AArch64 and cross compilation! 👍

[awvwgk]

I believe the issue is due to cross compilation, in this case we cannot easily check whether support is available by compiling and running a test program, instead those options should be defined with the cross compilation environment.

One solution from the test-drive side could be to disable the features in case of cross compilation.

[sheepforce]

No its not an issue of cross compilation, also does not work on our native AArch64 CI builders. I was just able to reproduce the issue with cross compilation and tested with it initially :)