All this does is enforce the default behaviour, I see no problem.

@micaeljtoliveira Given your feedback in easybuilders/easybuild-easyconfigs#6922, thoughts on this change, do you agree it makes sense to set both --build & --host, regardless of the buggy way in which the Hypre configure script is "written"?

@boegel I'm no expert in cross-compilation or this specific part of the autotools, but if their documentation is to be trusted, setting both --build and --host should be fine. Personally I would avoid doing this systematically as a workaround for a bug in a specific package and I would patch Hypre instead, but this is obviously up to you.

In case you are wondering, this is how the patch should look like for Hypre 2.11.2 (assuming I understood the problem correctly):

2330a2331,2332
> if test "x$build_alias" = "x"
> then
2367a2370,2374
> 
> fi
> 
> if test "x$host_alias" = "x"
> then
2399a2407,2409
> 
> fi
> 
7230c7240
<    if test $host = $build
---
>    if test $host_alias = $build_alias

@micaeljtoliveira Thanks for the feedback, much appreciated!

This problem may not be specific to Hypre at all, who knows... I'm pretty sure there's software out there we haven't played with yet. ;)

Also, since there should be no harm in setting both (especially since it doesn't prevent you from actually cross-compiling thanks to the build_type and host_type easyconfig parameters now supported by ConfigureMake), I think this is fine.

This approach has the added benefit that we don't need to touch the Hypre easyconfigs (which worked fine before we started setting (only) --build), and I couldn't find a clear reference to people reporting the invalid value of canonical host error for Hypre, so it seems to me we're in uncharted territory when only setting --build, which scares me a bit...

I agree with you that it's not entirely clear what the best way forward is here. But my gut feeling (which may be wrong, time will tell) suggests that also setting --host in ConfigureMake will result in less headaches going forward.