Added linux-musl support

[petermz]

Hi,
The only thing that prevents fastr from building on musl is the #include "fpu_control.h" line somewhere in libf2c sources. This header file is missing in musl. I'm adding this (effectively empty) header file as a patch to f2c/libf2c-patch

[steve-s]

Hello,

thank you for the PR. How does it work if the extern __fpu_control exposed by the glibc header is not there? Does linux-musl export it also but not in that header?

[petermz]

No, musl doesn't have it in its headers. My theory is that __fpu_control use sites are #ifdef'ed out and not compiled with musl

[steve-s]

Thanks for the explanation.

Have you signed the Oracle Contributor Agreement? We use it to make the copyright of contributions clear. You can sign it at that link.

[graalvmbot]

Hello peterz, thanks for contributing a PR to our project!

We use the Oracle Contributor Agreement to make the copyright of contributions clear. We don't have a record of you having signed this yet, based on your email address peterz -(at)- disroot -(dot)- org. You can sign it at that link.

If you think you've already signed it, please comment below and we'll check.

[petermz]

Hi,
I believe I should be covered by Bellsoft's umbrella OCA.

I've identified another undefined symbol and added it to unimplemented.c, hence the second commit. (It shows on musl because musl linkers use -z now by default, and libRnative fails to load).

[steve-s]

I've identified another undefined symbol and added it to unimplemented.c, hence the second commit. (It shows on musl because musl linkers use -z now by default, and libRnative fails to load).

good, thanks. Could you please also add the standard Oracle GPL3 copyright that we use into the new file. Then I'll spin our internal tests and merge if these changes pass. (Example of the copyright: https://github.com/oracle/fastr/blob/master/com.oracle.truffle.r.test.native/embedded/src/main.c)

I believe I should be covered by Bellsoft's umbrella OCA.

yes that should be OK, could you please give some proof that you're from that company (send me an e-mail from an account on that domain or add secondary e-mail here on GitHub)? Sorry I need to follow the procedures.

[petermz]

I already have [email protected] as secondary email here, and I've just sent you an email from that address.
Added copyright header, too. Thanks!

[steve-s]

Ah, sorry I forgot that we're using prebuilt F2C in our CI and check that F2C still builts correctly only in post merge checks, which gave us this error:

In file included from ./fpu_control.h:26:
./fpu_control.h:26:2: warning: #include_next in file found relative to primary source file or found by absolute path; will search from start of include path [-Winclude-next-absolute-path]
#include_next "fpu_control.h"
 ^
./fpu_control.h:26:15: error: #include nested too deeply
#include_next "fpu_control.h"
              ^
199 warnings and 1 error generated.
make[2]: *** [makefile:30: uninit.o] Error 1
make[2]: Leaving directory 'fastr/libdownloads/f2c/src/libf2c'
make[1]: *** [Makefile:60: all] Error 2
make[1]: Leaving directory 'fastr/com.oracle.truffle.r.native/f2c'
make: *** [Makefile:70: all] Error 2

Seems that the #include_next did not work as expected?

[petermz]

Strange it didn't show in my testing, but now I can reproduce this. Does it sometimes work and sometimes not?

Also, how do you build using prebuilt F2C?

[steve-s]

Looking at it now, I think we have a bug in com.oracle.truffle.r.native/f2c/Makefile we do not clean the right f2c copy (there's one in libdownloads that mx downloads for us, and one in f2c top level dir, where we copy the former). Edit: no, actually we wipe out the f2c copy, so that should be alright. So if you do rm -rf f2c and then mx build you should be actually building f2c from scratch (edit: mx clean should do that too).

In the same makefile you can see that we check for environment variable F2C_BINARY and if that's set we assume it points to pre-built f2c and we just copy it over and skip the building altogether.

[petermz]

I have a fix: #176