-pthread appended to CFLAGS even though the compiler is IBM XLC

[dcassioli]

https://github.com/curl/curl/blob/master/configure.ac#:~:text=AC_CHECK_FUNC(pthread_create%2C%20%5BUSE_THREADS_POSIX,PTHREAD%3D%22%20%2Dpthread%22

In configure.ac (and the generated configure script) the -pthread flag gets appended to CFLAGS even in the case of IBM XLC compiler. The issue was experienced in AIX 7.1 with XLC version 11.1.

The -pthread flag has no meaning in XLC, and it's probably getting interpreted as -p (enable profiling) -t (list of toolchain programs to prepend the prefix specified by the -B flag to, i.e., 'a' for 'assembler', 'l' for 'linker', etc.), and then 'hread' is parsed as argument for -t (which is garbage, so the flag is simply discarded with a warning).

In discussion: https://curl.se/mail/lib-2019-03/0049.html a possible fix is proposed.

[bagder]

The patch back then seems to suggest to check for *-ibm-aix* and treat it like hpux is already treated in there: it simply skips adding a command line option but then also disables threading (and the threaded resolver). It's a very crude "fix".

1.Isn't there a chance that gcc/clang can run on AIX and have such an option?
2. Don't you rather want the corresponding xlc take used so that you can use the threaded resolver?

[bagder]

I think it would make sense if you would write up a PR for the change you think makes sense for you on that platform. We don't have many AIX users or developers around.

[dcassioli]

The patch back then seems to suggest to check for *-ibm-aix* and treat it like hpux is already treated in there: it simply skips adding a command line option but then also disables threading (and the threaded resolver). It's a very crude "fix".

You are right, of course, I'm aware it's not a real fix but it allows a user to pass the -qthreaded flag to C/CPPFLAGS manually (not much of an improvement, I admit) and, most of all, it doesn't backfire and result in having profiling enabled.

1.Isn't there a chance that gcc/clang can run on AIX and have such an option?

I've found some literature about gcc working on AIX, but it's not supported by IBM and it feels like a "hack". I'd rather have XLC fully supported (apart from this little issue, curl is building fine).

2. Don't you rather want the corresponding xlc take used so that you can use the threaded resolver?

I'd love to. The main reason why I opened this issue was to (indirectly) ask whether some patch was already being implemented; if you don't mind, I'm going to test a possible solution and submit it as a PR.

Thank you for your patience and your quick reply.

[bagder]

I've found some literature about gcc working on AIX, but it's not supported by IBM and it feels like a "hack". I'd rather have XLC fully supported (apart from this little issue, curl is building fine).

I fully support that. I was only suggesting that if gcc (might) (or in fact another compiler in general) run on AIX, then checking for the OS in configure is the wrong way to detect the compiler.

I'm going to test a possible solution and submit it as a PR.

Awesome!

[asedeno]

@dcassioli can I get your help with #8541? I do not have an AIX system to test against.

[dcassioli]

@dcassioli can I get your help with #8541? I do not have an AIX system to test against.

Sure, but I won't be able to test it until Monday. I'll let you know as soon as it's done.