[openmpi 5.0] `error: 'OPAL_RECURSIVE_MUTEX_STATIC_INIT' undeclared here (not in a function); did you mean 'OPAL_MUTEX_STATIC_INIT'?`

[barracuda156]

Something is not right here:

  CC       instance/instance.lo
/opt/local/var/macports/build/_opt_PPCSnowLeopardPorts_science_openmpi/openmpi-gcc13/work/openmpi-5.0.0/ompi/instance/instance.c:62:47: error: 'OPAL_RECURSIVE_MUTEX_STATIC_INIT' undeclared here (not in a function); did you mean 'OPAL_MUTEX_STATIC_INIT'?
   62 | static opal_recursive_mutex_t instance_lock = OPAL_RECURSIVE_MUTEX_STATIC_INIT;
      |                                               ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
      |                                               OPAL_MUTEX_STATIC_INIT
make[2]: *** [instance/instance.lo] Error 1

@shintaro-iwasaki Could this be related to bcd376a commit?

[jsquyres]

@barracuda156 Could you please provide the information asked for in the Github Issue template?

https://github.com/open-mpi/ompi/blob/main/.github/ISSUE_TEMPLATE/bug_report.md

[eschnett]

I see the same error (this may be unrelated to the original poster) when building on platforms that use musl as their C library instead of the GNU libc. The underlying macro you are using to initialize a recursive mutex does not exist there because it is a GNU extension and not part of the POSIX standard.

I encounter this when building OpenMPI 5 for Yggdrasil to make it available for the Julia language; see here. This patch corrects the problem for me.

[jsquyres]

@eschnett Thanks for the additional context. @bosilca has posted a PR for this issue: #12034

[barracuda156]

@jsquyres Sorry, I got overwhelmed with issues to sort :) I guess, this is already understood as an actual bug and getting addressed?

P. S. When I looked through the source code with BBEdit search, I could not find what defines that macro. Now we know why.

[jsquyres]

@barracuda156 No, we didn't recognize this as an issue until it was filed and @eschnett filed the additional context (I don't think we have core community members who test on non-glibc / non-macOS platforms, and so it wasn't a problem for us). But @bosilca filed a PR that should fix it.