I'd like @makr73's opinion on this. Not too long ago #29334 added support for AIX, and this is now taking it away again.

Hm ... I don't make any stunts to get this compiled on an AIX 7.2. I use, however, XLC 16.1 for AIX. Maybe the GCC or clang do some things differently, here. I don't have them available. I also don't have xlclang, which would be another test worth trying. If xlclang is also working without stunts I'd say: make it compiler dependent. Do we know someone who has such an environment available?

I've installed xlclang onto one of my development AIX 7.2 machines and ran this simple test:

#include  <sys/types.h>
#include  <sys/socketvar.h>
#include  <sys/socket.h>
#include  <net/proto_uipc.h>

int main() {
    struct mmsghdr msgs[1];
    
    /* This causes implicit declaration warning - no header declares sendmmsg */
    sendmmsg(0, msgs, 1, 0);
    
    return 0;
}

$ /opt/IBM/xlC/16.1.0/bin/xlclang -o sendmmsg-test.out sendmmsg-test.c
sendmmsg-test.c:10:5: warning: implicit declaration of function 'sendmmsg' is invalid in C99
      [-Wimplicit-function-declaration]
    sendmmsg(0, msgs, 1, 0);
    ^
1 warning generated.

Using xlc everything works without the warning:

$ /opt/IBM/xlC/16.1.0/bin/xlclang --version                           
IBM XL C/C++ for AIX, V16.1.0  (5725-C72, 5765-J12)
Version: 16.01.0000.0021

$ /opt/IBM/xlC/16.1.0/bin/xlc -o sendmmsg-test.out sendmmsg-test.c 
-bash-5.3$ echo $?
0

So it seems like this only works properly with xlc for some reason!

UPDATE

Actually if you enable C99 you will see that the imiplict function declaration is occuring!

$ /opt/IBM/xlC/16.1.0/bin/xlc -qlanglvl=stdc99 -o sendmmsg-test.out sendmmsg-test.c
"sendmmsg-test.c", line 10.5: 1506-1421 (W) Implicit function declaration for function "sendmmsg".

xlc just defaults to an older std than xlclang/clang. It still does the implicit function declaration under the covers!

Perhaps we should only allow enabling sendmmsg / recvmmsg when xlc is used?
Is there an xlc only define we can check?

With that update above looks like all compliers have the implict function declaration going on

@makr73

Interesting. The doc says that xlc supports all of the ISO C99 standard plus most IBM extensions. Implicitly it is also documented that xlc supports implicit function declaration. If invoked with std99 it will set a macro __C99_REQUIRE_FUNC_DECL to indicate "that implicit function declaration is not supported."
@mattcaswell , I wasn't aware that the compiler is doing the stunts for me. Given the latest findings I support the move to disable sendmmsg/recvmmsg again. Just take care that it doesn't break the test suite again.
@abmusse, thank you for doing the deep dive!

@mattcaswell Please let me know if you need to me to do anything else to help get this moving forward.

@esyr - do you agree this is trivial?

Yes, agree with CLA: trivial.

I think this can be accepted as a CLA trivial change.

@abmusse - please could you add "CLA: trivial" in a line on its own to the commit description (assuming you agree that it is trivial from a copyright perspective).

Alternatively please submit a CLA: https://openssl-library.org/policies/cla/

Yes I agree this is a trival change!

Done added the following to the original comment

CLA: trivial

Done added the following to the original comment

The bot is insisting on putting "hold: cla required". The reason is that you just edited the original comment rather than changing the actual commit message.

Please either update the actual commit message or, alternatively, whoever commits this can fix it during merge if you're not sure how to do that.

24 hours has passed since 'approval: done' was set, but as this PR has been updated in that time the label 'approval: ready to merge' is not being automatically set. Please review the updates and set the label manually.

Done added the following to the original comment

The bot is insisting on putting "hold: cla required". The reason is that you just edited the original comment rather than changing the actual commit message.

Please either update the actual commit message or, alternatively, whoever commits this can fix it during merge if you're not sure how to do that.

Can I force push the amended commit to include CLA: trivial line?

Can I force push the amended commit to include CLA: trivial line?

I guess so, github doesn't reset reviews on pushes that involve only commit message changes (as if it treats them as insignificant or something…).

Dropped branch: 3.3 label, as openssl-3.3 is EOL.

Pushed to master, openssl-4.0, openssl-3.6, openssl-3.5, and openssl-3.4. Thank you.