preload: fix build issue with clang 19

[clan]

librdmacm/preload.c:796:9: error: at most one overload for a given name may lack the 'overloadable' attribute

[jgunthorpe]

That does not seem correct since it broke everything, I don't think attribute overload is the right answer here

[clan]

updated.
remaining error of ci are reported by cgcc, I don't have the environment to test.

BTW: it seems CI check the first commit only? I mean, if commit B is pushed during checking commit A, CI will continue check commit A, and commit B is ignored?

[jgunthorpe]

That seems like a more believable change, but it probably needs a cmake fragment to detect what the libc is using for it's function signature.

[clan]

That seems like a more believable change, but it probably needs a cmake fragment to detect what the libc is using for it's function signature.

do you mean to detect glibc & musl? or earlier glibc version which don't have __SOCKADDR_ARG?

[jgunthorpe]

I don't know where this is coming from, but the usual way to handle it is to have cmake detect what is correct and then have the C code match it.

[jgunthorpe]

I suppose the sparse failure means the cmake test is not quite right

[clan]

2025-03-05T09:06:33.0044262Z -- Performing Test HAVE_SOCKADDR_ARG
2025-03-05T09:06:33.0869996Z -- Performing Test HAVE_SOCKADDR_ARG - Success
...
2025-03-05T09:06:43.8696098Z /usr/bin/cgcc -DVERBS_DEBUG -D__CHECK_ENDIAN__ -Drspreload_EXPORTS -I/__w/1/s/build-sparse/include -I/usr/include/libnl3 -I/usr/include/drm -std=gnu99 -Wall -Wextra -Wno-sign-compare -Wno-unused-parameter -Wmissing-prototypes -Wmissing-declarations -Wwrite-strings -Wformat=2 -Wcast-function-type -Wformat-nonliteral -Wdate-time -Wnested-externs -Wshadow -Wstrict-prototypes -Wold-style-definition -Werror -Wredundant-decls -O2 -g -DNDEBUG -fPIC -MD -MT librdmacm/CMakeFiles/rspreload.dir/preload.c.o -MF librdmacm/CMakeFiles/rspreload.dir/preload.c.o.d -o librdmacm/CMakeFiles/rspreload.dir/preload.c.o -c /__w/1/s/librdmacm/preload.c
2025-03-05T09:06:43.8697310Z /__w/1/s/librdmacm/preload.c:807:60: error: expected structure or union
2025-03-05T09:06:43.8708740Z /__w/1/s/librdmacm/preload.c: In function 'recvfrom':
2025-03-05T09:06:43.8709127Z /__w/1/s/librdmacm/preload.c:806:56: error: 'src_addr' is a pointer; did you mean to use '->'?
2025-03-05T09:06:43.8709414Z   806 |                 rrecvfrom(fd, buf, len, flags, src_addr.__sockaddr__, addrlen) :
2025-03-05T09:06:43.8709642Z       |                                                        ^
2025-03-05T09:06:43.8709831Z       |                                                        ->
2025-03-05T09:06:43.8710012Z /__w/1/s/librdmacm/preload.c:807:60: error: 'src_addr' is a pointer; did you mean to use '->'?
2025-03-05T09:06:43.8710237Z   807 |                 real.recvfrom(fd, buf, len, flags, src_addr.__sockaddr__, addrlen);
2025-03-05T09:06:43.8710558Z       |                                                            ^
2025-03-05T09:06:43.8710731Z       |                                                            ->
2025-03-05T09:06:43.8710962Z /__w/1/s/librdmacm/preload.c:798:33: error: parameter 'src_addr' set but not used [-Werror=unused-but-set-parameter]
2025-03-05T09:06:43.8711243Z   798 |                  __SOCKADDR_ARG src_addr, socklen_t *addrlen)
2025-03-05T09:06:43.8716772Z /__w/1/s/librdmacm/preload.c:812:1: error: control reaches end of non-void function [-Werror=return-type]
2025-03-05T09:06:43.8717181Z   812 | }
2025-03-05T09:06:43.8717434Z       | ^
2025-03-05T09:06:43.8717707Z cc1: all warnings being treated as errors

Where can I have the cgcc installed and test? It seems the problem is caused by incorrect preprocess result.

[clan]

    uses with any of the listed types to be allowed without complaint.
    G++ 2.7 does not support transparent unions so there we want the
    old-style declaration, too.  */
-#if defined __cplusplus || !__GNUC_PREREQ (2, 7) || !defined __USE_GNU
+#if 1
 # define __SOCKADDR_ARG                struct sockaddr *__restrict
 # define __CONST_SOCKADDR_ARG  const struct sockaddr *
 #else

find the reason finally, buildlib/sparse-include/*/sys-socket.h.diff, why?

[jgunthorpe]

I can't remember what the sparse issue is that required that change, but shouldn't cmake should still be able to detect the difference? Hmm. maybe the order of the header patching prevents that from working.
I suggest adjusting your CMake logic to also check HAVE_SPARSE and make the right choice.

[clan]

I can't remember what the sparse issue is that required that change, but shouldn't cmake should still be able to detect the difference? Hmm. maybe the order of the header patching prevents that from working. I suggest adjusting your CMake logic to also check HAVE_SPARSE and make the right choice.

My first intution is that we should do the patch first before run cmake which made the check reasonable rather than depend on some hard coded value.

update: add ${BUILD_INCLUDE} for __SOCKADDR_ARG test so test and spare run based on same source

[jgunthorpe]

that makes sense and looks right