[CMake] Define _FILE_OFFSET_BITS=64 on Solaris

[ro]

Summary

This is the compantion patch to https://reviews.llvm.org/D64482, needed to ensure
that builds with host compilers that don't yet predefine _FILE_OFFSET_BITS=64 on
Solaris succeed by always making the host and freshly built clang consistent.

Tested on x86_64-pc-solaris2.11. Ok for trunk?

[ro]

Ping? It's been a week and together with it's clang companion this patch is needed to unbreak make check-all with gcc 9. So it would
be good to get this into LLVM 9.

[rnk]

I would place this macro setting with the other code that sets this macro, in cmake/modules/HandleLLVMOptions.cmake.

[ro]

In D64483#1590120, @rnk wrote:

I would place this macro setting with the other code that sets this macro, in cmake/modules/HandleLLVMOptions.cmake.

This doesn't work unfortunately: I still get the same link failure for Sanitizer-i386-Test with gcc 9. For the original patch I'd followed how things are done
for AIX (_XOPEN_SOURCE=700, _LARGE_FILE_API) which is similar.

[ro]

How should we procede here, given the proposed alternative doesn't work? It would be good to get this and the already approved
companion patch into trunk and, if possible, into the 9.x branch.

[ro]

Ping^2? It's been two weeks now and I'd like to know how to proceed with this patch.

[hans]

Seems fine I guess, since this matches what's already there for AIX.

[ro]

In D64483#1605840, @hans wrote:

Seems fine I guess, since this matches what's already there for AIX.

Thanks. Would this and its companion D64482 be appropriate for the 9.x branch, too? If so, what's the procedure to get it there?

[hans]

Thanks. Would this and its companion D64482 be appropriate for the 9.x branch, too? If so, what's the procedure to get it there?

Probably. They need to land on trunk first, then let me know and I'll merge them.

[MaskRay]

llvm/trunk/CMakeLists.txt

837 ↗

(On Diff #212309)

Doesn't ${CMAKE_SYSTEM_NAME} MATCHES "SunOS" imply UNIX?

[ro]

llvm/trunk/CMakeLists.txt

837 ↗

(On Diff #212309)

I guess so: just copied it from the AIX clause above.

[ro]

In D64483#1605895, @hans wrote:

Thanks. Would this and its companion D64482 be appropriate for the 9.x branch, too? If so, what's the procedure to get it there?

Probably. They need to land on trunk first, then let me know and I'll merge them.

Both patches are in now. However, there's a buildbot failure
that claims to be related, though I cannot see yet how this can possibly happen.

[ro]

In D64483#1605989, @ro wrote:

In D64483#1605895, @hans wrote:

Thanks. Would this and its companion D64482 be appropriate for the 9.x branch, too? If so, what's the procedure to get it there?

Probably. They need to land on trunk first, then let me know and I'll merge them.

Both patches are in now. However, there's a buildbot failure
that claims to be related, though I cannot see yet how this can possibly happen.

The next builds of both affected buildbots [[http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast] and http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux
are fine again, so this seem to have been transient failures unrelated to my patches.

From the POV, they seem good to go for a backport.

[rnk]

In D64483#1591428, @ro wrote:

In D64483#1590120, @rnk wrote:

I would place this macro setting with the other code that sets this macro, in cmake/modules/HandleLLVMOptions.cmake.

This doesn't work unfortunately: I still get the same link failure for Sanitizer-i386-Test with gcc 9. For the original patch I'd followed how things are done
for AIX (_XOPEN_SOURCE=700, _LARGE_FILE_API) which is similar.

I don't understand how that could be, but I can't investigate since I don't have solaris set up. HandleLLVMOptions.cmake is included from llvm/CMakeLists.txt right above the code you are modifying, and it calls add_definitions as well, so it should work exactly the same.

[hans]

In D64483#1606208, @ro wrote:

In D64483#1605989, @ro wrote:

In D64483#1605895, @hans wrote:

Thanks. Would this and its companion D64482 be appropriate for the 9.x branch, too? If so, what's the procedure to get it there?

Probably. They need to land on trunk first, then let me know and I'll merge them.

Both patches are in now. However, there's a buildbot failure
that claims to be related, though I cannot see yet how this can possibly happen.

The next builds of both affected buildbots [[http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux-fast] and http://lab.llvm.org:8011/builders/sanitizer-x86_64-linux
are fine again, so this seem to have been transient failures unrelated to my patches.

From the POV, they seem good to go for a backport.

Yes, I've merged both in r367526 and r367527, respectively.

[ro]

In D64483#1609776, @hans wrote:

In D64483#1606208, @ro wrote:

From the POV, they seem good to go for a backport.

Yes, I've merged both in r367526 and r367527, respectively.

Excellent, thanks.

CMakeLists.txt

 
 # Build with _XOPEN_SOURCE on AIX, as stray macros in _ALL_SOURCE mode tend to
 # break things. In this case we need to enable the large-file API as well.
 if (UNIX AND ${CMAKE_SYSTEM_NAME} MATCHES "AIX")
           add_definitions("-D_XOPEN_SOURCE=700")
           add_definitions("-D_LARGE_FILE_API")
 endif()
 
+# Build with _FILE_OFFSET_BITS=64 on Solaris to match g++ >= 9.
+if (UNIX AND ${CMAKE_SYSTEM_NAME} MATCHES "SunOS")
+          add_definitions("-D_FILE_OFFSET_BITS=64")
+endif()
+
 # Work around a broken bfd ld behavior. When linking a binary with a
 # foo.so library, it will try to find any library that foo.so uses and
 # check its symbols. This is wasteful (the check was done when foo.so
 # was created) and can fail since it is not the dynamic linker and
 # doesn't know how to handle search paths correctly.
 if (UNIX AND NOT APPLE AND NOT ${CMAKE_SYSTEM_NAME} MATCHES "SunOS|AIX")
   set(CMAKE_EXE_LINKER_FLAGS
       "${CMAKE_EXE_LINKER_FLAGS} -Wl,-allow-shlib-undefined")

cmake/config-ix.cmake

 
 # Do checks with _XOPEN_SOURCE and large-file API on AIX, because we will build
 # with those too.
 if (UNIX AND ${CMAKE_SYSTEM_NAME} MATCHES "AIX")
           list(APPEND CMAKE_REQUIRED_DEFINITIONS "-D_XOPEN_SOURCE=700")
           list(APPEND CMAKE_REQUIRED_DEFINITIONS "-D_LARGE_FILE_API")
 endif()
 
+# Do checks with _FILE_OFFSET_BITS=64 on Solaris, because we will build
+# with those too.
+if (UNIX AND ${CMAKE_SYSTEM_NAME} MATCHES "SunOS")
+          list(APPEND CMAKE_REQUIRED_DEFINITIONS "-D_FILE_OFFSET_BITS=64")
+endif()
+
 # include checks
 check_include_file(dlfcn.h HAVE_DLFCN_H)
 check_include_file(errno.h HAVE_ERRNO_H)
 check_include_file(fcntl.h HAVE_FCNTL_H)
 check_include_file(link.h HAVE_LINK_H)
 check_include_file(malloc/malloc.h HAVE_MALLOC_MALLOC_H)
 if( NOT PURE_WINDOWS )
   check_include_file(pthread.h HAVE_PTHREAD_H)