Set defines by OS name if cross-compiling, especially for Android#901
Set defines by OS name if cross-compiling, especially for Android#901ribbons wants to merge 1 commit into
Conversation
Compilation of v3.4.3 fails due to mkfifoat() being referenced but it is not available on Android until API level 23 (we are targeting 21 and above). Add patch (submitted at RsyncProject/rsync#901 for the consideration of upstream) to test specifically for the existence of mkfifoat() before trying to call it. Extend the build script and workflow to automatically apply patches in the root of the repository to the fetched source and reference the number of patches in the version number of the release.
Compilation of v3.4.3 fails due to mkfifoat() being referenced but it is not available on Android until API level 23 (we are targeting 21 and above). Add patch (submitted at RsyncProject/rsync#901 for the consideration of upstream) to test specifically for the existence of mkfifoat() before trying to call it. Extend the build script and workflow to automatically apply patches in the root of the repository to the fetched source and reference the number of patches in the version number of the release.
|
@ribbons I'm planning to have android builds as part of the main rsync repo, both normal builds and specific termux deb pkgs. Are you interested in helping with that? I'd like android to be tested in CI to ensure it doesn't bitrot |
|
@tridge That sounds great! Sure thing - happy to help (though with the caveat that I've not used Termux in anger before)! If you're happy to add another third-party action, I've had pretty good results using On the topic of this PR, do you think this is the right approach or is it getting the correct result the wrong way? I was a bit on the fence before but seeing your workaround in 8839314 made me think it might be better to adjust configure.ac to automatically set HAVE_SOCKETPAIR, MKNOD_CREATES_FIFOS and MKNOD_CREATES_SOCKETS when cross-compiling and the target is android ? That would then both fix this specific issue and give the correct feature detection for anyone downstream performing builds. |
|
@ribbons I'm leaning towards the auto-set of options, but I'd be happy to declare you as being the rsync on android maintainer and let you decide on the approach. The aim would be that rsync.samba.org downloads has usable android builds for the various architectures and environments that android uses, maybe in an android subdirectory with a readme explaining when to use which one and how to install. |
ribbons
force-pushed
the
have-mkfifoat
branch
2 times, most recently
from
9e6e418 to
8a10973
Compare
|
@tridge great, thanks - I've switched this PR to that approach instead and so |
ribbons
changed the title
The following defines set by autoconf tests always default to unset if cross-compiling: - HAVE_C99_VSNPRINTF - HAVE_SECURE_MKSTEMP - HAVE_SOCKETPAIR - MKNOD_CREATES_FIFOS - MKNOD_CREATES_SOCKETS To avoid all builds for Android needing to override the same cached values, add some extra logic to set HAVE_C99_VSNPRINTF and HAVE_SECURE_MKSTEMP when cross-compiling and OS name matches *android*. Go even further with HAVE_SOCKETPAIR, MKNOD_CREATES_FIFOS and MKNOD_CREATES_SOCKETS - as it looks like the underlying tests should pass for any Linux based system, enable them when cross-compiling and OS name matches *linux*.
2 participants
If
MKNOD_CREATES_FIFOSis not defined by configure (such as when cross-compiling),mkfifo()is available butmkfifoat()is not then compilation fails since 30656c5.(I discovered this when trying to cross-compile v3.4.3 for Android, targeting API level 21 which provides
mkfifo()but notmkfifoat()).This PR adds a configure function check for
mkfifoatand updates the preprocessor conditional that wraps the call to it in syscall.c to referenceHAVE_MKFIFOAT.