[jdk17u-dev] RFR: 8316304: (fs) Add support for BasicFileAttributes.creationTime() for Linux

[Thomas Stuefe]

On Thu, 25 Jan 2024 15:05:49 GMT, Severin Gehwolf <sgehwolf at openjdk.org> wrote:

> This is an unclean backport of the filesystem enhancement on Linux to use `statx` where available. The JDK 21u patch didn't apply cleanly because JDK 20+ has [JDK-8283335](https://bugs.openjdk.org/browse/JDK-8283335) which isn't needed for this backport. The relevant code has been adjusted to work without that enhancement being present (see different commits). Copyright dates needed manual adjusting and the test needed some trivial changes related to test lib and import so that it works (failed to compile on 17u).
> 
> As for the code changes in `UnixNativeDispatcher.c`, the JNI method `Java_sun_nio_fs_UnixNativeDispatcher_stat1()` returning the mode for a file/directory doesn't exist in JDK 21 (again [JDK-8283335](https://bugs.openjdk.org/browse/JDK-8283335)). Therefore, I've adjusted the code to also use `statx` where available for symmetry.
> 
> I'll follow up with the related build fix (for the test libs) in a dependent PR (JDK-8319922).
> 
> Please review!
> 
> Testing:
> - [ ] GHA (still running).
> - [ ] Build test on various Linux (aarch64, Alpine x64, ppc64le, x64) and Unix OSes (AIX, Mac) so as to not break the build there.  On-going.
> - [x] Manual testing for `BasicFileAttributes.creationTime()`. Looks good.

One thing I wonder is whether there could be cases where statx may fail but stat64 may succeed; in that case, we would have a regression now. Looking at the manpages, it seems that the set of errno values is the same for both, so it may be a nonissue. If we wanted to be super careful though, we could always retry stat64 if statx fails. Would also shorten the functions a bit.

Up to you, especially since that would introduce changes in the downport release.

I'm reviewing now...

-------------

PR Review: https://git.openjdk.org/jdk17u-dev/pull/2175#pullrequestreview-1844016995