On Sun, Feb 16, 2025 at 10:30:17AM +0100, Samuel Thibault wrote:
> Hello,
>
> Zhaoming Luo, le dim. 16 févr. 2025 15:59:41 +0800, a ecrit:
> > In the test[0] of vim. I think the issue is that the sleep time is not
> > long enough. Sleep for 200ms is possible to pass the test, though the
> > possibility is not that high. I think sleep for 2 seconds might be a
> > good option.
>
> Indeed, file timestamps still have 1s-granularity for now.
>
> > Should we just submit a patch to Debian?
>
> As usual, better submit usptream. There is no reason for Debian to
> maintain such a patch for upstream code.
>
Indeed, this is about the backward compactability of vim.
Vim should use has('nanotime') to wrap it, just like [0].
The issue I have now is that Vim set has('nanotime') == 1 when it
finds st_mtim.tv_nse of struct stat[1][2], and indeed we have it in
glibc[3], so from the perspective of Hurd, we do support nanotime.
However, our current filesystem driver ext2fs does not support the
sub-second precision. Maybe we can propose adding a ./configure option
to manually turn off the has('nanotime')?
[0]:
https://github.com/vim/vim/blob/master/src/testdir/test_filechanged.vim#L15-L18
[1]: https://github.com/vim/vim/blob/master/src/configure.ac#L3972-L3995
[2]: https://github.com/vim/vim/blob/master/src/evalfunc.c#L7030-L7035
[3]:
https://sourceware.org/git/?p=glibc.git;a=blob;f=sysdeps/mach/hurd/bits/stat.h;h=fcaa12334845e2759c4dc13cf410f39c4442b494;hb=HEAD#l60
Best,
Zhaoming
