Re: Atomic file locking update

[Samuel Thibault]

Svante Signell, le Thu 27 Nov 2014 13:49:53 +0100, a écrit :
> As discussed on #debian-hurd IRC, these changes are conflicting with the
> record file locking patches. When you upgrade Hurd again, I need to
> rebase these patches another time.

As discussed on IRC too, AIUI the record locking patches just get rid of
the whole content that I'm patching.  So the rebase should be quite
trivial: just tell git to keep your version.

> Maybe we should keep your patches and forget about record file
> locking?

Again, as discussed on IRC, no: we do need the fine-grain record
locking.  Making tdb using coarse whole-file locking is just a
brownpaper fixup that we want to get rid of eventually.

> As written on IRC the record lock patched hurd/glibc does not play well
> with tdbtorture.

Then it needs more care. But that doesn't mean it can't be fixed.

> On the other hand, I've been running at least 5 VMs
> with these patches since August without problems.

But were you running applications which were actually using record
locking?  Few applications use locks at all, actually :)

> Of course these patches ar not up to GCS standards, so let's forget
> about the whole issue. I continue to rebase and apply them privately
> for every hurd/libc release to come.

I'm not sure to understand what you mean here.

Samuel