On the testsuite we explicitly disable the sleeps via the environment variable
designed for this (except for a few specific tests). We compensate for this
problem in quite a few places by making sure that we also change the number of
characters in the file.
Bert
> -----Original Message-----
> From: Julian Foad [mailto:julianf...@apache.org]
> Sent: donderdag 14 juni 2018 19:35
> To: Philip Martin <phi...@codematters.co.uk>
> Cc: dev@subversion.apache.org
> Subject: Re: shelf-tests failure on macOS
>
> Philip Martin wrote on 2018-06-14:
> > Philip Martin <phi...@codematters.co.uk> writes:
> > > Julian Foad <julianf...@apache.org> writes:
> > >> The most unusual thing about these failing tests is the data they
> > >> write. Strings of 5 or 6 low-value bytes, like:
> > >>
> > >> '\0\1\2\3\4\5'
> > >> '\5\4\3\2\1\0'
> > >
> > > Those are the same size. Could it be a filesystem timestamp resolution
> > > issue? The file changes but the timestamp does not and some change is
> > > not detected?
> >
> > An HFS+ filesystem has a coarse 1s timestamp resolution so that would
> > explain why the problem shows up there but not systems with subsecond
> > resolution.
>
> Sounds like the correct explanation.
>
> > Neither shelve or unshelve appear to use svn_sleep_for_timestamp().
>
> Actually shelve already does, in the relevant code path, which is inside
> svn_client_shelf_unapply() where it uses svn_client_revert4() which does.
>
> Unshelve sleeps now, since r1833532. Hopefully that's the end of it.
>
> Thank you very much for the insight, Philip!
>
> - Julian
