Re: svn commit: r1619068 - in /subversion/trunk/subversion: include/ include/private/ libsvn_fs/ libsvn_fs_base/bdb/ libsvn_fs_fs/ libsvn_fs_x/ libsvn_ra_svn/ libsvn_subr/ svnserve/ tests/ tests/libsvn_fs_fs/

On Wed, Aug 20, 2014 at 4:16 PM, Evgeny Kotkov <evgeny.kotkov_at_visualsvn.com>
wrote:

> > Author: stefan2
> > Date: Wed Aug 20 10:30:26 2014
> > New Revision: 1619068
> >
> > URL: http://svn.apache.org/r1619068
> > Log:
> > Remove the "checked" option from svn_mutex_t. Update all callers
> > and remove the regression test.
> >
> > The reason is that there seems to be no portable way (e.g. via a
> > series of conversions) to use APR atomic CAS operations with
> > apr_os_thread_t.
> >
> > * subversion/include/private/svn_mutex.h
> > (svn_mutex__init): Remove CHECKED option.
> >
> > * subversion/include/svn_error_codes.h
> > (SVN_ERR_RECURSIVE_LOCK): Remove as it has no users anymore.
>
> This is not entirely true, as we now also expect in our freeze_freeze() and
> freeze_same_uuid() tests. With this change, both of them will deadlock
> while
> running 'make check SERVER_MINOR_VERSION=8'.
>

Seems that Brane fixed that last week and you updated the commentary.
Thanks!

I also think that if we remove these checks, we should also revert a
> follow-up
> to it (http://svn.apache.org/r1593015). We did enable the shared data
> mutexes
> for all platforms in order for the test suite to pass, but this is no
> longer
> necessary, and I do not see any reasons not to restore the 1.8 behavior
> here.
>

Yes, we should revert and I did in r1620930.
Thanks, Evgeny.

-- Stefan^2.
Received on 2014-08-27 18:42:16 CEST