On Tue, 12 Apr 2005, Greg Hudson wrote:
> The FSFS locking fix comes in three logical parts:
>
> (1) Cache the uuid in memory (in FSFS) for technical reasons.
> (2) Add the serialized_init machinery to the FS loader library, and
> use it in the FSFS back end to serialize access to the write
> lock.
> (3) Add the svn_fs_initialize() API and calls to it. If it is not
> used, libsvn_fs makes a best-effort fallback which (a) leaks a
> little heap memory if the svn libraries are unloaded, and (b)
> could, if one is very unlucky, fail to serialize accesses to the
> write lock.
>
> The third part is proving troublesome; client programs like the "svn"
> command-line client need to invoke it in case they use ra_local, but
> they may not be linked against libsvn_fs if --enable-dso is used. The
> solution will involve adding an svn_ra_initialize() for client
> programs to use, and that will take some time and a little design
> work.
>
Locking at your backport to 1.2, I see you removed the svn_fs_initialize
API. I don't understand why. ONly because our own *client* can't make use
of it doesn't hinder our servers from using it. And other users wanting to
be thread safe could use it.
I think it should go in onto the backport branch together with Karl's
r14156.
Regards,
//Peter
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Wed Apr 13 21:55:32 2005