[svn.haxx.se] · SVN Dev · SVN Users · SVN Org · TSVN Dev · TSVN Users · Subclipse Dev · Subclipse Users · this month's index

RE: svn commit: r39153 - in trunk/subversion: libsvn_wc tests/cmdline

From: Bert Huijben <rhuijben_at_sharpsvn.net>
Date: Sat, 5 Sep 2009 21:51:09 +0200

> -----Original Message-----
> From: Hyrum K. Wright [mailto:hyrum_at_hyrumwright.org]
> Sent: zaterdag 5 september 2009 20:27
> To: svn_at_subversion.tigris.org
> Subject: svn commit: r39153 - in trunk/subversion: libsvn_wc
> tests/cmdline
> Author: hwright
> Date: Sat Sep 5 11:26:48 2009
> New Revision: 39153
> Log:
> When opening access batons, recursively lock all its subdirectories to
> an
> infinite depth as well. In wc-ng, we don't have the notion of
> recursively
> opening a database; it's either open or it's not. By making this
> change now,
> we can remove the existing complexity of determining how many levels to
> lock
> and where to lock them.

Opening a database doesn't have to be the same as locking everything
contained in it. It is completely possible to support other models with a
single SQLite database and properly designed transactions.

> We also gain another advantage. When attempting to open a baton, we
> can check
> and see if that baton is already opened, and just return it if it is.

This is not an advantage. Just returning it will make the new 'owner' close
the access baton that is really owned by some other part of our code!!!
This just introduces another kind of bugs to search for (and breaks
backwards compatibility with old code that assumed you couldn't open an
access baton twice!)

> By
> always recursively locking, we can guarantee that a found baton is also
> recursively locked, which is what the original caller may have
> intended.
> All this simplifies our current access baton usage, and the path to
> removal.

I think we should handle this always recursive locking at the caller side
(with a new function called by svn_client_*()), or just calling
svn_wc__adm_open_in_context with depth -1; not deeply buried inside the
access baton implementation.

I'm 99.9% sure the 1.6 client would crash on this new behavior, so this
would be a breaking change and would probably require switching to
developing a 2.0.

> Oh, and it incidentally fixes copy test 76.

Yes, this tested for a too shallow lock.


Received on 2009-09-05 21:51:43 CEST

This is an archived mail posted to the Subversion Dev mailing list.

This site is subject to the Apache Privacy Policy and the Apache Public Forum Archive Policy.