So, things are moving around somewhat. This is the current status of the
1.3 branch as far as I know. Please read through and confirm that I'm
not missing anything - Oh, and that nobody's forgotten about any of this
stuff :-)
* Ruby bindings segfault. Has this been put to bed? I see the
STATUS has a couple of approved changes that I'll merge once I'm
done with this mail. Is there any work or polishing that still
needs doing, or are we all set? Ah, David James tells me that fix is
satisfactory as far as 1.3.0 is concerned.
* Crashes and deadlocks in the python bindings. This seems to be
rapidly turning into a can of worms, as mod_python is revealing
some problems that date back to a long time with our bindings.
Two cases in point. The first, is that calling apr_terminate
crashes on windows, because it is called using atexit(), ie. at a
point where the python interpreter has shut down and can no longer
process callbacks that get invoked. We are working on a fix to
this, which we'll be getting out for testing shortly.
The second has to do with multithreading. The current code is not
careful enough with memory management and locking, which crashes
multithreaded programs calling our bindings. Both ViewCVS and
Trac, if running over mod_python (one of ViewCVS' 2 operation
modes, and one of Trac's 3 with FastCGI), are using our bindings in
a multithreaded fashion, and are at risk.
We should note a couple of things: first, this is not a new issues.
There were also memory leaks and crashes in the 1.2 bindings. The
only difference is that the bindings are now doing a lot more
things behind the scenes, which make the problems more likely to
occur. Secondly, it seems that the only reports of crashes during
execution we have come from mod_python on win32.
David James is in the process of summarizing what needs to be done
in another mail. We need people to read that and help bang out a
fix, as it seems to be the only major blocker to rolling the RC.
I'll be setting up a mod_python test server for the devs tonight,
so that people can try all this out properly.
* #1911 still needs the third vote. I'll be looking into it
(honest), but if someone who already knows svndumpfilter logic can
take a look, it might get approved faster.
That's all I can see offhand. I've spent some time discussing the
python issues with David James and people on IRC, and am now going to
run off to set up a user mode linux I can unleash onto the devs, for
testing.
Oh, I almost forgot to renew my prod to Ben Reser. Ben, could you make
sure you have SWIG 1.3.25 ready for when we're ready to build RC1 ?
Thanks all.
- Dave.
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Tue Oct 18 19:36:36 2005