Re: CVS update: subversion/subversion/libsvn_fs fs.c

> This error cleanup/recovery isn't quite right. It gets me past the
> txn_checkpoint() failure, though.

Thanks for catching this. I think I've got the logic right in my
sources now; I'd commit it immediately, but I've got other incomplete
changes in fs.c.

If allocate_env returns, we always need to call env->close, even when
the call to env->open fails. So the jumps to error_env aren't right,
I think.

> Note that I zeroed out fs->env in the cleanup. I put that *after* the
> checkpoint, right before the close. It might need to go before the
> checkpoint call, but I'm not intimately familar with the DB semantics to
> determine where it made the most sense.

Checkpointing is just an ordinary operation that you could perform at
any time on a healthy env. We shouldn't zero fs->env until we've
called fs->env->close, so I think your change is correct.
Received on Sat Oct 21 14:36:14 2006