Philip Martin <philip@codematters.co.uk> writes:
> Should this detection of first command failure be used differently?
> There are two possibilities:
>
> 1. During normal operation, the first command failure could cause the
> log file to be removed. Then the working copy would not remain
> locked and there would be no need to run cleanup at all. Of course
> this would mean that the log file would not be available for
> analysis.
Yeah, that last sentence is the only fly in an otherwise attractive
ointment...
> 2. How robust is this first command failure? Is it possible for a log
> file to run part way though, and then fail in such a way that
> subsequent attempts to run it will fail on the first command?
Urgh, I hadn't thought of that.
Yes, I think that *is* possible, which means r7592 isn't safe...
> It might be more robust if the first command failure were
> detected during normal operation and, rather than removing the
> log file, some positive indication of first command failure could
> be written into the working copy. Cleanup could then use this
> positive indication of first command failure when deciding to
> remove the log file.
... but this proposal would make it safe (and is the only way I can
see to make it safe, off the top of my head). I'll finish the rest of
issue #1581, then either implement your above proposal or revert
r7592, as time permits.
Thanks, Philip.
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Mon Nov 3 04:05:51 2003