"Eric S. Raymond" <esr@snark.thyrsus.com> writes:
> For anybody on the GPSD mailing list who missed it, the last hour of
> traffic is among other things a demonstration that moving gpsd to a
> Subversion repository was a Good Thing.
>
> When Rob's report of total breakage came in, we were at rev 3027.
> The way I zeroed in on the bug was to svn -r up 2900, build, test
> on a live GPS, and verify that it worked. Thereafter I binary-searched
> the revision sequence looking for the commit that broke communication
> with the GPS.
>
> Because of Subversion's unitary commits on the whole tree, there was
> only one revision sequence for the whole tree. Under RCS or CVS I
> could only have searched this way if I'd had the foresight to tag
> every single group of commits. But with Subversion I found the bad
> commit in less than ten minutes and had it backed out in another
> thirty seconds.
>
> OK, it helps that the bad commit was late last night rather than three
> weeks ago. But the search cost would only have risen as the log2 of
> time. This is a very efficient way to pin down breakage.
>
> Thank you, Subversion maintainers. Take a bow; you've earned it.
<developers bow in unison>
Thanks! :-)
Really I guess Jim Blandy should take that bow, since he designed the
Subversion repository model, including, among other things, atomic
commits with monotonically increasing revision numbers.
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe@subversion.tigris.org
For additional commands, e-mail: dev-help@subversion.tigris.org
Received on Thu Jul 28 20:41:40 2005