Karl Fogel <kfogel_at_red-bean.com> writes:
> "David Glasser" <glasser_at_davidglasser.net> writes:
>> I don't have time to think in detail right now, but couldn't the
>> opposite situation happen: where the user has the text of r1 but the
>> fake metadata of r2, and commits happily not ever seeing r2? (ie, if
>> the update is not a back-update.)
>
> Let me see if I can construct a repro for that. I believe not: that is,
> I believe that commit would be rejected. But let's make sure.
Oops, should have read the whole thread: Mark Phippard already tried
this, and the repository rejected the commit as I <pats self on back>
predicted.
In summary: commits can still only happen against HEAD. Basically, the
real lossage here is that a user might be led to believe that no changes
had happened between two revisions, when in fact changes had happened.
This is annoying, but does not rise to the level of "data corruption" in
the sense that we usually mean it, IMHO.
(There's a separate investigation needed into how the client reports the
base revision to the server, as I mentioned in my earlier mail, but
that's more an independent issue that this bug happened to unmask. I'm
looking into that now.)
-Karl
---------------------------------------------------------------------
To unsubscribe, e-mail: dev-unsubscribe_at_subversion.tigris.org
For additional commands, e-mail: dev-help_at_subversion.tigris.org
Received on 2008-06-02 21:35:37 CEST