[cvsnt] Mergepoint issues on 220.127.116.11 b2382
tony.hoyle at march-hare.com
Mon Jan 15 12:56:01 GMT 2007
Gerhard Fiedler wrote:
> I wish Tony H or Arthur would explain better the situation(s) they have
> seen that lead to data loss. IIRC, so far it was only a "trust me on that"
> or something the like... I think it would be extremely helpful if the
> situations that lead to data loss were described somewhere in a way that
> allows them to be reproduced.
It was a bug report from a cvsnt user that initially alerted me to the
problem. There were a couple following that - but it was a while ago.
Basically what I believe happened was they had two independent branches
of work. Merged A->B, taking only a few changes and ignoring the rest,
did some changes to B then merged B->A
What they *wanted* to happen was that only the changes they made in B
got merged back to A. What actually happened was as seen in the other
threads - it took the original mergepoint on A as a common ancestor and
wiped out the differences between A and B.
Yes they could revert it (so it wasn't true data loss) but it's not
intuitively obvious that it should do that (probably obvious once you've
read this thread but definately not to a new user).
As far as they were concerned things that they wanted to keep had been
deleted. It's also possible that some of the 'i committed xxx but it's
gone' messages that used happen occasionally may be attributed to this,
but can't prove either way.
btw. evs now has an option to enable this behaviour (not on by default).
Not decided how strong to make the warning though.. whether to say
'use this and you might lose stuff' or simply document that it only
works correctly in certain usage scenarios.
More information about the cvsnt