RE: [stellation-res] Text merge test case

["Mark C. Chu-Carroll" <mcc@xxxxxxxxxxxxxx>]

On Sat, 2002-10-26 at 21:37, Jonathan Gossage wrote:
> I am finding many instances where conflicts should be generated and are not.
> This is worrisome because if a SCM is going to err it should be on the side
> of reporting spurious conflicts, rather than silently making mistakes.

Definitely send them to me. What I've seen so far look like
edge-case bugs - that is, not fundamental problems with the
algorithm, but silly errors in the implementation dealing
with cases that didn't come up in our use of the system. (I've found
three different problems triggered by off-by-ones in trailing segments
where the trailing segments were one line long. They work if the
trailing segment is longer. The
case you sent me earlier was a case where the code assumed
there'd be at least one common LCS segment. Definitely bugs,
but bugs in the form of my stupidity in not being careful
enough to make sure the code covered all possibilities.)

I believe the merge algorithm is sound. Hopefully, the
cases your seeing are either more of these edge-case
type problems, or they're differences in behavior between
merge algebra type merges and delta-transformer merges.
(There are cases where CVS would call a conflict, and
an algebraic merge would find a valid resolution.)  

Of course, if there is a fundamental error in our merge, I'd rather know
it now than later. 

	-Mark

-- 
Mark Craig Chu-Carroll,  IBM T.J. Watson Research Center  
*** The Stellation project: Advanced SCM for Collaboration
***		http://www.eclipse.org/stellation
*** Work Email: mcc@xxxxxxxxxxxxxx  ------- Personal Email: markcc@xxxxxxxxxxx