PRCS tries to match files first by their file descriptor (entity identifier) and then by their file name. If the two matches agree, or if there is only one match, then the result is used. If the result is ambiguous PRCS prints out a message telling the user that they have probably done something pathological, that they shouldn't have, and then allows them to choose which match to observe.
In three passes, PRCS selects a merge tuple as follows:
The three passes insure that all 14 possibilities:
http://www.cs.berkeley.edu/~jmacd/merge.html
are correctly processed. I realize that this won't apply for you directly because you aren't picking a common ancestor project version.
-josh
Quoting Jonathan S. Shapiro (shap@eros-os.org):
> On 12 June, I posed a question about computing merge triples. You can find
> the original note at
>
> http://www.eros-os.org/~majordomo/dcms-dev/0204.html
>
> I'ld really appreciate responses to this. It's likely that in the next week
> or two I will have time to implement merge, and reactions to the questions
> posed in that note would be extremely helpful.
>
> I'ld especially appreciate reactions from Josh, John Randolph, and Bill
> Frantz (and everyone else, of course, but my sense is that these three have
> had occasion to wrestle with this stuff before).
>
> I really want to get merge done so that I can do a 0.1 release.
>
> Thanks
>
>
> shap