Hacker News Reader

I've just finished the introduction part and there is one thing he got wrong. the changesets vs revisions thing. Changesets and revisions are completely equivalent as far as merging is concerned (and almost anything else) because one is trivially computed from the other. and git which has even better merging support tracks file contents not diffs, which is another proof that this argument is irrelevant.

View on HN

If you speak of Mercurial's changesets and revisions, they are clearly not trivially computed from one another, being that a revision is clone-specific, and a changeset is a hash computed on a single file by looking at all it's ancestors' states. REPO1 and REPO2 can both have 5 revisions and only one is common to both — a merge is possible, but the changesets considered will be very different.

If you're comparing SVN's revisions with Hg's changesets, still no dice, for the same reason: changesets are unique for a file, given that file's history, and the hash trail allows for a clear way to figure out where changes happened, and merge from that point. A SVN revision just tells you how many commits have happened up to that point.

Reply on news.ycombinator.com