Darcs compared to Arch ====================== Simpler to use; perhaps harder to completely understand. Always local; always fast. Patch commution is slow and perhaps doesn't clearly do what people want. Too slow! Can't reliably get back to any previous point. Explicitly not addressing source archive/librarian function. Loads everything into memory. Written in Haskell. A really simple pre-commit check hook is remarkably useful. http://www.scannedinavian.org/DarcsWiki/DifferencesFromArch Sometimes useful to be able to set email per-branch, for people who work on different projects under different personas. Token replace ------------- Very cute; possibly handy; not absolutely necessary in most places. Somewhat limited by the requirement that it be reversible. This is one of very few cases where it does seem necessary that we store deltas, rather than tree states. But that seems to cause other problems in terms of being able to reliably sign revisions. This can perhaps be inferred by a smart 3-way merge tool. Certainly you could have it do sub-line merges. Partial commit -------------- darcs allows you to commit only some of the changes to a single file. This is like the common feature of commiting only a subset of changed files, but taken to a higher level. It is useful more often than one might think: it is common to fix some documentation 'on the wing' and while strictly it should be in a separate commit it is not always worth the hassle to back out changes, fix the docs, then do the real change. Similarly for making a separate branch. Although the idea is very good, the current darcs implementation is limited to selecting by patch hunk, which means that neighbouring changes cannot be separated. Fixing this probably means having some kind of pluggable GUI to build the file-to-be-committed or an edited patch, possibly using something like meld, emacs, or dirdiff. Another approach some people might like is editing the diff file to chop out hunks. I don't think this needs to be on by default, as it is in darcs. It is usual to commit all the changes. For this to work safely, it is good to have a commit hook that builds/tests the tree. Of course this needs to be evaluated against the tree as it will be committed (taking account of partial commits), not the working tree. Schwern ------- * http://www.scannedinavian.org/DarcsWiki/SchwernLikesDarcs * http://www.scannedinavian.org/DarcsWiki/SchwernHatesDarcs I think we hit most of these; more evidence for the hypothesis that what people really like about darcs is the simple interface not the patch-commutation model.