On Sat, Oct 9, 2010 at 8:52 PM, Brent Yorgey <byor...@seas.upenn.edu> wrote:
> On Sat, Oct 09, 2010 at 03:43:42PM -0700, Jason Dagit wrote: > > > > The Haskell.org server doesn't have to be upgraded. Maintainers can > install > > a newer darcs locally (cabal install darcs), do the upgrade locally and > then > > copy the repository back to the haskell.org server. > > Really? But then the version of darcs on code.haskell.org wouldn't be > able to read those repositories, which would cause darcs over ssh to > fail -- wouldn't it? Or am I missing something? > I answered you immediate question in a different email. Now I'll do a bit of explanation on why people want hashed. I should also mention that darcs 2.0.2 (which is the version on c.h.o) does read darcs 2 repositories as support for that feature exists in all darcs 2.x clients. And I personally think that c.h.o should be upgraded to darcs 2.4.4, but that issue is orthogonal to the issue of getting maintainers to upgrade to hashed format repos. Darcs currently has 3 options for repo formats: * old fashioned * darcs 1 hashed * darcs 2 hashed The old fashioned format will eventually be removed as it's not as robust or safe as the hashed variants. The big difference between darcs 1 hashed and darcs 2 hashed is how patches behave when commuted. The darcs 2 hashed variant is a refinement that better handles some corner cases that result from conflicts. Something like 90% of the code in darcs is shared between darcs 1 hashed and darcs 2 hashed. The hashed formats are more robust because: * the hashes guard against accidental modification of patches that are stored on disk * the hashes guard against a buggy darcs client that misapplies patches (the hashes would be off, git has similar safe guards) * the hashed formats have better atomicity and hence better transactional properties If you try the hashed format and find it unpleasant for any reason it's something to report as a bug. We really want (and need) people to switch to hashed, so if it's not good enough we need to know why so we can fix it. Jason
_______________________________________________ Haskell-Cafe mailing list Haskell-Cafe@haskell.org http://www.haskell.org/mailman/listinfo/haskell-cafe
