Hashed pristine format

(or my understanding thereof).

The file _darcs/tentative_pristine contains a pointer to the “pristine root”, in form of a hash reference, like this… (Actually, there is a reference to pristine root in _darcs/hashed_inventory as well. I currently do not understand the difference, but there surely is some.)

(Sidenote on tentative_pristine: this file seems to contain a “not-yet live” root pointer into pristine.hashed. Darcs.Repository.Hashed.revertTentativeChanges will take the pointer from _darcs/hashed_inventory and write it over _darcs/tentative_pristine. Likewise, Darcs.Repository.Hashed.finalizeTentativeChanges seems to take what’s in _darcs/tentative_hashed_inventory and _darcs/tentative_pristine and write it over _darcs/hashed_inventory (atomically))

morn@eri:~/dev/darcs/mainline -> cat _darcs/tentative_pristine 
pristine:0000001784-5853cb1b7025f159eed27dc8a2dd81f519171bc4aa85cc7ca01c44fded99315e

If you look the referenced file, you will see something like this:

morn@eri:~/dev/darcs/mainline -> zcat _darcs/pristine.hashed/0000001784-5853cb1b7025f159eed27dc8a2dd81f519171bc4aa85cc7ca01c44fded99315e | head
file:
hpc.README
0000001139-9b29de9930f2e19ab1ef4042919a34c24dfc44bc544ef6fc9c1b20a2d0b6bf32
directory:
bugs
0000000887-7b74bd1cf418f0c82f9af9f50ddfa0a9bb10a075de02403e05b058d5fb2074b0
directory:
release
0000000484-24a4464e3e1f9a56e39247cafcd9e359575bd60cfcacad6ccc766341291ceb05

Basically, this is a directory listing, with filenames accompanied by their pristine hashes. It is rather simple and somewhat git-ish (if you know git, the above object looks like a “tree” git object…). Peeking inside a “file” (git would say blob) object:

morn@eri:~/dev/darcs/mainline -> zcat _darcs/pristine.hashed/0000001139-9b29de9930f2e19ab1ef4042919a34c24dfc44bc544ef6fc9c1b20a2d0b6bf32 | head

Cleanup

When a new root pointer is written, some files under _darcs/pristine.hashed may become unreachable. Darcs.Repository.Hashed.cleanPristine takes care of that. It calls HashedIO.listHashedContents which seems (not really checked) to compute reachability on the hashed.pristine from a given root and give a flat list of reachable files. (Okey, that’s probably fairly obvious.)

See also Petr’s page “Designing Storage For Darcs”.