On Fri, Sep 09, 2005 at 12:50:51PM +0100, Bruce Stephens wrote: > Technically I imagine that would be possible, if files have a > permanent identity as proposed by Nathaniel recently, it might be > possible to expose that sufficiently to attach certs to them. File > attrs could be done like that, I guess? On the other hand, you'd > probably want them to be versionable, so that wouldn't work well.
The current plan is for files to get a permanent identity in the implementation, but not to expose this at all -- the permanent identities aren't transferred or even guaranteed to be stable (just globally consistent -- so you could regenerate identities for the whole tree at once, say). And attributes will be "done like that", inasmuch as attribute changes will be first class and attached to logical files; internally, they'll be indexed by the permanent identities just like content hashes. In principle, one could add certs on (revision, filename) pairs, but it seems a bit brittle (if your use case wants to cert a revision and a _set_ of files, you're right back where we started), and when you define a new cert type, you get to pick the format used for it -- and it's trivial for this format to include a filename or several filenames or whatever you like. -- Nathaniel -- "Of course, the entire effort is to put oneself Outside the ordinary range Of what are called statistics." -- Stephan Spender _______________________________________________ Monotone-devel mailing list [email protected] http://lists.nongnu.org/mailman/listinfo/monotone-devel
