Re: [Monotone-devel] Re: bidirectional CVS gateway

[Christof Petig]

graydon hoare schrieb:
> Christof Petig wrote:
>
> > IIRC this monotone cert (or file) needs to store every CVS' file's 
> > revision for a MT revision. Perhaps I should start with tagged 
> > revisions and try to match them. [Does monotone support moving a tag 
> > to a different revision? IIRC no. This looks like to be the missing 
> > stone for a full synchronization. (synchronizing trees might still work)]
>
>
> this sounds like the beginnings of a plan for how to do it. perhaps we 
> could store a "CVS" attribute in .mt-attrs which notes which CVS version 
> number each file is at. this would at least make the CVS->monotone job 
> easier, since you could compare those numbers against a CVS working copy.

I still want to stay away from storing synchronization information in 
tree if possible.

This way there's no internal state which might get outdated (or 
misleading). If I find that synchronization speeds up by a factor of two 
or above I might add this caching later.

> it seems to me that the monotone->CVS direction is the hard one, because 
> monotone may literally have a history which is inexpressible in CVS 
> terms. in that case you need to decide which part of monotone's history 
> to elide (or decide to support linear monotone branches). also there is 
> the problem of deciding who "wins" when there's divergence: if CVS 
> diverges from monotone, monotone will be happy to just record that as a 
> fork. if monotone diverges from CVS, CVS will want monotone to merge 
> with its head before it lets us commit.

In any case a pull should precede a push to a CVS server. (perhaps pull 
even omits the second step)

> hmm. I wonder. can our "branch linearizing process" just be the process 
> of reconciling with a CVS server? that'd kill two birds with one stone, 
> and relieve us of ever having to write code to manage the locking and 
> unlocking of our "trunk".. then all the centralization and concurrency 
> problems are just "CVS' problem" (or SVN's problem).

IMHO choosing a head for every branch (or choosing which branch is the 
main an which is a side fork) is a prerequisite to CVS push. You have to 
linearize before you can reconcile.

   Christof

signature.asc
Description: OpenPGP digital signature