On Wed, Jul 24, 2013 at 11:59 AM, Jörn Hees <dev@xxxxxxxxxxxx> wrote: > On 24.07.2013, at 10:52, Antoine Pelisse <apelisse@xxxxxxxxx> wrote: >> I think the best way would be to create the shared repository in >> .git/hg/$share, with $share being a path that can't be a remote name >> (so that it doesn't conflict with remote directories), > > Maybe ".git/hg/.share"? According to Documentation/git-check-ref-format.txt, I'm not sure if we should start with a dot, or end with it. >> That way, the share can be created even if .git/hg already exists >> (because of a previous import, before the shared machinery existed, or >> because you already have a local remote). > > I like the idea of having independent remotes (fetching one, doesn't update another). http://mercurial.selenic.com/wiki/ShareExtension warns about this, and i wasn't sure it wouldn't cause intricate bugs. > This is why I opted for the explicit cloning, no shared history for several remotes. I think the goal of using sharing here is that Mercurial and Git may use different schemes to handle branches. Mercurial may lead you to have separate repositories for each branch (They seem to do it for its own development [1]). All these branches actually share most of the same history, and are fully related, and we usually handle this situation in Git with one repository with multiple branches. Using "hg share", we allow a smooth transition from Mercurial model to Git model by merging all Mercurial repositories into one, and then map this single repository to the Git repository. IOW, the goal is to have only one copy of each "hg object" that are shared amongst many "remotes" (and potentially import them only once, though I don't think it currently works for me). >>> Changing gitdir to dirname causes shared_path == >>> .git/hg/<remote_name>/hg. The call to hg.share with local_path == >>> .git/hg/<remote_name>/clone works again. >> >> I think that will be a problem, because then the shared_path will no >> longer be shared, will it ? > > Yupp, the shared_paths won't be shared, so it's not as optimal as possible, but it will work at least ;) If we decided to remove the sharing idea, I think we should revert Felipe's commit rather than leave the shared_path variable, and call hg.share() on repository we don't even mean to share. That would be very confusing. [1]: http://mercurial.selenic.com/wiki/DeveloperRepos -- To unsubscribe from this list: send the line "unsubscribe git" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html