Re: ELPA commit freeze

[Dmitry Gutov <dgutov@yandex.ru>]

On 20.08.2013 08:16, Stefan Monnier wrote:
>> The packages/js2-mode subtree history is wrong,
>
> That's for sure: it pretends that all changes until the conversion to
> Git happened directly in `elpa'.

Aside from that, git log -- packages/js2-mode doesn't show the commits 
from the upstream repo at all, only the merge commit. Can you explain that?
I remember that Bzr somehow omits histories of the merged branches (by 
default?), but 'git log' usually shows all commits that ended up in the 
current branch history.

>> so the wrong history got pushed.
>
> I'd expect this history to be "the true js2-mode history plus
> a redundant copy of that history from the packages/js-2mode directory of
> the Bzr branch".  Is that the case, or do you get "the true js-2mode
> history plus the complete history of the whole elpa branch".

Something along the lines of the latter. Here's what the push contained:

https://github.com/mooz/js2-mode/compare/2c27e3eb847c...05424b8245b6 
(the view is limited to 250 commits, but there were obviously more of 
them). 2c27e3eb847c is the current upstream HEAD.

And here's the report from the automatic build on Travis-CI: 
https://travis-ci.org/mooz/js2-mode/builds/10312341 After checking out 
the new version, it couldn't find .travis.yml, because the tree 
contained the whole elpa:

https://github.com/mooz/js2-mode/tree/05424b8245b62e95e2b376cbc63ed182cb1c8bee

> In the case of a "duplicate history", I think it's about as good as it's
> going to get, because of the fact that the "subtree merge" was only made
> at the end.

Do you think it's too late to rewrite history? Apparently, git 
filter-branch allows to remove a directory from history.

So we could check out the version before externals were introduces, 
erase all their respective directories, then apply all commits made to 
the "administrative" part of the tree, and then add the subtrees 
properly. Some cleanup commits would have to be re-applied, but there's 
not a lot of them.

If it's too late, I suppose we can live with that, but this way we a) 
give up an easy way to sync back, b) accept that we'll see each 
non-upstream commit in externally maintained packages's histories twice: 
once for when it's made, second after it's cherry-picked, committed to 
upstream and then merged back into elpa.

 > To get something better, we'd have to reconstruct the elpa
 > branch bit by bit, performing the merges as if they'd been done right
 > back when we added/updated all those externally maintained branches.

I'm not sure which merges you mean, but the way I described we'd give up 
older changes from before conversion to Git, that still aren't merged 
into upstreams now.

>> You did. See 293db6e (Fix up copyrights and the checking code).
>
> I think this was before I fixed the email-massaging scripts to adjust to
> the new format.  I'll install one of my pending minor cleanups to the
> js2-mode, so we can make sure it works.

Ok.