[RFC] Amend commit messages

timeless timeless at gmail.com
Thu Feb 24 12:39:41 CST 2011


On Thu, Feb 24, 2011 at 8:19 PM, Jason Harris <jason at jasonfharris.com> wrote:
> Shouldn't the history editing tools then be better here? Which is the real issue. The immutability of history is not really what you care about here it seems like... :)

history editing doesn't help if you only find out history isn't ok
after you push to another host because some other client pulled from
that host and complained to someone. it's already partially out in the
wild.

picture:

<manager> - <anal-bot>
|               |
<dev> - <internal-dvcs> - <publisher> - <public-dvcs>
                |
<other-dev>

steps:
1. dev pushes something to internal-dvcs.
2. other-dev pulls from internal-dvcs
3. anal-bot pulls from internal-dvcs
4. anal-bot sends a report to manager
5. manager sends a report to dev

6. dev cries

6-1. dev needs to do something to satisfy manager | anal-bot, but
<internal-dvcs> is append only.

7. publisher doesn't publish to <public-dvcs> until much later (or
never if ordered not to by manager because of anal-bot report).

6-2. If dev can fix commit message to satisfy <anal-bot>, then life is
better for dev.

7-1. if <publisher> can use |hg convert| to fold those messages later
before publishing to <public-dvcs>, then publication is more likely to
happen. the alternative is roughly an export of a single delivered
version (because of <anal-bot>/<manager>).

Note that as there's a general mapping available to <publisher> for
<internal-dvcs> and <public-dvcs>, the fact that the changeset ids
don't match doesn't really bother anyone too much. If this didn't
happen, they'd still not match, but the tree topology wouldn't match
either (single commit dumps for an entire release).


More information about the Mercurial-devel mailing list