Tip (okay, loooong tip): forcing changes to contain signoffs

[Bryan O'Sullivan]

On Tue, 2006-04-11 at 14:14 -0700, Danek Duvall wrote:

> Is it not possible to break the changeset, make the required changes, and
> re-create the changeset again before resubmission?

It is quite easy to do this using the Mercurial Queues (mq) extension.

> Or is it simply that
> this is a manual step that not everyone might want to go through?

I think that was more Daniel's point.

> It just seems like you'd want the signoff to be more tightly bound to a
> changeset than having changeset pairs, if you will.

That's really a matter of how you like to structure your work, not
something that's inherently good or bad per se.

For example, Linux kernel changes have to contain a "Signed-off-by:"
line in the comments of the actual changeset, because Linus decreed that
It Must Be So, but it's equally valid to decouple the process of
creating a change from the review and signoff.

> It seems to me that a reasonable workflow for us in ON is something like:
> 
>   - engineer completes development work, testing, paperwork, etc.
>   - files an RTI -- Request to Integrate, which is essentially a request
>     for a signoff
>   - advocate (the signer-off) attaches signature to engineer's changes
>   - engineer creates changeset with embedded signature
>   - engineer pushes changeset to repository which checks for valid RTI

You can do this, but you'll *have* to use mq (or something very similar)
in that case, because changesets are (as you've observed) normally
immutable.

> I hope that my questions get more intelligent as time goes by, but
> please point me at the appropriate documentation if it appears that I've
> forgotten it.

Welcome aboard :-)

	<b