Re: Is it time to drop ChangeLogs?

[Phillip Lord]

Paul Eggert <address@hidden> writes:

> On 07/06/2016 06:06 PM, Ted Zlatanov wrote:
>> using a pull request system would reduce the
>> need for writing every commit message as if it was going into a
>> ChangeLog
> I don't see why; if we're going to keep the current format, pull requests
> should use the format too.
>
> I have the sense that pull requests work better in projects with a large
> number of occasional committers and a small number of full-time developers who
> triage and review. Emacs development doesn't work that way: among other
> things, there are no full-time developers, and we don't have enough reviewer
> time. So it may not be a good fit for the pull-request model. (It might make
> sense to change Emacs's development model but that's a larger topic....)

I think that this is not true. I have recently sent patches into, for
example, ensime which is a fairly small development. Also, Cask uses an
always PR model which means that all changes get looked at by at least
two people.

Where PRs work better over direct commits is when ever someone wants
comments and feedback. There are two main reasons for this. One is where
commits need to be checked by someone else before going in; the emacs-25
branch is in this state at the moment.

The other is where someone wants feedback on their work, because they
are new, or because they are fiddling with parts of Emacs which they
understand poorly. I've found this extremely useful, for instance, with
my changes to the undo system, which involved extensive communications
mostly with Stefan. With Cask, also, I have found this very useful,
since I've had to make changes to in three different languages.

At the moment, we have a poor workflow for supporting this.

 - I can push a branch onto the Emacs git. But, this is not squashable,
   so the final state before the merge is hard to do
 - There is no system for queuing pull requests, so sometimes things get
   forgotten
 - I can send patches, but this is clunk compared to pushing a branch
   within version control.
 - There is no system for viewing feedback about an individual patch.
 - There is no system for adding inline comments to patches

Installing something like gerrit or kallithea would be nice (I have no
direct experience of using either, but they are similar to other
systems). However, this would be considerable work.

Perhaps, as a half way house, we could use the resources that we have.
PRs could go to the bug reporting system. This will, at least, keep all
the conversations in one place. If we can tag these with "has patch"
here as well, it will give an queue also. We would still need to do
something about the Emacs git, in terms of squashability; in practice,
this would probably require something like gitolite as allowing non-FF
pushes on all branches would be a bad thing.

This would not give a nice web interface, nor inline comments, but it's
a start.

Phil