Re: [RFC/PATCH 00/18] WIP implement cherry-pick/revert --continue

[Jonathan Nieder <jrnieder@gmail.com>]

Hi Christian,

Christian Couder wrote:

> Many patches in this series are replacing calls to "die()" by
> "return error()", because the TODO and DONE files are written
> only when cherry-pick fails. This is efficient but perhaps it
> would be simpler and safer to write them before each cherry-pick
> just in case it fails, so that the "die()" calls don't need to
> be removed.

Another possibility would be to use set_die_routine()/atexit()/
sigchain_push_common(), but the "always write" solution does seem
simpler.

> (17):

Perhaps too many. :)

I like where this is going.  My main complaint is the commit messages;
given a clear explanation of the design it should not be too hard for
others to help write documentation, enhancements, and tests, but
without it is much harder.

Nit: the style of commit message in patch 16 is unnecessarily
demoralizing.  It basically says "track down the history in this repo
that may not exist in 2050 if you want to know what this patch is
about".  I think it would be better to say

	This code was written as part of the git sequencer
	Google Summer of Code project, 2008

and let the rest of the commit message tell the important details.
Readers can google for the detailed history.

Thanks for your work.
Jonathan