[PATCH 2 of 3] commit: add a -M/--reuse-message option to copy a commit message from a

[Matt Mackall]

On Mon, 2015-05-18 at 10:53 -0700, Durham Goode wrote:
> 
> On 5/17/15 6:57 PM, Matt Mackall wrote:
> > On Fri, 2015-05-15 at 15:54 -0700, Tony Tung wrote:
> >> # HG changeset patch
> >> # User Tony Tung <tonytung at fb.com>
> >> # Date 1429655274 25200
> >> #      Tue Apr 21 15:27:54 2015 -0700
> >> # Node ID dc122dd80665762d8febe2db1a08ce00a63d5ab8
> >> # Parent  18cadf9d058931ef00e5272d15cb5cf2ebc3a248
> >> commit: add a -M/--reuse-message option to copy a commit message from a
> >> revspec
> >>
> >> One way to split up a diff that includes a refactor involves resetting to
> >> the ancestor, and then committing the refactor first.
> > For future reference, "resetting" is git-speak that I don't understand,
> > despite having looked at the git-reset manpage countless times. Please
> > don't make me learn git to understand your commit messages, because git
> > makes me sad. Please instead use "updating to", "reverting to", or
> > whatever version of --hard/--soft/--bouncy you mean corresponds to.
> >
> > I vaguely recall discussing this option with Ryan, perhaps at the
> > sprint. I understand the value of the feature, but I think it's below
> > the threshold for UI clutter.
> >
> > One way of getting around that is to make a more generically useful
> > feature with lower UI footprint. For instance, log historically has a
> > ton of individual options to specify which revisons to show:
> > --keyword/date/follow/follow-first/only-branch/only-merges/no-merges/limit/user/prune, etc.
> >
> > We were far past the point where things were cumbersome, the
> > combinations didn't make sense, and it still wasn't expressive enough
> > for all the things users wanted, and new things only worked for log.
> >
> > Now compare that with revsets: basically everything supports revsets,
> > they're vastly more powerful and expressive, and adding new features to
> > revsets doesn't clutter the help and implementation of every command.
> >
> > So a more "generic" version of this feature might look like this:
> >
> >   hg commit -u @255 -m @255 -d @255  # copy the metadata from cset 255
> >
> > ..but would have further reaching consequences:
> >
> >   hg log -d @.  # show commits with the same date as the current one
> >
> >   hg log -r "files(@bookmark)" # show commits that touch the same files
> >
> > The per-command implementation would look something like:
> >
> >    if opts["message"]:
> >       message = cmdutil.expandopt(repo, "description", opts["message"])
> >
> > There are a bunch of problems with the precise form of the above,
> > starting with the fact that we obviously already use "@" for other
> > things, but I think it illustrates why a generic approach is more
> > valuable and maintainable.
> >
> I like the idea of a generic solution.  What about 'hg commit -m 
> {255}'?  Templates use {} to indicate "replace this with the desired 
> value", so we could do the same in commands and revsets, where appropriate.

I'm actually thinking "<foo>", so there's no confusion with templates.

> Would you be ok if it was only done for commit -m initially?  That seems 
> like the 95% case.

Provided it's not open-coded.

-- 
Mathematics is the supreme nostalgia of our time.