[PATCH RFC] log: add --from for showing history from a certain rev

[Matt Mackall]

On Fri, 2015-02-06 at 13:17 -0800, Durham Goode wrote:
> On 2/6/15 12:37 PM, Sean Farley wrote:
> > Matt Mackall writes:
> >
> >> On Fri, 2015-02-06 at 11:26 -0800, Sean Farley wrote:
> >>> Durham Goode writes:
> >>>
> >>>> # HG changeset patch
> >>>> # User Durham Goode <durham at fb.com>
> >>>> # Date 1423249495 28800
> >>>> #      Fri Feb 06 11:04:55 2015 -0800
> >>>> # Node ID 86c462eee946ce6ce63140fa494e765196107126
> >>>> # Parent  e1dbe0b215ae137eec53ceb12440536d570a83d2
> >>>> log: add --from for showing history from a certain rev
> >>>>
> >>>> Right now it's very obtuse to show the history of a particular rev (hg log -r
> >>>> 'reverse(::foo)'). This adds a --from option that allows "hg log --from foo".
> >>>>
> >>>> Sending as an RFC since the naming might be controversial and there might be
> >>>> better ways of integrating it with existing commands (like using this to show
> >>>> history for bookmarks, hg log --from mybook vs hg log --bookmark mybook).
> >>> As we found out on IRC, both myself and another person got confused with
> >>> the naming of '--from': we both thought it meant 'foo::'.
> >>>
> >>> I do like the idea of this patch for logging the history of a symbol. In
> >>> most commands, we have '-b' for branch and '-B' for bookmarks. Since we
> >>> have this new namespace API, why not use that?
> >>>
> >>> hg log --name/-N foo
> >> Log will probably never get another filtering option. It has too many.
> > Fair enough.
> >
> >> We should probably instead figure out why -fr foo doesn't do what's
> >> wanted.
> As it stands, -r implies 'show me exactly these' revisions.  So -f 
> doesn't modify -r at all (nor should it I think).

I don't understand that. In my mind, there are basically two reasonable
behaviors for hg log -f -r <spec>:

 ::(<spec>)  # follow the revisions in the spec

or the slightly weird and less useful:

 ::. and (<spec>)  # filter the revset relative to .

A third almost meaningless possibility is:

 ::. or (<spec>)  # show the union of two unrelated things

You seem to be arguing for:

 <spec> # silently ignore -f

..and then adding a whole new flag to do what -f is more or less
documented to do.

For comparison, look at how -m/-M interact with revsets:

$ hg log -q -M -r '::.' | wc
  73691   73692 1389014
$ hg log -q -r '::. and not merge()' | wc
  73691   73691 1389005
$ hg log -q -m -r '::.' | wc
   3712    3712   70519
$ hg log -q -r '::. and merge()' | wc
   3712    3712   70514

-- 
Mathematics is the supreme nostalgia of our time.