Concept of the day: filesets?

[Matt Mackall]

On Tue, 2010-11-16 at 13:38 -0600, Kevin Bullock wrote:
> On Nov 16, 2010, at 1:15 PM, Matt Mackall wrote:
> 
> > What if we applied the revset query language concept to file patterns?
> > 
> > We have long had the ability to do:
> > 
> >  hg ci "re:(foo|bar).*"
> > 
> > We can specify re/glob/path just about wherever filenames are taken. If
> > we add a new set type, we can now do:
> > 
> >  hg ci "set: modified() and contains('mpm') and exec() and not glob('**.py')" 
> 
> I love this idea.
> 
> > Want to check in only files not in a subrepo?
> > 
> >  hg ci "set: not subrepo()"
> 
> The subrepo() predicate would presumably match .hgsub and .hgsubstate as well?
> 
> Brainstorming predicates/functions:
> 
> Basic status: added(), modified(), deleted(), removed(), ignored(), unknown(), clean()
> Extended status: copied(), renamed()
> Merge status: unresolved(), resolved()
> Paths: glob(str), re(str)
> Contents: contains(str/regex?)
> Metadata: user(str) [modified by user]

find(1) (the classic implementation of predicate matching on files)
knows about these:

tests (N can be +N or -N or N): -amin N -anewer FILE -atime N -cmin N
      -cnewer FILE -ctime N -empty -false -fstype TYPE -gid N -group NAME
      -ilname PATTERN -iname PATTERN -inum N -iwholename PATTERN -iregex PATTERN
      -links N -lname PATTERN -mmin N -mtime N -name PATTERN -newer FILE
      -nouser -nogroup -path PATTERN -perm [+-]MODE -regex PATTERN
      -readable -writable -executable
      -wholename PATTERN -size N[bcwkMG] -true -type [bcdpflsD] -uid N
      -used N -user NAME -xtype [bcdpfls]

We probably don't want all of those, but a few are interesting.

Also, we'll want a syntax to let us include file lists/patterns from
other sources, for instance:

 set:include(files.lst)

> What about making at least some predicates take revsets too? So you could do:
> 
>  hg log 'added(date(2010-01-01 to 2010-02-01))'

Reasonable concept. But you'll still need the set: prefix. I'll note
that some revset operators already take filespecs, so this would be
mutually recursive.

> First commands I would expect to take these filesets: st, log, ci, and
> anything that takes -X and -I... I'm sure there's others to add to
> that list.

Anything that uses match objects (just about everything) would get
support for these automatically.

-- 
Mathematics is the supreme nostalgia of our time.