[PATCH] revset: introduce the summary predicate

[Yuya Nishihara]

On Thu, 12 Jan 2017 00:40:36 -0500, Matt Harbison wrote:
> On Wed, 11 Jan 2017 09:15:28 -0500, Yuya Nishihara <yuya at tcha.org> wrote:
> > I also like the idea of deprecating
> > grep() since grep() sounds like searching the file contents.
> 
> I don't have a strong feeling on that.  If somebody makes a new search  
> function though, I wonder if it should be like revset.matching() (but with  
> stringmatcher support), where the user can control the fields searched, in  
> order to avoid this sort of ambiguity.  I wouldn't want to fold grep()  
> into author() because of the clashing case sensitive/insensitive you  
> mention below.

Fair enough. (and my feeling wasn't strong neither.)

> >> That leaves only 'author' and 'desc' as not providing the full
> >> functionality of the others.
> >>
> >> > C) If we do A + B, that means 'desc' is the only oddball left.  I  
> >> don't
> >> > like the idea that case sensitivity for a raw pattern and a 'literal:'
> >> > prefixed pattern would differ.  They are both literals in my mind, and
> >> > it would be the one remaining exception.  The 're:' prefix could  
> >> follow
> >> > regular rules.
> >
> > I won't insist that 'literal:' must be case-sensitive (because of (A).)
> > However, I would guess 're:' is also case-insensitive if 'literal:' is.
> > In my mindset, desc() is a case-insensitive matcher in that case.
> >
> > So I lean towards adding case-insensitive desc('re:'), which would be at  
> > least
> > consistent in that desc() always ignores cases.
> 
> The only reason I would guess 're:' is case sensitive, is because I've  
> never run into one that hasn't been.  I do like the consistency argument  
> though, so let's try that.  I wonder if in addition to the  
> 'icase-literal:' you suggested, if this also means we need a 'case-re:',  
> since it doesn't look like you can force re.I off.  I don't see any real  
> benefit for author(), but I can maybe see it for desc().  The series I'm  
> about to submit hints at the ability to add these with one or two lines.   
> See what you think.

Honestly I just postponed the consideration about case-sensitive desc() by
this design. ;) I'm not a fan of 'case-re:' because 're:' is case-sensitive
in most revset functions. I'd rather add new case-sensitive desc() or more
general function.

I found (?-i:...) syntax, but that's Python 3.6 thing, sigh.

https://docs.python.org/3/library/re.html