'hg log FILE1 FILE2' order depends on the order of the files?

[Yuya Nishihara]

On Sat, 13 Feb 2016 01:53:59 +0900, Yuya Nishihara wrote:
> On Thu, 11 Feb 2016 20:47:32 +0000, Pierre-Yves David wrote:
> > On 02/11/2016 06:20 PM, Stanislau Hlebik wrote:
> > > Maybe I'm doing smth wrong, but when I tried to run
> > > sort(::0902ae52b1289c6e88bbbb55eb59464a1ebbce91, -rev) result still
> > > depends on the file order.
> > 
> > That's probably because the file arguement are added tot he revset after 
> > the fact (and probably as first element for some reason) try adding your 
> > file as revset element directly.
> 
> This looks like a regression around smartset + logrevset.
> 
> "hg log -r R A B" constructs the following expression,
> 
>   revs = R
>   expr = filelog(A) or filelog(B)
> 
> and evaluates expr against revs.
> 
>   matcher = revset.match(ui, expr)
>   matcher(repo, revs)
> 
> This is equivalent to "R and (filelog(A) or filelog(B))", which builds the
> following set:
> 
>   <addset
>     <filteredset <R>>,
>     <filteredset <R>>>
> 
> Therefore, the ordering of <R> isn't preserved.
> 
> Perhaps, <addset> should be wrapped by <filteredset self=R> if R is dominant,
> but the problem won't that simple.

Reported as
https://bz.mercurial-scm.org/show_bug.cgi?id=5100