question about revset usage

[Mads Kiilerich]

On 09/21/2014 10:45 AM, Faheem Mitha wrote:
> If I use revsets as follows:
>
> faheem at orwell:/usr/local/src/mercurial/hg$  hg log -r 
> 'author(mpm at selenic.com)'
>
> The order of the csets is reversed - the oldest csets are at the top, 
> unlike the normal result of `hg log` where the newest csets are at the 
> top. I.e.
>
> faheem at orwell:/usr/local/src/mercurial/hg$  hg log -l1 -r 
> 'author(mpm at selenic.com)'
> changeset:   0:9117c6561b0b
> user:        mpm at selenic.com
> date:        Tue May 03 13:16:10 2005 -0800
> summary:     Add back links from file revisions to changeset revisions
>
> I rarely use revsets, but in any case, I've not noticed this behavior 
> before, nor could I find any documentation about it. I also can't 
> think of any good reason for this behavior. If this is correct 
> behavior, and if I am not missing something obvious, I suggest it be 
> documented somewhere.

The unsorted order of revsets is not documented and can thus be 
considered undefined and can thus not be wrong. If you want a special 
order, use sort().

One reason to have it that way is that different predicates most 
efficiently will yield results in different orders. For example, 
everything involving unpacking of revlogs is most efficiently done with 
increasing rev numbers, while ancestors obviously yield descreasing rev 
numbers.

I agree that it would be nice if everything used the same order 
(decreasing!) unless another sorting was requested but I am afraid that 
it would break a lot of existing revsets out there so it might not be 
feasible to change it even though it is undocumented and somewhat 
surprising behaviour. Documenting it beyond the fact that it is 
undefined might not be feasible.

(In Kallithea we take care to always show revisions in the "right" order 
... also because graphs become very confusing if shown upside down.)

/Mads