[PATCH] localrepo: fix prepush warnings about new heads (issue 2131)

[Peter Arrenbrecht]

On Thu, Apr 8, 2010 at 7:54 PM, Sune Foldager <cryo at cyanite.org> wrote:
> On 08-04-2010 17:24, Peter Arrenbrecht wrote:
>> On Thu, Apr 8, 2010 at 5:22 PM, Peter Arrenbrecht
>> <peter.arrenbrecht at gmail.com> wrote:
>>> # HG changeset patch
>>> # User Peter Arrenbrecht <peter.arrenbrecht at gmail.com>
>>> # Date 1270740102 -7200
>>> localrepo: fix prepush warnings about new heads (issue 2131)
>>
>> @crew, please do not push this before getting OK from Sune.
>> Thanks,
>> -parren
>
> This is a nice patch :-). I think the 'get the total counts' approach
> stems all the way back from before our original changes to accommodate
> named branches, and we just kept it like that. This way is, if not
> clearer (although I also think so), at least shorter and at least as
> quick which is nice.
>
> About issue 2131: I introduced that in a regression in 1e819576e926,
> which was supposed to optimize some stuff (well, it did ;-)). So instead
> of just pushing your refactor I did the following:
>
> I made a partial backout of my buggy patch and added to it your tests.
> This fixes issue 2131, so I pushed that as such.
>
> Then I'd like to push your patch without the tests and with a changed
> subject line, now that it's "just" a refactor of the code. I'll see if I
> can catch you on IRC later; I'll be on all evening, and we can talk
> about it. Otherwise, re-submit or just tell me what to change to subject
> to. I can also come up with it, but since it's your patch... :)

Please don't commit just yet. I think it still has a bug. When you add
a new head to branch A while also adding a node to branch B which is a
descendant of a (now former) head on branch A, we would not note the
new head on branch A. Means we'd have to compute nolongeraheadnodeset
per branch, not globally. Shall look into this.
-parren