Clarification: is the hgmerge program always called?

[Matt Mackall]

On Tue, May 16, 2006 at 03:58:02PM -0700, Vadim Gelfer wrote:
> On 5/16/06, Vadim Gelfer <vadim.gelfer at gmail.com> wrote:
> 
> >i worked on it in past few days. i will put bundle up somewhere soon
> >when new drop is ready.
> 
> here is patch against crew repo. remember, merge code is written by
> bram cohen. i just plug it into mercurial.
> 
> to enable:
> 
> [extensions]
> hgext.histmerge =
> 
> to enable demo command (called hmdemo):
> 
> [extensions]
> hgext.histmerge.demo =
> 
> demo will find all merges in a repo where parents of file have not
> same contents as child and compare result of histmerge on each file
> with result of 3-way merge.
> 
> i have run demo command on repo with 12500 revs. 5% of changes have
> merges on files. 95% of cases, histmerge and 3-way merge give same
> results. about 60% of rest of time, histmerge gives better merge than
> 3-way.

Interesting. This is about what I was thinking would happen, but it's
definitely nice to see it quantified. Is your repo a public project?
If not, can we run this against one? Eg, the Linux kernel history?

It'd also be good to characterize/classify what those improved cases
look like.

-- 
Mathematics is the supreme nostalgia of our time.