Mercurial performance tracking tool

Yuya Nishihara yuya at tcha.org
Sun Sep 4 06:25:34 EDT 2016


On Thu, 1 Sep 2016 11:16:29 +0200, Philippe Pepiot wrote:
> It has detected a regression between 1% and 10% on almost all revset 
> benchmarks: 
> https://jenkins.philpep.org/hgperf/#summarylist?sort=2&dir=desc (related 
> changeset is https://www.selenic.com/hg/rev/abe4eecc). I don't have the 
> mercurial knowledge required to understand this regression, so I let you 
> analyze if something needs to be done or not.

Interesting. That +20usec is the cost to make new parsed tree recursively.
These revsets are small so the parsing cost appears dominant. For instance,
"(0)" is 10us slower than "0" on my Core i5 laptop.


More information about the Mercurial-devel mailing list