[PATCH 6 of 6 RFC] perf: define formatter locally if Mercurial is earlier than 2.2

FUJIWARA Katsunori foozy at lares.dti.ne.jp
Sat Jun 18 12:28:47 EDT 2016


At Sat, 18 Jun 2016 03:29:01 +0200,
Pierre-Yves David wrote:
> 
> On 06/17/2016 08:59 PM, Matt Mackall wrote:
> > On Fri, 2016-06-17 at 19:04 +0200, Pierre-Yves David wrote:
> >> The code itself seems mostly good to me (I need a second pass) but I'm a
> >> bit confused about why we want to make perf.py backward compatible? Old
> >> version of Mercurial can use the old version of the perf extension,
> >> isn't that enough? I'm curious about your use case here.
> > 
> > When you add new perf tests, you want to run them against historic hg. I hit
> > this regularly.
> 
> This quest seems a bit doomed to me as perf.py is likely to keep
> evolving alongside the code base of Mercurial, breaking BC and using new
> feature. Especially now that there is some people working on improving
> our performance tracking.
> 
> The way I usually handle this by applying a small patch with the new
> perf thing when doing the initial comparision (same as I do with
> regression tests)
> 
> That said, I've nothing against this series if it help a bit, but I
> think we should document (near the compatibility hack) the motive better
> so that future contributor understant what is going on here.

I think:

  we have to:

  - make perf.py loadable with as old version hg as possible

  - make historical perf tests work with as old version hg as possible

  we don't have to:

  - make newer perf tests work with old version hg
    (of course, it is good, if a newer test works with old hg)

    therefore, even if perf.py itself can be enabled with an old hg,
    it isn't ensured that all perf tests are executable.

As far as I confirm, we can make recent perf.py loadable with
Mercurial 1.1 or so by this series and some additional patches.


> -- 
> Pierre-Yves David
> 

----------------------------------------------------------------------
[FUJIWARA Katsunori]                             foozy at lares.dti.ne.jp


More information about the Mercurial-devel mailing list