[PATCH] OS X: try cheap ascii .lower() in normcase before making full unicode dance

[Matt Mackall]

On Wed, 2013-01-30 at 00:19 +0100, Mads Kiilerich wrote:
> On 01/29/2013 11:41 PM, Matt Mackall wrote:
> > On Tue, 2013-01-29 at 20:09 +0100, Mads Kiilerich wrote:
> >> On 01/29/2013 06:23 PM, Kevin Bullock wrote:
> >>> On Jan 29, 2013, at 10:42 AM, Mads Kiilerich wrote:
> >>>
> >>>> # HG changeset patch
> >>>> # User Mads Kiilerich <madski at unity3d.com>
> >>>> # Date 1359475301 -3600
> >>>> # Branch stable
> >>>> # Node ID b7e3f9a8796fc31b6d81d134072b4daa3e0d42ec
> >>>> # Parent  a58d8936647aa270854cd919fe8e8b2da1c1c669
> >>>> OS X: try cheap ascii .lower() in normcase before making full unicode dance
> >>> LGTM
> >> crewed
> > Surprised you both think this is stable-appropriate.
> 
> That might have been a misjudgement.
> 
> It fixed a performance regression back from 1fa41d1f1351, made in stable 
> and released in 2.0.1. (It traded some performance for correctness - 
> which is fine ... but better performance and same correctness should 
> also be fine).
> 
> It also uses a pattern already used in encoding.py, so I consider it 
> quite low risk.

It is indeed practically zero risk, I absolutely agree. But stable is
less about managed risk (which people are notoriously bad at estimating)
than managed code churn (which is quite a bit easier to get a handle
on). This is why I also don't want to see code cleanups in stable.

> If you disagree then we can back it out.

We can leave it, but in the future I'd like folks to err towards less
churn.

-- 
Mathematics is the supreme nostalgia of our time.