[PATCH] group_mapping: Avoid a talloc

[Volker Lendecke]

On Thu, Jan 09, 2014 at 11:01:28AM +1300, Andrew Bartlett wrote:
> The issue I have is that the previous discussion ([PATCH] s3: set native
> os according to Windows and NBT_ANNOUNCE_VERSION defines) was over code
> used in session setup, and this instance is the only justification was:
> "Hi!Please review & push!".  If someone says 'this is a hot path', then
> I'm happy to go with that, but I just don't see how this is 'hot'
> compared with the tdb access that must follow?
> 
> I'm still concerned, but a little explanation and justification would
> greatly reduce my worries here, because it feels like the standard is
> 'remove all talloc where we think we can use a fixed string', rather
> than 'remove avoidable talloc in the hot path'.

For me it's definitely the former. I will not actively
remove talloc without some proof that it shows up in
profiles, but when I touch code I will be very careful about
talloc.

Every single time I do profiling, talloc shows up really
high on the list. After having to learn this session over
and over again I said to myself that it's enough. talloc is
expensive and if there is a simple and clear stack-based
alternative, use the alternative.

No doubt it's satisfying to gain a few percent after some
performance work, but I would be much happier if we never
got into the position to gain a measurable difference with
some small changes.

Don't get me wrong: talloc is an extremely worthwhile
abstraction and it is the single most valuable single that
Tridge has brought to Samba, but it comes with significant
cost.

Volker

-- 
SerNet GmbH, Bahnhofsallee 1b, 37081 Göttingen
phone: +49-551-370000-0, fax: +49-551-370000-9
AG Göttingen, HRB 2816, GF: Dr. Johannes Loxen
http://www.sernet.de, mailto:kontakt at sernet.de