Re: Hunting for performance regressions in libstdc++

[Paolo Carlini <pcarlini at suse dot de>]

Hi Howard,

> On Jun 16, 2005, at 7:53 AM, Jonathan Wakely wrote:
>
>> You aren't, by any chance, using std::set::insert() with hints ?
>>
>> We changed the behaviour of that function recently to conform to the
>> standard, which contradicts the traditional meaning of the hint. If
>> you are supplying a hint that points just before the insertion point
>> you will not get any speed-up.
>
> You are fighting an uphill battle with this change.
>
> http://www.open-std.org/jtc1/sc22/wg21/docs/papers/2005/n1780.html

Maybe you don't remember ;) but I was in Lillehammer when *finally* that
DR has been discussed and you volunteered to write down the document: at
that time was still *open* (years after the submission ;) and apparently
nobody really paid attention before.

Anyway, we implemented the mentioned change weeks *before* the meeting
and, the needed feedback from the committee lacking, we could only
implement the letter of the standard (the same choice of RW, by the way).

Of course, as soon as the DR makes progress we are going to implement
its resolution. Ah, FWIW, in Canada, you will get my vote in case of
straw polls (I don't think will be necessary, right?)

Paolo.