GC performance - was Re: [gclist] GC topics

[Scott.Nettles@LDL.VENARI.CS.CMU.EDU]

>>> Well, I guess I must be dumb, or misinformed, so please give me advice.
>> 
>> Misinformed is my guess :-).
>> 
>Well, obviously, reading what came next, I was not that misinformed.

Let's not be snotty. After all, you were the one that said "dumb, or
misinformed" and he used a smiley to show he was teasing.

> even you seem to admit that conservative GC *is* slower than precise
> GC, 

I don't think there was any such admission. The factors involved are to
complicated to answer in the abstract. In one limiting case, that in
which no collections are ever done, conservative GC is clearly cheaper,
after all it does no work to help the GC find pointers. Flaming people
about performance without offering solid results does not reflect well
on a person.

>does not apply), precise GC has clear advantages.

And clear disadvantages. Note I work on precise collection, and for my
work there are advantages, but...

>work, but on some architectures, the impact on performance would be
>negligible, if not negative.

Proof? And do those architectures matter?

>I read. I do believe it is not a generally good solution, except

Belief != proof.

GC is too complicated and important a topic to dismiss useful
techniques without careful study. Especially when we NEED those
techniques in some important arena's.

Scott