Re: Developer machine spec for Linux kernel w/ darcs

On Wed, Dec 01, 2004 at 01:11:55PM +0100, Tim O'Callaghan wrote:
> 
> > Not really.  It's very hard to see *why* things are being held in memory
> > from memory profiling. 
> 
> Maybe i'm missing something here, but other than standard library
> stuff, how can you not know why things are being held in memory?
> properly scoped variables should be freed/GC'd when they are out of
> scope right?

Laziness makes it more complicated to work out what is holding on to
what.

> While i understand the GC has to check references, i
> assmue profiling can help minimise the number of internal references?
> 
> > I know Ian has been doing some work recently, and
> > that he's been using profiling, but his changes have been improving (I
> > think) both memory and cpu usage, so I'm not sure which he's profiling.

Time profiling is the easiest way to find the problems I'm currently
looking at (unnecessary conversion between Doc, String and PackedString).

It may be there's also stuff being kept around unnecessarily, in which
case heap profiling will probably be better.

Thanks
Ian