Re: gc.c -- possible logic error?

On Tue, 2 Oct 2007, Tanaka Akira wrote:

> In article <[e-mail]>,
>   Hugh Sasse <[e-mail]> writes:
> 
> >> I measured the example by valgrind --tool=massif.
> >> (massif is a heap profiler comes with valgrind.)
> >
> > Thank you for this.  I don't have valgrind on this Solaris machine.
> 
> valgrind works only on Linux.

I promptly looked that up, and discovered that. Oh well. Thank you.

> I don't know a heap profiler on Solaris.

Nearest thing I know of is truss, and that only monitors system calls.

> 
> The image generated by valgrind:
> http://www.a-k-r.org/tmp/ostruct.png

That's rather nice :-).  Can you get it to plot that with the heap
at the bottom of the display?  I'm finding it hard to see if the
heap decreases at all at the end.  It certainly does increase during
the early stages, so it is not as if everything has been allocated
to one chunk.  But when GC.start should make things go away, they
don't really seem to disappear.

I've seen no evidence that Ryan Davis's memory analyser:

http://blog.zenspider.com/archives/2006/08/ruby_memory_vis.html

ever got released, and the picture looks like a Mac gui, so I 
don't think I could use that anyway.
> -- 
> Tanaka Akira
> 
        Thank you,
        Hugh