Re: GC statistics (was Re: big project ported)

[Tom Tromey <tromey at cygnus dot com>]

>>>>> "Godmar" == Godmar Back <gback@cs.utah.edu> writes:

Godmar> The reason I said it would translate nicely is because libgcj
Godmar> appears to use almost the same data structures for fields and
Godmar> methods as kaffe does (and for utf8s and for the constant pool
Godmar> and for classes too.)  Who knows, there may have been some
Godmar> blind cross-pollinization going on somewhere.  Personally and
Godmar> honestly, I'd very much wish to continue it.

gcj was originally written to work with kaffe.  I don't remember the
whole story, but at some point we decided to write our own libraries
instead.  The layout of the Class object is probably the only place
where kaffe's influence can really still be seen (and even that was
rewritten from scratch; I did it by reading the copmiler).

There has been one proposal to change the Class structure to use Java
objects where possible, to make reflection code simpler (and there was
some other reason I forgot).  I don't know if we'll follow up on that
or not.

>> The Boehm collector has built-in support for bitmap-based marking;

Godmar> Yes, I know.  I was wondering why you didn't use that---I
Godmar> thought the reason was that you may have to handle other
Godmar> cases, such as marking classloaders or thread objects
Godmar> etc. that Boehm would not cover.

At the time I was worried about objects larger than what is allowed by
the bitmaps.  Also I was concerned about correctness more than
efficiency at the beginning.  I'm more willing to experiment now that
it is known to work.

T