Re: [boehm-gc] Import 6.3 alpha 1

[Jeff Sturm <jsturm at one-point dot com>]

On Tue, 29 Jul 2003, Ranjit Mathew wrote:
>     http://gcc.gnu.org/ml/java/2002-06/msg00073.html
>     http://gcc.gnu.org/ml/java/2002-06/msg00252.html
>     http://gcc.gnu.org/ml/java-patches/2002-q2/msg00490.html
>
> AFAICT, this is still not integrated into libgcj.

That patch was far from perfect for reasons I outlined, though I had hoped
more ideas would emerge to lower the costs of scanning roots.  One
approach is reducing the size of the static data areas generated by gcj.
Another is to scan static data selectively, with GC_exclude_roots or some
other technique, as in the patch you referenced.

There would be other benefits to shrinking static data, but it seems
difficult given the current class layout.  If class data used a
position-independent encoding, much of it could perhaps move to read-only
sections.  But certain members of the class record (e.g. string table)
will probably always have to be initialized at runtime.

> Could this not help the issue raised in frequent
> messages on the GCJ list complaining about almost
> unbounded heap growth like, say, this one:

The root set is in part responsible for determining the frequency of
collection.  Fewer collections result in a larger heap.  But I'm unsure
that root set size can alone be responsible for *unbounded* heap growth.

That said, a patch like this could also avoid DYNAMIC_LOADING bugs
on certain platforms.  ISTR win32 having some trouble in that area.

Jeff