Re: GC Merge?

[Richard Henderson <rth at cygnus dot com>]

On Thu, Sep 02, 1999 at 11:39:37AM -0700, Mark Mitchell wrote:
>   Jeff, Richard, do you recall any particular design problems with the
> current GC implementation?  In other words, short of run-of-the-mill
> bugs, what were the reasons this patch was never incoprorated?  

I don't recall any particular problems, other than the fact that it
was turning out to be annoyingly time consuming to find all of the
roots that needed registering, and to kill the one-off caches tucked
away in dusty corners.

Replacing my Amazingly Stupid Collector [tm] with something that doesn't
completely suck performance-wise may be high on your list of things to
do once you get things in shape for mainline.  Using the Bohem collector
if available is problably the fastest way to do that, though it has the
disadvantage of not needing roots marked.  Which will lead to bitrot of
the non-conservative collector if the gcc developers take to using it. 

Not having a non-conservative collector available probably isn't an option,
as there are too many places were we wouldn't be able to build gcc any
longer.  Of course, it may be that the only places left that don't support
the Bohem collector are the same places where gcc just doesn't work any
longer either -- old Pyramid and other such cruft.

Anyway, in theory a non-conservative collector could be faster and more
memory efficient.


>   Are there any objections to my plan?

I hope you're not jumping from the frying pan into the fire.

It's a Very Good Thing to have done, and I will certainly help out,
but I worry about taking on something like this under time pressure.
Of such are ulcers and burnouts made.


r~