Quoting Anton Vodonosov (avodonosov@...):
> 22.10.2012, 18:29, "David Lichteblau" <david@...>:
> > Is there anything special done by this run.sh? Previously saved core
> > file, additional asdf systems loaded, custom dynamic space size, etc.?
> No, just setting SBCL_HOME:
> > (For comparison, an x86-64 build needs roughly 1/3rd more
> > memory on this test, and hence fails completely reliably during GC (with
> > either version of SBCL) when unduly restricted to the same amount of
> > dynamic space.)
> So it's not a bug, it's just not enough memory?
Well, running with a larger --dynamic-space-size would likely help, e.g.
600 or 700. But that does not mean that it is not a regression.
On x86 it is always possible that the conservative GC is being more
conservative than usual, in the sense that it happens to see
unfortunately many possible roots. Unless one considers said
conservativeness a bug in itself, those effects would be merely by
But I'm interested because safepoint builds currently have an actual bug
when it comes to timely GC, so obviously there is some concern that this
problem could have leaked into normal builds (even though I don't see
how that could be).
One more thing to double-check: Can you send me the *features* of this
binary? Or even make the binary available, to double-check whether the
same build fails on my systems in the same way?