>>>>> "ab" == Alastair Bridgewater <alastair.bridgewater@...> writes:
ab> This looks familiar. As though it might be
ab> https://bugs.launchpad.net/sbcl/+bug/999972 all over again. I don't
ab> have a fix for you, but if it's the same problem then I have a
ab> workaround (workaround described along with the bug).
I am using GCC version 4.7.0. I can't reproduce the GC errors using
your backtrace-test, even running it 100 times longer. However, I can
reproduce it running random-integer-testing (but it seems to depends
on the build; git revisions a little after my initial report no longer
failed, and now they are failing again).
I tried Martin's suggestion (as I understood it -- see below), with no
apparent difference in behaviour.
diff --git a/make-host-2.lisp b/make-host-2.lisp
index 72ce598..e45ab96 100644
@@ -23,7 +23,9 @@
;; SAFETY = SPEED (and < 3) should provide reasonable safety,
;; but might skip some unreasonably expensive stuff
;; (e.g. %DETECT-STACK-EXHAUSTION in sbcl-0.7.2).
- (safety 2) (space 1) (speed 2)
+ ;; (safety 2) (space 1) (speed 2)
+ ;; emarsden2012-06-09
+ (safety 3) (space 1) (speed 2)
;; sbcl-internal optimization declarations:
;; never insert stepper conditions