Regression failures with verify_gens etc
Affects | Status | Importance | Assigned to | Milestone | |
---|---|---|---|---|---|
SBCL |
New
|
Undecided
|
Unassigned |
Bug Description
Some continuous build should run the test suite with heap verifications.
I've recently observed three failures (all irreproducible of course).
Perhaps these are due to enabling sb-sprof in parallel-exec.
If the failure rate goes to 0 with that off, it's probably a safe bet that sprof is to blame.
(Ironically the defstruct failure is while trying to produce the sprof report, but the sprof graph creation code is extremely unlikely to be the root cause of anything)
1. defstruct.impure
BEGIN gc_gen(0,1)
/pinned objects(g0): large=1 (113800 words), small=17
Verify after GC(0) [threads] [RO] [static] [dynamic]fatal error encountered in SBCL pid 2269301 tid 2269301:
GC invariant lost, file "verify.inc", line 323
GC: stw_delay=31,50,79 µs (min,avg,max) pause=83493,
0: fp=0x7f5e64727310 pc=0x1001e0c96b SB-SPROF:
1: fp=0x7f5e64727378 pc=0x1001e5940c SB-SPROF::MAKE-NODE
...
2. compiler.impure
BEGIN gc_gen(0,1)
/pinned objects(g0): large=0 (0 words), small=21
Verify after GC(0) [threads] [RO] [static] [dynamic]Ptr 0x1004799953 @ 100491e860 (lispobj 100491e867,pg2339) sees junk
fatal error encountered in SBCL pid 1975592 tid 1975592:
Verify failed: 1 errors
3. hash-table.impure
BEGIN gc_gen(0,0)
/pinned objects(g0): large=0 (0 words), small=94
CORRUPTION WARNING in SBCL pid 2003898 tid 2003898:
Memory fault at (nil) (pc=0x561546bc6be5, fp=0x7f9cac7be1c0, sp=0x7f9cac7be180) tid 2003898
The integrity of this image is possibly compromised.
Exiting.
possibly unrelated random failure-
::: UNEXPECTED-FAILURE :FLUSH- VECTOR- CREATION due to SIMPLE-ERROR:
( FUNCALL F)
"Expected the form
NOT to cons, yet running it for 10000 times resulted in the allocation
of 82096 bytes (8.210 per run)."