Dear all,
Recently I find quite often that after "go test -fuzz" reported an error, the command shown to re-run the test does not reproduce the failure. For example, just now I got the following:
voss@dumpling [..nt/tounicode2] go test -fuzz FuzzToUnicode
fuzz: elapsed: 0s, gathering baseline coverage: 0/2 completed
fuzz: elapsed: 0s, gathering baseline coverage: 2/2 completed, now fuzzing with 10 workers
fuzz: elapsed: 3s, execs: 280461 (93481/sec), new interesting: 24 (total: 26)
fuzz: elapsed: 5s, execs: 410320 (72348/sec), new interesting: 60 (total: 62)
--- FAIL: FuzzToUnicode (4.80s)
--- FAIL: FuzzToUnicode (0.00s)
tounicode_test.go:129: template: tounicode:26:2: executing "tounicode" at <Single $cs .>: error calling Single: runtime error: integer divide by zero
Failing input written to testdata/fuzz/FuzzToUnicode/2566873a28045c1b
To re-run:
go test -run=FuzzToUnicode/2566873a28045c1b
FAIL
exit status 1
FAIL seehuhn.de/go/pdf/font/tounicode2 4.957s
voss@dumpling [..nt/tounicode2] go test -run=FuzzToUnicode/2566873a28045c1b
PASS
ok seehuhn.de/go/pdf/font/tounicode2 0.122s
The fuzzer reported a failure, and when I re-run the test I get a "PASS".
How to debug such a problem?
I understand that my fuzz function may somehow be non-deterministic, but so far I have not found any cause of non-determinism and re-running the test 10 times gives me 10 passes. Also, for a bug the fuzzer reported earlier, I found it very hard to believe that the given input could make the fuzz function reach the site of the error message. Could it be that the fuzzer sometimes looses track of which input belongs to which fuzzing run, and then reports the wrong input?
If somebody wants to play with this, here is how to reproduce (up to randomness) the fuzzing run shown above:
Any suggestions on how to debug such errors would be most welcome.
Many thanks,
Jochen