You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
One harness's goto-instrument failure aborts the whole run and discards every other harness's results #4966
When goto-instrument fails while preparing one harness, kani / cargo kani / kani autoharness stop with error: goto-instrument exited with status exit status: 6, and nothing else is reported. Not the results of the harnesses that already finished, the results table, or the summary.
Seen with kani autoharness -Z autoharness -j 3 --harness-timeout 120s -p kraken-lab over https://github.com/krakenfx/kraken-cli, with goto-instrument limited to 12 GB of address space. One harness (hash::sha256_hex, see #4965) hit the limit during --ensure-one-backedge-per-target:
Trying to force one backedge per target
Out of memory
...
error: goto-instrument exited with status exit status: 6
The run had already started 38 harnesses, most of them finished, but the output ends at that error: no Autoharness Summary, no Complete - ... line, and an exit code that doesn't distinguish this from a compilation failure. The only way to get the other results was to rerun with that harness excluded.
Expected. Treat a goto-instrument failure as that harness's failure, like a CBMC failure: report it as failed with the reason (e.g. "goto-instrument ran out of memory" for exit status 6 with Out of memory), and go on with the rest. A run over a whole crate shouldn't be only as reliable as its most expensive harness.
I hit the same issue while running kani autoharness --std ./library --harness-timeout 60s over the whole verify-rust-std library. In my case a goto-instrument --dfcc step hung for hours (#4967). When I killed it, the whole run stopped:
error: goto-instrument exited with status signal: 9 (SIGKILL)
No summary table was printed, and the results of the harnesses that had already finished had to be recovered from the log.
Activity
Tianshu-Huang commentedon Oct 5, 2026
I hit the same issue while running
kani autoharness --std ./library --harness-timeout 60sover the whole verify-rust-std library. In my case agoto-instrument --dfccstep hung for hours (#4967). When I killed it, the whole run stopped:No summary table was printed, and the results of the harnesses that had already finished had to be recovered from the log.