Verification failures only with --isolating-assertions? #5427
-
I have a lemma which consistently verifies correctly with random seeds.
I am not sure how to think about this.
Dafny 4.6 |
Beta Was this translation helpful? Give feedback.
Replies: 3 comments
-
Is this related to #5424? Is it possible to share the file that fails to verify? |
Beta Was this translation helpful? Give feedback.
-
No, I guess it's not really related... though to my mind this is a good example of The code is gas.dfy. Sorry that I didn't include it earlier, I assumed it might be something easily dismissible. This verifies correctly:
Add
Note that with many other random seeds it fails the 10 iterations in a row. |
Beta Was this translation helpful? Give feedback.
-
I think it's uncommon that |
Beta Was this translation helpful? Give feedback.
I think it's uncommon that
--isolate-assertions
makes a proof less stable, but it's not unimaginable. If you haveassert A; assert B
, then the solver might learn things when provingA
that it then uses to proveB
.--isolate-assertions
will turn the program into two proofs, oneassert A
and anotherassume A; assert B
, so then the learning of provingA
can no longer be used when provingB
. There might be some propertyC
that is useful both for provingA
and forB
, that if you assert it as well, will make your proof stable even when using--isolate-assertions
.