Investigate error log on ci light-test padding_len overflow about zkevm-circuits HOT 9 CLOSED

Hi @davidnevadoc do you think is it related to the recent change #1445 ?

from zkevm-circuits.

also + @ed255 maybe have any quick answer :)

from zkevm-circuits.

Hi @davidnevadoc do you think is it related to the recent change #1445 ?

Yes, I think this comes directly related to this discussion #1445 (comment)

My guess is that there's a check that is too "overprotective" and logs an error even though the parameters are safe. The other option is that we're skipping enabling some necessary rows (and thus skipping their constraints), which could mean a soundness issue. So I believe this needs some investigation. I would also suggest to avoid logging errors without panicking in such situations; if the circuit parameters not enough to hold the full witness, we should panic.

from zkevm-circuits.

Yes, this was introduced in #1445 .
I was a bit on the fence about this issue because I too got this error. However, I think it is due to what @ed255 mentioned: An overprotective check.
By having max_rws +3 here

from zkevm-circuits.

Yes, this was introduced in #1445 . I was a bit on the fence about this issue because I too got this error. However, I think it is due to what @ed255 mentioned: An overprotective check. By having max_rws +3 here

zkevm-circuits/bus-mapping/src/circuit_input_builder.rs

Line 377 in 3575aab

max_rws: max_rws + 1,

the error will go away. But I could not logically explain such a magic number so I left the +1 which is what I think makes sense according to the specs.

Thanks for the hint. After I applied the max_rws +3 commit 360ad81 on word-lo-hi the error indeed gone.

However agree that we should investigate what's the root cause for why magic number + 2

from zkevm-circuits.

Yes, this was introduced in #1445 . I was a bit on the fence about this issue because I too got this error. However, I think it is due to what @ed255 mentioned: An overprotective check. By having max_rws +3 here

zkevm-circuits/bus-mapping/src/circuit_input_builder.rs

Line 377 in 3575aab

max_rws: max_rws + 1,

the error will go away. But I could not logically explain such a magic number so I left the +1 which is what I think makes sense according to the specs.

Hi @davidnevadoc , after L377, there is send_end_block which still push 2 RW op plus context lookup, so if wanna double confirm if we set max_rws = max_rws + 1 but not counting rw op happened in set_end_block is it still correct?

from zkevm-circuits.

Tried to fixed in PR #1515

from zkevm-circuits.

there is send_end_block which still push 2 RW op plus context lookup

Oh there it is!! You are right, I missed those and that is why I left the +1. Now that we know where the other 2 come from I'm comfortable having the +3 👍

from zkevm-circuits.

there is send_end_block which still push 2 RW op plus context lookup

Oh there it is!! You are right, I missed those and that is why I left the +1. Now that we know where the other 2 come from I'm comfortable having the +3 👍

Actually +1 + 1 (overall +2) work in this case, 😄 the pr should fixed it generally 👆

from zkevm-circuits.