overflow classification and error is inconsistent on divide by zero in mulDiv #201
Labels
effort: medium
Default level of effort.
priority: 2
We will do our best to deal with this.
type: refactor
Change that neither fixes a bug nor adds a feature.
work: complicated
Sense-analyze-respond. The relationship between cause and effect requires analysis or expertise.
What version of PRBMath are you using?
5959ef5
What version of Solidity are you using?
0.8.19
Describe the bug
the unchecked division at the start of mulDiv actually jumps into a divide by zero check even though it is "unchecked".
this is probably a good thing because
mulDiv(0, 1e18, 0)
andmulDiv(1, 1e18, 0)
and other examples causeprod1
to be0
which is treated as a "non overflow case" according to the comments, but we still don't want to allow dividing by zero here.further down in the same fn, divide by 0 triggers an "overflow" error if
prod1 >= denominator
and the denominator is0
.i don't think this is dangerous but it leads to inconsistent errors which makes fuzzing downstream quite fiddly if we want to assert on the error selector with foundry.
the same error should be thrown for "divide by 0" regardless of the internal
prod0
andprod1
logic.The text was updated successfully, but these errors were encountered: