Document/unify approach for division by 0

[vezenovm]

Problem

In this PR (#2475) extra checks were added to guarantee that divisions by 0 on fields and integers both resulted in unsatisfied constraints. However, the strategy differs between Field's and integers. We should attempt to unify our strategies for checking against division by 0 or at a minimum clearly document why the strategies are different.

Check PR #2475 and its description for more information as to why the strategies differ for checking division by zero. The gist of it though is that for field division we multiply against the field's inverse while for integer division we use euclidean division. We have a check when computing the inverse which checks that the inverse returned from brillig multiplied by the original value being inverted is equal to one. Doing a Brillig inversion on zero does not panic when executing the Brillig VM and this constraint fails as expected. Now for integer division, which does a euclidean division, we get a panic when executing the Brillig VM. Thus to avoid the panic, we execute checks for the inputs to euclidean division in ACIR gen before executing code for euclidean division.

Happy Case

It should be clear to new contributors why we check for division by zero in this specific manner.

Alternatives Considered

We should most likely have the Brillig VM not panic on an attempted division by zero and rather have the caller (nargo) handle that error. This may lead to clearer errors at least for euclidean division, and we could remove the inputs check before every euclidean division. However, this does still lead to a divergence in strategies as Field division by 0 will still fail with a constraint failed. We could potentially return the same error from the Brillig VM in both these instances in the name of a having a unified strategy.

Additional Context

No response

Would you like to submit a PR for this Issue?

No

Support Needs

No response