[GnoVM] Top-level error does not revert state

[leohhhn]

[GnoVM] Top-level error does not revert state

[deelawn]

I'm very skeptical that state should be reverted when an error is returned from a top level realm function; it seems inconsistent with go's behavior. For example, if I modify a value in go that exists in a scope outside of the function from which it is modified, I would expect that value to be persisted regardless of whether or not the function that modified it later returns an error. Returning an error usually means that the operation could not be fully completed as intended and any state modifications outside the scope of the function that returned the error will remain. Panicking means something went horribly wrong and we need to get rid of all state unless the panic is recovered. It seems like the current implementation most closely aligns with how the terms are used in go.

[ajnavarro]

I'm with @deelawn here for several reasons:

• Consistency: rollback only when returning an error on top-level methods but not on the rest is not consistent. It might mislead users.
• Unexpected: as @deelawn said, is not consistent with Go behavior.
• Limiting choices: We already have a way to roll back the Realm state, that is, throwing a panic. I can see use cases where a Realm's dev wants to return an error without rollbacking the whole state, like when saving some metadata about who voted, or metrics about the amount of people that used the Realm, as examples.

[petar-dambovaliev]

Errors are just values, they have no special semantics within the language.
Just like you can return an integer, you can return an error. We shouldn't rollback when people create integers, right?
Panic, on the other hand, is special and does what you suggested. This proposal is basically to make errors be treated as panics. I am against.

[leohhhn]

Okay. got it. Documentation already states that panics revert state, but I will make sure to add mentions and examples of errors not reverting state, which might be what blockchain developers could expect. Thanks.

[zivkovicmilos]

Thank you for giving your thoughts @deelawn @ajnavarro @petar-dambovaliev

I think we can find good middle ground here -- don't introduce this top-level error logic, but make it clear in the documentation that panics are equivalent to reverts. @leohhhn, what do you think?