Make UInt panic on overflow if overflow_checks is enabled

[u59149403]

Currently UInt wraps on overflow. Please, make it panic on overflow if cfg(overflow_checks) is enabled, i. e. exactly same thing rustc does with standard numbers. Here is why.

First and foremost, ruint is used mainly for representing monetary values. Wrapping behavior is unnatural for monetary values. So, UInt should panic on overflow. But still some users may want to skip checks for speed reasons. So, we should provide way to configure behavior. The most natural thing to do is to make UInt panic when standard numbers panic and wrap when standard types wrap. So, we should use cfg(overflow_checks).

Of course, this is breaking change, so it should be done in 2.0.0.

I use alloy for managing my cryptocurrency, i. e. for transferring it around. Unnoticed overflows may cause loss of money, I absolutely don't want this. alloy directly uses ruint's UInt. So, I plan to create my wrapper around U256, which will panic on overflow. And I also will use clippy::arithmetic_side_effects to make sure that I will not use arithmetic directly on ruint/alloy's U256. But ideally this should be fixed in ruint itself

[u59149403]

See rust-lang/rust#91130 and rust-lang/rust#111466

[prestwich]

ruint is intended to support the modular arithmetic required for cryptographic applications. Adding overflow checks based on cfg(overflow_checks) would mean that consumers of the library could inadvertently break ruint when used as a subdep. E.g. if you used a hypothetical ruint-ecdsa, it would be broken if you enabled cfg(overflow_checks) for your project

Is there a reason that using the checked_* API is not sufficient for your project?

[u59149403]

Is there a reason that using the checked_* API is not sufficient for your project?

As I said this requires me to depend on special clippy lints. But okay, this is your project, so feel free to close this issue if you want

[prestwich]

when writing consensus code, I would strongly recommend always using checked arithmetic, even with primitive number types, rather than panicking. Accidentally leaving reachable panics in consensus code can lead to network-wide DOS

for now, closing this as wontfix

[Yen]

when writing consensus code, I would strongly recommend always using checked arithmetic, even with primitive number types, rather than panicking. Accidentally leaving reachable panics in consensus code can lead to network-wide DOS

for now, closing this as wontfix

Would it be possible to reassess this as I don't think it's the correct behaviour. I fully agree that people shouldn't be relying on behaviour like this to catch errors, but I was running into issues with overflows and had a hard time debugging the issue as the assumption was that the behaviour here would be the same as the standard rust types. I think the decision of the way to do things should not made here as it has already been decided by the rust team with the standard types. I would argue that replicating the default rust behaviour is far more important than trying to enforce good practice with a different behaviour in this library that nobody will know about unless they are hit by this case wile debugging and end up on this issue like I did.

I think the above comment for using checked arithmetic is a very reasonable thing to say as an argument against the rust standard types behaviour, but here it's far more important to have consistent behaviour with those types and to panic in debug.

[Yen]

For reference, the issue I had debugging was that I was debugging under the rust assumption that in debug mode, overflows will panic. It took a long time checking many other parts of the code before I realised that it was overflowing, and it was the different behaviour in this library that broke that assumption.