Math operations with Int does not warn overflow

[felipemarkson]

Hi, I guess that could be a bug, but overflows shouldn't be warned?

julia> typemax(Int) + 1
-9223372036854775808

julia> typemax(Int)^2
-1

julia> typemax(Int)*2
-2

Compilation: Generic Linux on x86 64-bit
Version: Julia v1.6.5 (LTS)

[oscardssmith]

dup of #22539

[felipemarkson]

Yeah, sure, I saw that Issue. But this shouldn't be warned?
I think this silent overflow can cause huge bugs in production projects.
Not everybody knows that you need to use BigInt or could be a function that causes the overflow and the programmer is not warned.

[felipemarkson]

I could think of a way to warn that.
Should I create a PR?

[oscardssmith]

The problem isn't warning, the problem is that warning without tanking performance. This is probably achievable for exponentiation, but definitely isn't for addition/multiplication.

[felipemarkson]

I understand that, but could you please appoint me where the code could be changed to implement it?

I want to make some tests and benchmarks.

[felipemarkson]

dup of #22539

I also think this is not a duplication of that issue.

[oscardssmith]

Base/int.jl:87 for addition, and 88 for multiplication. Good luck.

[KristofferC]

A warning is a problem. It should either error or do nothing, but not warn.

[felipemarkson]

I don't know.
In my experience with embedded systems, you can make some deep tricks with overflows to get some performance.
But I guess Julia has not the intent to support these devices.

[felipemarkson]

@oscardssmith Could you re-open this issue?
So, I can make some tests and demonstrate them here.

[oscardssmith]

OK. That said, be warned that erroring (or warning) is a breaking change that can't happen until 2.0, and is unlikely to be merged unless the error has no runtime cost (which it will).

[KristofferC]

There is nothing really to discuss here. You can take a look at #21600 for erroring for pow.

[felipemarkson]

But this shouldn't be an open issue to PRs references it?

[Seelengrab]

The reason this issue was (originally) closed so quickly is because there already has been a tremendous amount of discussion about this, e.g. on discourse. Julia just isn't at that stage of development anymore, where such things are considered.

[felipemarkson]

Okay, but this is a problem and should be documented somewhere.
Not for discussion, but so other people can consult it and learn about it.

[Seelengrab]

Okay, but this is a problem and should be documented somewhere.

This is already prominently answered/documented, in the FAQ of the documentation.

[felipemarkson]

Ok, thanks!

I'm closing this issue since this discussion was already done and is documented in the Integers Documentation and in the FAQ of the documentation.