Add SafeCmp lib - a library to safely compare felts

[kaliberpoziomka]

🧐 Motivation Currently using assertion functions in math.cairo and comparison functions in math_cmp.cairo for felts is confusing, and therefore insecure. Those assertion/comparison functions can be split into two groups:

• first: assert_le, assert_lt, assert_nn, is_le, is_lt, is_nn
• second: assert_le_felt, assert_lt_felt, is_le_felt, is_lt_felt

Functions in the first group should only be used in Uint256 context - so with Uint256.low and Uint256.high, since they operate in range [0, 2**128), and Uint256 members are in the same range [0, 2**128). However, developers tend to use functions from the first group outside the context of Uint256, this creates a lot of insecure scenarios. To provide few examples of unexpected behavior:

• is_le(1, 2**128 + 2) returns false FALSE
• is_nn(2**128) returns FALSE (even though negative numbers are felts in range (P/2, P-1])

Functions from the second group return proper results, however they assume that arguments are unsigned felts, i.e. represent numbers in range [0, P). This creates confusion when working with signed felts, i.e. in range (-P/2, P/2). Currently there are no assertion/comparison functions that would allow safely compare two signed felts or assert that the number is non-negative.

📝 Details SafeCmp library would contain functions for safe assertion and comparison of unsigned felts and signed felts. This would be done by:

• providing functions that would allow for safe assertions/comparisons of signed felts
• providing names of functions that would clearly indicate in what range felt arguments are expected (signed/unsigned)

[EvolveArt]

Really supportive of this initiative.

With Cairo 1.0 all of this might get more clearer as a native typing system is added into the language and we have a safe intermediate representation. However I'm pretty sure the issue you're mentioning is currently present on a few contracts out there as we have no fuzzing tools open source (do we?) to check this out.

[martriay]

Closing this since it will not be relevant anymore after the ongoing Cairo 1.0 migration. If you think this is a mistake, feel free to open a new issue.