fix[codegen]: bounds check for signed index accesses

[charles-cooper]

What I did

fix #3762

How I did it

How to verify it

Commit message

Commit message for the final, squashed PR. (Optional, but reviewers will appreciate it! Please see our commit message style guide for what we would ideally like to see in a commit message.)

Description for the changelog

Cute Animal Picture

[trocher]

Would it make more sense to prevent indexing with signed types entirely or is it something that should be possible?

[charles-cooper]

Would it make more sense to prevent indexing with signed types entirely or is it something that should be possible?

i think people want to use signed integers without conversion, a common usage is like

    for ix: int256 in ...:
        ret += array[ix]

and it would be annoying to have to write it with a convert(), like

    for ix: int256 in ...:
        ret += array[convert(ix, uint256)]

[trocher]

The order of the args of clamp2 seems incorrect and should probably be instead 0, ix, bound, ix.typ.is_signed

Additionally, even with the other order, the check is incorrect for signed integer types when the array's size is greater than or equal to 2**255. [sle, arg, hi] performs a signed comparison but hi, the length of the array is an unsigned integer interpreted as a signed integer.

For example, compiling:

arr: public(uint256[max_value(uint256)-1])

@external

def foo():
    self.arr[1] = 3

raises

vyper.exceptions.StaticAssertionException: assertion found to fail at compile time. (hint: did you mean `raise`?) [assert,
  [and,
    [sge, 1 <1>, 0],
    [sle,
      1 <1>,
      115792089237316195423570985008687907853269984665640564039457584007913129639934]]]

[charles-cooper]

The order of the args of clamp2 seems incorrect and should probably be instead 0, ix, bound, ix.typ.is_signed

ah sorry, i should have marked as draft until i have tests passing

Additionally, even with the other order, the check is incorrect for signed integer types when the array's size is greater than or equal to 2**255. [sle, arg, hi] performs a signed comparison but hi, the length of the array is an unsigned integer interpreted as a signed integer.

For example, compiling:

arr: public(uint256[max_value(uint256)-1])

@external

def foo():
    self.arr[1] = 3

raises

vyper.exceptions.StaticAssertionException: assertion found to fail at compile time. (hint: did you mean `raise`?) [assert,
  [and,
    [sge, 1 <1>, 0],
    [sle,
      1 <1>,
      115792089237316195423570985008687907853269984665640564039457584007913129639934]]]

hmm, so i guess it must be sge 0 but lt hi. nice catch, thank you!

[codecov-commenter]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 85.21%. Comparing base (b53a020) to head (635a3d7).
Report is 3 commits behind head on master.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #3817      +/-   ##
==========================================
+ Coverage   85.15%   85.21%   +0.05%     
==========================================
  Files          92       92              
  Lines       13907    13912       +5     
  Branches     3116     3117       +1     
==========================================
+ Hits        11843    11855      +12     
+ Misses       1571     1564       -7     
  Partials      493      493              

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[trocher]

LGTM :)