Use const_eval instead of constant for more lints?

[llogiq]

• eq_op won't join here because we expressly only want syntactically equal expressions
• bit_mask uses it
• float_cmp uses it
• approx_const doesn't even need constant because we only want to match literals
• zero_div_zero might probably be able to use it
• min_max might use it (but we currently bail on refs, so it's debatable whether it should)
• loops could benefit from it
• identity_op shouldn't use it because it bails on refs and it's so low-payoff that a full const_eval isn't worth it

What do you folks think?

[Manishearth]

We don't have a way of bypassing const resolution in const_eval though (do we?). We want to resolve constant expressions without doing const lookup because of cfgs and tweaking.

[llogiq]

Agreed. We could easily simplify consts.rs to work without a Context, though. const_eval should work in the other cases.

[Manishearth]

Sounds good to me