container.len() == 0

[rust-lang]

container.len() == 0

[lambda-fairy]

As a first approximation, we can warn when the self type is in collections.

If it's easy enough to check for an .is_empty() method, we can do that instead.

[llogiq]

So, I now know how to get at the type (though I'm not sure what the exact contents would be), so how do we find out if the type has an is_empty() method?

[Manishearth]

I assume you go through the method map? Everything regarding type check and resolution is done through the tcx (type ctxt). Check out middle::ty for a ton of helper functions.

[llogiq]

I know how to get the type definition (via expr_ty), but I'm a bit lost on how to go from there to the MethodMap. Do I construct a MethodCall via its expr and use that to query the MethodMap? Also what do I do with the MethodCallee once I have it?

[Manishearth]

I'm not sure. Iterating through the tcx.impl_or_trait_items entry for the defid of the given type might work.

You'll have to explore middle::ty and ctxt to get this I guess :)

[llogiq]

I've run into some wrestling with typeck, but at least I feel like I'm on the right track.

I'll see once the code compiles, anyway. 😄 However, it's getting late, so I'll resume tomorrow.

[llogiq]

Now it compiles, but the type lookup fails for the simple case let x = [1, 2]; if x.len() == 0 { … }. I cannot figure out how to get from the DefLocal to a slice of its ImplOrTraitItemIds. Probably the whole approach is wrong.

[Manishearth]

I've never really played with this so you're basically on your own here, sorry :) Poking people in #rust-internals may help

[llogiq]

The problem is I'd either need to copy large swaths of typeck and patch rustc to allow me to access some private fields/methods, or just wait until the core team makes this available for lints.

The latter part opens up some of my time to try and implement some other things, so I guess I'll wait.

[llogiq]

The lint in #60 basically encodes that every impl that has .len () should have .is_empty(), so we can get by without method lookup at the cost of possible false positives.

This is intended as a stop-gap solution, but may turn out to be good enough should we surmise that the cost of method lookup is too high.