small improvements to interning

[Eh2406]

In #5121 Eh2406@411355a I tried to use the InternedString In more places, but test failed so that commit was backed out. This PR is an attempt to redo that in tiny steps, so as to track down exactly what caused the error.

The first commit hear, is just some housekeeping. This should still be green.

[rust-highfive]

r? @alexcrichton

(rust_highfive has picked a reviewer for you, use r? to override)

[Eh2406]

This second commit is the smallest change I have found that reproduces the test fails. It took several trys to find something this small.

I still don't see why it is problematic. @alexcrichton any thoughts on how this is wrong?

[Eh2406]

I don't know what is going on with nightly and beta, that is not the error I see locally.

[alexcrichton]

Aha! I believe the problem is that the hash value for an InternedString is making its way into the fingerprint for a crate which means that if it changes the crate is recompiled. That I think means that this would always recompile all the crates!

Perhaps the Hash implementation should look at the contents rather than just the raw pointer?

[alexcrichton]

Ah and for nightly/beta feel free to just remove the #![deny(unused)]

[Eh2406]

Thanks! That seems to have fixed it. It was nice to have constant time hashing, but I think the inconsistent hash is too much of a foot gun to leave in.

[alexcrichton]

Closed?

[Eh2406]

Ya you identified the bug, which fixed the more ambitious versions. When I have had a chance to polish one of them, it will come in as a new PR.

[alexcrichton]

Aha I see!