Change in behavior of `include!` in doctests on nightly

[sgrif]

I haven't had a chance to create a minimal repro case, but I'm seeing these in Diesel's builds: https://travis-ci.org/diesel-rs/diesel/jobs/252088111#L824. The issue only occurs on nightly, and it seems to only affect doctests which are more than 2 directories deep.

[sgrif]

Any chance I can get some eyes on this? It's going to become a regression in beta soon.

[Eijebong]

I tried bisecting but ran in some issues with rust not being able to compile anything that had a proc_macro in it after the first iteration.

error: libproc_macro-19bb163a3bc62cba.so: cannot open shared object file: No such file or directory
  --> /home/eijebong/.cargo/registry/src/github.com-1ecc6299db9ec823/dotenv-0.10.1/src/lib.rs:11:1
   |
11 | extern crate derive_error_chain;
   | ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^

error: Could not compile `dotenv`.

EDIT:
Oh and if I fix this by using LD_LIBRARY_PATH, I get

error: proc-macro derive panicked
  --> /home/eijebong/.cargo/registry/src/github.com-1ecc6299db9ec823/dotenv-0.10.1/src/lib.rs:22:17
   |
22 | #[derive(Debug, error_chain)]
   |                 ^^^^^^^^^^^
   |
   = help: message: index out of bounds: the len is 0 but the index is 4294967180

error: Could not compile `dotenv`.

[Mark-Simulacrum]

@sgrif Is there a chance you or someone else could provide a more minimal example? Preferably something that would work with straightforward rustdoc invocation...

[Eijebong]

@Mark-Simulacrum Here is a repo https://github.com/Eijebong/repro_rust_43153

cargo test --doc passes on stable but doesn't on nightly (and beta for some reasons, I have no idea why diesel tests are passing...)

EDIT: They're not passing on beta anymore.

[sgrif]

Thanks for getting a minimal repro @Eijebong

[Mark-Simulacrum]

Bisection shows 4d526e0 - #40939, cc @jseyfried, @alexcrichton -- I'm not sure if this bisection is entirely accurate, but it does seem potentially related.... I'm going to attempt to look into fixing, but don't count on me.

[jdm]

This looks like a similar symptom ("No such file or directory") to the warning I reported in #43371.