Regression in finding crates during doc tests

[dtolnay]

Problem

A couple of my crates' builds started failing at nightly-2019-10-03. I don't have a chance tonight to dig into what is going wrong, but I figured I would report it early since this appears to be a regression.

#7417 looks possibly related.

Steps

cargo clean && cargo test in either https://github.com/serde-rs/json or https://github.com/dtolnay/quote. They both work on nightly-2019-10-02 but fail on nightly-2019-10-03, in different ways.

serde_json: https://travis-ci.org/serde-rs/json/jobs/592841446

...

   Doc-tests serde_json
error[E0463]: can't find crate for `serde_derive` which `serde` depends on
   --> /home/travis/build/serde-rs/json/src/lib.rs:325:1
    |
325 | extern crate serde;
    | ^^^^^^^^^^^^^^^^^^^ can't find crate

quote: https://travis-ci.org/dtolnay/quote/jobs/592846455

...

---- src/ext.rs - ext::TokenStreamExt::append_all (line 20) stdout ----
error[E0460]: found possibly newer version of crate `proc_macro2` which `quote` depends on
 --> src/ext.rs:19:1
  |
2 | extern crate quote;
  | ^^^^^^^^^^^^^^^^^^^
  |
  = note: perhaps that crate needs to be recompiled?
  = note: the following crate versions were found:
          crate `proc_macro2`: /home/travis/build/dtolnay/quote/target/debug/deps/libproc_macro2-e61df99964fc712a.rlib
          crate `proc_macro2`: /home/travis/build/dtolnay/quote/target/debug/deps/libproc_macro2-e61df99964fc712a.rmeta
          crate `proc_macro2`: /home/travis/build/dtolnay/quote/target/debug/deps/libproc_macro2-37a3c8079ede86c5.rmeta
          crate `quote`: /home/travis/build/dtolnay/quote/target/debug/deps/libquote-9e0852a9faf16dfb.rlib

[ehuss]

Yea, it is definitely due to #7417. The trybuild test recursively invokes cargo which causes some crates to be rebuilt due to the addition of RUSTFLAGS. These partially recompiled crates are no longer compatible with the original run, but they have the same hash.

I don't think there has been much consideration made for running cargo within tests. Would it be possible for trybuild to use a dedicated target directory?

[dtolnay]

Thanks! That makes sense, sorry about the noise. I will fix trybuild to use an independent target directory.