capping lints does not work for rustdoc

[kentfredric]

As I mentioned in #66622 , there's a fun thing that happens:

--cap-lints warn can be over-ridden, apparently by attributes in macros, when run under rustdoc, which pretty much defeats the points of --cap-lints

With rand 0.5.6:

RUSTFLAGS="--cap-lints warn" cargo test --doc  --verbose
... 
 test src/prng/chacha.rs - prng::chacha::ChaChaRng::new_unseeded (line 115) ... ok
test src/prelude.rs - prelude (line 17) ... ok
test src/rngs/jitter.rs - rngs::jitter::JitterRng (line 72) ... FAILED
test src/lib.rs - sample (line 1020) ... ok
test src/rngs/adapter/read.rs - rngs::adapter::read::ReadRng (line 35) ... ok
test src/rngs/jitter.rs - rngs::jitter::JitterRng::new_with_timer (line 318) ... ok
...
failures:

---- src/rngs/jitter.rs - rngs::jitter::JitterRng (line 72) stdout ----
error: trait objects without an explicit `dyn` are deprecated
  --> src/rngs/jitter.rs:79:33
   |
10 | fn try_main() -> Result<(), Box<Error>> {
   |                                 ^^^^^ help: use `dyn`: `dyn Error`
   |
note: lint level defined here
  --> src/rngs/jitter.rs:71:9
   |
2  | #![deny(warnings)]
   |         ^^^^^^^^
   = note: `#[deny(bare_trait_objects)]` implied by `#[deny(warnings)]`

error: aborting due to previous error

It seems that, this line, in src/lib.rs takes precedence over lint-capping :(

#![doc(test(attr(allow(unused_variables), deny(warnings))))]

[kentfredric]

To add insult to injury, I've discovered this also happens for lints that are defaulted on, as demonstrated by byteorder 0.5.3

running 6 tests
test src/lib.rs - ByteOrder (line 114) ... FAILED
test src/lib.rs - ByteOrder (line 104) ... ok
test src/lib.rs -  (line 15) ... ok
test src/lib.rs -  (line 28) ... ok
test src/new.rs - new::ReadBytesExt (line 15) ... ok
test src/new.rs - new::WriteBytesExt (line 143) ... ok

failures:

---- src/lib.rs - ByteOrder (line 114) stdout ----
error: literal out of range for `i16`
 --> src/lib.rs:118:33
  |
7 | BigEndian::write_i16(&mut buf, -50_000);
  |                                 ^^^^^^
  |
  = note: `#[deny(overflowing_literals)]` on by default

error: literal out of range for `i16`
 --> src/lib.rs:119:13
  |
8 | assert_eq!(-50_000, BigEndian::read_i16(&buf));
  |             ^^^^^^

error: aborting due to 2 previous errors

Couldn't compile the test.

failures:
    src/lib.rs - ByteOrder (line 114)

test result: FAILED. 5 passed; 1 failed; 0 ignored; 0 measured; 0 filtered out

Nothing I can do to resolve this other than patching the source. ( this issue is fixed in newer byteorders, but due to some deep dependency chain, the old byteorder is still required, and so as a linux vendor, we desire tests to run, but this failure isn't a useful one )

[jyn514]

Duplicate of #67533 - cap-lints is currently ignored for doctests.

[kentfredric]

Technically I think the other bug is the dupe, being newer, but not sure what standard conventions for this are on github

[jyn514]

Ah well, you're right and I just happened to see the other first 😆 feel free to comment there if it's missing info you had here.