Empty doc tests should be linted

[phrohdoh]

Run tests for https://play.rust-lang.org/?version=stable&mode=debug&edition=2018&gist=8ac29ea8121d53906e7a77cd01ccbe70 and you'll get the following diagnostic:

warning: invalid start of a new code block
 --> /playground/src/lib.rs:1:1
  |
1 | //! Hello
  | ^^^^^^^^^

But to be as helpful as possible the diagnostic should, if possible, point to the start of the invalid code block.

[euclio]

This error was removed in #60140 because I didn't think it could actually be produced. Looks like it was possible to trigger by writing an empty doc test. The error message is misleading, though. It's a valid code block, just empty.

On master, this code produces a passing doc test now. IMO, that is the correct behavior. There should probably be a test for it.

[phrohdoh]

There should probably be a test for it.

I agree completely.

Would this mean re-introducing this message but improving the message (replacing invalid with empty or something similar)?

I believe that is the best course of action as warning on an empty code block is completely valid and desired IMO.

[euclio]

The original code only caught this incidentally. I think this check would make more sense as a rustdoc lint instead of a hard warning. Could be added to check-code-block-syntax or a new pass entirely.

[euclio]

@phrohdoh could you edit this issue to reflect the current status? It should be something like "Empty doc tests should be linted".

[tommilligan]

@euclio @phrohdoh I'd like to take this issue - should have a PR this week for review.