redundant_semicolon and clippy::no_effect tripped without emitting line numbers

[brson]

Using nightly-2019-08-26.

In this TiKV PR we upgraded the compiler. Our tidb_query component tripped the redundant_semicolon lint (and seemingly in turn clippy::no_effect) but rustc/clippy did not tell us the line number.

I couldn't find the code that triggered the lint and had to just allow both for the whole crate.

This issue has been assigned to @nathanwhit via this comment.

[brson]

It can be reproduced by cloning that PR, removing the commit that adds the allows, and running make clippy.

[estebank]

Introduced in #62984. It has caused similar effects in other cases like #63947. CC #63679.

[Centril]

cc @nathanwhit @varkor

[nikomatsakis]

Check-in from compiler team:

Marking as P-high " as this is a regression, though admittedly one without particularly serious consequences (the lint can always be allowed).

@nathanwhit or @varkor, perhaps one of you can take a stab at this? Please claim the issue if you do start working on it (even if it's just to write mentoring instructions).

[nikomatsakis]

Also, it'd be great to get a minimized version of the problem. Therefore, marking as E-needs-mvce.

[nathanwhit]

I'll work on this. Looking into it a bit, I think it has to do with the lint's interaction with custom proc macro attributes. Specifically, when the item TokenStream passed to the attribute contains a redundant semicolon, every span in the TokenStream is the dummy span. This occurs in both this issue and #63947. It's definitely due to the changes to the parser, but I'm not sure why the changes would have that effect on the spans.
@rustbot claim