Correctly parse checkUsage when a multiline warning is encountered

[AshesITR]

fixes #507

• implement fix
• add regression test
• add NEWS bullet

[AshesITR]

Diff coverage needs a test for declared_globals to pass.

[infotroph]

I've only tried this fix on one (very messy) file so far, but lintr is exiting with a new error before it gets to the multiline expression I wanted to test:

Error in rep.int(character, length) : invalid 'times' value
Calls: <Anonymous> ... print.lint -> cat -> highlight_string -> fill_with -> paste0
Execution halted

When linting with the released version of lintr the next message emitted is a "Lines should not be more than 80 characters" -- if rep.int doesn't immediately tip you off I'll try to narrow this down to a reprex tonight.

[AshesITR]

@infotroph

This seems vaguely related to #742.
Can you narrow down which linter produces NA column_numbers?
as.data.frame(lint(...)) should help you find the issue.

[AshesITR]

We're still getting NA locations if the info doesn't match on the first line of the checkUsage info.
I'll work on making the regex more robust.

[infotroph]

Now working without error and with correct messages for the case I reported above, but for completeness I can report narrowed the offending lines down to an if-else with unbound names in both conditions:

fn <- function() {
  if (badname) {
  } else if (badname2) {
  }
}

Again, as of 9209e7a lintr is correctly warning on both lines and not erroring.

[AshesITR]

Thanks for checking this out @infotroph.
@MichaelChirico PTAL, I think it's ready to merge now.

[MichaelChirico]

the test here is a symbol name that's not matched by our regex? so that the linter has to default to where the expression starts

[AshesITR]

The test causes the boundary, res$name, boundary regex to not match against the source code, yes.