On Mon, 17 May 2021 16:54:07 GMT, Hannes Wallnöfer <hann...@openjdk.org> wrote:
> The primary problem was that key `compiler.err.dc.ref.bad.parens` had the
> wrong message "')' missing in reference". That error message is not used when
> the right parenthesis is missing, but when it occurs before the end of the
> signature. The missing parenthesis case is caught earlier by the parens
> balancing code in `DocCommentParser#reference` and uses message
> `compiler.err.dc.unterminated.signature` ("unterminated signature").
>
> I changed the message for `compiler.err.dc.ref.bad.parens` to "unexpected
> parenthesis".
>
> A secondary problem was that `DocCommentParser#reference` also threw
> "unterminated signature" when there were too many closing angle brackets or
> parentheses. These cases are now passed through
> in`DocCommentParser#reference` so that `ReferenceParser#parse` can throw a
> more appropriate error ("unexpected parenthesis" or "unexpected input").
>
> Here's a list of what references now cause what error messages:
>
> - `#foo(int` `unterminated signature`
> - `#foo((int))` `unexpected parenthesis`
> - `#foo(int))` `unexpected parenthesis`
> - `#foo(int)x` `unexpected parenthesis`
> - `F<T` `unterminated signature`
> - `F<T>>` `unexpected input`
>
> I added two new tests for the cases with too many closing brackets/parens
> that used to fail with "unterminated signature".
Thanks for reviewing, Pavel!
> 1. I think this error is a counterintuitive:
> ```
> #foo(int)x
> error: unexpected parenthesis
> ```
>
>
> It's `x` that is unexpected, not the parenthesis. It looks correct here, but
> only coincidentally:
>
> ```
> #foo(int))
> error: unexpected parenthesis
> ```
Yes, you can look at it both ways, either there is a closing parenthesis where
it shouldn't be, or there is text following a correctly placed closing
parenthesis. We tend to see one or the other depending on whether the text up
to the closing parenthesis looks like a proper signature, but it's not as clear
to the parser.
I still like the more specific message. We can use a more generic "unexpected
text" message, but I think it is less useful.
> 1. The latter case in your list, `F<T>>`, yields "unexpected text", not
> "unexpected input".
That's my mistake, the resource key ends with `unexpected.input` but the
English message is "unexpected text".
> 2. Why is there an asymmetry between wording for angle brackets and that
> of parenthesis?
See my answer to point 1, I like the message to be as specific as possible.
-------------
PR: https://git.openjdk.java.net/jdk/pull/4068
