Decide if TokenMissingError and MismatchedDelimiterError should have end_line/column or begin_line/column

[josevalim]

And whatever we choose should be passed down to the diagnostics and we need to document these fields.

The big question is: should the line/column point by default to where the delimiter starts or where it is missing/mismatched?

Perhaps this doesn't matter for IDEs. IDEs will receive both line+column PLUS the span, so they can always highlight the whole bit. If this is the case, then I would have line/column be the starting position and include end_line/end_column, so it is closer to the diagnostic.

/cc @lukaszsamson

[lukaszsamson]

LSP diagnostics are based on ranges and so far I assumed the following. I omit the mapping from UTF8 1-based to UTF16 0-based columns for simplicity

1. if position is 0 I map it to range {1, 1} - {1, 1}
2. If position is pos_integer line I find the first nonwhitespace char and {line, first_char}-{line, first_char}
3. If position is tuple {line, column} I return {line, column}-{line, column}
4. If position is range I return exactly that range

Besides the main diagnostic range, message and severity there are 3 things we can use to make them more helpful.

1. A collection of tags https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/#diagnosticTag. VSCode marks unused and deprecated warnings specially. So far I use regex for extracting that.
   Notice gray font on unused and strikethrough on deprecated

2. A collection of related info https://microsoft.github.io/language-server-protocol/specifications/lsp/3.17/specification/#diagnosticRelatedInformation. This makes it possible to make the diagnostic in line 5, and point to line 1 starting the unclosed token and line 8 when document ends. You can click on the related info and it will navigate to relevant position in the editor. Here I extract it from end_line, closing_delimiter or via regex from hint in description

BTW here the new diagnostics look rather ugly as the UI does not use fixed width font

3. Error code and URL link to description - no such thing in elixir right now

[josevalim]

Thanks. You will have the exception now, so hopefully you can extract all of that more reliably.

[viniciusmuller]

BTW here the new diagnostics look rather ugly as the UI does not use fixed width font

@lukaszsamson apart from the font, the problem mainly seems to be the fact that this vscode thing is trimming all the leading whitespace in the error lines, making it unaligned

I don't use vscode but I can try it later as well, but it would be cool if you tried how it looks like with a language such as rust, as it also have these leading whitespaces in order to make padding for the line numbers (although ours are a bit more indented to the right)

[lukaszsamson]

@viniciusmuller that's a known issue with vscode. The same diagnostics are rendered properly in other places microsoft/vscode#198981

[viniciusmuller]

Well, at least it looks very nice on the error hover pop-up