Handling of reserved identifiers and operators in the identifier parser

[sjakobi]

I've noticed that the identifier parser will accept e.g. 'case' or '..' as identifiers although both are reserved and can't be hyperlinked. This isn't much of a problem as the renamer will mark them as DocMonospaced instead of DocIdentifier but maybe there are arguments for correcting this nonetheless?

Relatedly, maybe haddock should emit a warning when it can't resolve an identifier and applies DocMonospaced?

[alexbiehl]

Good call. I have prepared haddock-api to emit renamer warnings already. Feel free to make a PR to output actual warnings.

[sjakobi]

Strangely, when I tried to use my patch for #819 to produce warnings for 'case' and '..', I didn't get any.

A bit of print-debugging revealed that e.g. 'case' was handled as a DocString "'case'". I don't understand why haddock-library's ParserSpec then parses them as DocIdentifiers. Any ideas?

[alexbiehl]

Maybe a difference between parseString and parseParas? Unlikely but an association I had.

[sjakobi]

Maybe a difference between parseString and parseParas? Unlikely but an association I had.

No, parseString and parseParas both parse reserved identifiers as DocIdentifiers.

[harpocrates]

Strangely, when I tried to use my patch for #819 to produce warnings for 'case' and '..', I didn't get any.

A bit of print-debugging revealed that e.g. 'case' was handled as a DocString "'case'". I don't understand why haddock-library's ParserSpec then parses them as DocIdentifiers. Any ideas?

This happens entirely on the haddock-api side in parseString and parseParas. Note that both of these have a call to P.overIdentifier (parseIdent d).

parseParas :: DynFlags -> Maybe Package -> String -> MetaDoc mod RdrName
parseParas d p = overDoc (P.overIdentifier (parseIdent d)) . P.parseParas p

parseString :: DynFlags -> String -> DocH mod RdrName
parseString d = P.overIdentifier (parseIdent d) . P.parseString

parseIdent calls out to GHC and returns Nothing for things that aren't valid identifiers according to the GHC parser. overIdentifier converts identifiers that were mapped to Nothing into DocString (here).

[sjakobi]

Thanks @harpocrates. I hadn't noticed that haddock-api also has functions called parseParas and parseString.

I'm wondering though if users of haddock-library shouldn't be able to expect the parser to handle the question of valid or invalid identifiers.