[ fix ] Backticked names cannot be used as prefix operators

[spcfox]

Description

Fixes #3550

Self-check

• If this is a fix, user-facing change, a compiler change, or a new paper
  implementation, I have updated CHANGELOG_NEXT.md

[gallais]

Is there any chance we could have a more helpful message suggesting the use of
a section? The following is valid:

(`minus` 5)

[spcfox]

Is there any chance we could have a more helpful message suggesting the use of a section? The following is valid:

(`minus` 5)

Not sure if this error can be easily added during parsing. I couldn't get it to work.

[spcfox]

@andrevidela, could we add this kind of error to the parser? Perhaps we shouldn't modify the parser itself and instead throw the error during desugaring?

I think it’s better to make it a parsing error, but I couldn't figure out how to add an informative error message.

[andrevidela]

In my experience, it's easier to catch this kind of error during desugaring, we have more contextual information than during parsing and it keeps the parser simple.

It also keeps the parse tree semantically simple, if we catch the backtick misuse at parsing time, we still produce a value of type OpStr which could be backticked at desugaring time, but since the error would have already been thrown by then it would be an impossible value to have. This would force us to handle a case that we know cannot happen. I'd rather have this handling code in the desugarer such that we're not saddled with a value of OpStr that doesn't faithfully represent the space of possible values accurately.

[spcfox]

Ok, now the error is handled during desugaring

[gallais]

Should the section not be

(`Elem` _) . fst

?

[spcfox]

Should we be guessing the parenthesis placement here? Both cases could potentially be correct

[spcfox]

The internal error occurred in the shunting algorithm, which reorders operators according to their precedence. That’s why I added this check before it. We could potentially run it, but I’m not sure it would be helpful

[gallais]

This application has a different meaning as _ is now for the first argument rather than the second one.

[spcfox]

We don’t know what the user intended. In the first case, I'm suggesting the most likely option — a section. But it’s also possible that the backticks were just a typo.

[andrevidela]

yeah we can't run the typechecker at this stage to know which one is the one that the user meant. I think this is fine as the user will probably try both and pick the one that works