Improve message for misplaced using

[som-snytt]

Offer extra feedback when soft keyword using is found in identifier position, because it is a common beginner mistake.

def f(i: Int, using j: Int) = i + j

Skipping to the intended j may also reduce subsequent "not found" errors.

In Vulpix, show actual errors when no errors are marked. It is unlikely that this is "TMI" for the developer. Automatic -rewrite is left for another day.

Fixes #15614

[odersky]

Otherwise LGTM

[som-snytt]

Thanks again for the quick review. A little bit of experienced feedback goes a long way.

Following is note-to-self:

Updated without tokens for using, erased. Worth noting that using is enshrined as a standard name, but tracking down usages is still tricky. Similarly, I wondered why then isn't rewritten. What is handled where? Every special case increases complexity.

Modified the verbose advice to say an error makes a symbol unavailable. That happens frequently during a refactor, where fixing one site results in further errors in the next run.

Internet suggests "misspelt" is becoming archaic. Also, although I love forms such as "spelt", "spelt" also means "wheat", and we know overloading is evil.

Edit: the message says "soft token" instead of "soft keyword", and on the ticket Bjorn asks for more specific advice about "mixing regular and using params." Maybe just say "the keyword must appear at the beginning of the param clause".

#15614 (comment)

[bjornregnell]

I think the error messages and explaining texts read good.

[som-snytt]

@odersky Please approve based on Bjorn's approval.

[bjornregnell]

I may be missing some context here, but why closed instead of merged? @som-snytt

stale

[som-snytt]

rebased, squashed, removed some words during ugly merge.

[som-snytt]

@bjornregnell Could you re-review and perhaps get official sanction?

The previous effort included heavier edit of the explain text, which I omit here because of the heavy merge.

The previous effort included handling erased, which is no longer needed.

The only trick in parser is error recovery in skipping bad token to the presumably intended ident.

I'll also update the introductory comment above.

[bjornregnell]

LGTM
Thanks for nice work in increasing quality of life on planet Earth @som-snytt
I think this is ready to be merged @odersky - who should do that?

[som-snytt]

Reopening to reconsider this forgotten bit.

[Gedochao]

@som-snytt is this still relevant? Do you plan to work on this in the near future, or should we close it?

[som-snytt]

I'm not sure it was ever relevant, but I have refreshed the PR.