Semicolon does not parse in match types

[neko-kai]

Compiler version

3.0.1

Minimized code

def Choice = (_: Any) match { case Int => Long; case Long => Int } // ok

type Choice[A] = A match { case Int => Long ; case Long => Int } // error

https://scastie.scala-lang.org/popPHRHGRTmfZ9n41Un3fQ

Output

'}' expected, but ';' found

Illegal start of toplevel definition

Expectation

Expected the semicolon to separate match type cases the same as for value matches. Couldn't find a previous discussion on this, so I assume this is not intended?

[odersky]

Looks like a bug, yes.

[odersky]

Correction. The error is according to the syntax. It's

MatchType         ::=  InfixType `match` <<< TypeCaseClauses >>>
TypeCaseClauses   ::=  TypeCaseClause { TypeCaseClause }
TypeCaseClause    ::=  ‘case’ InfixType ‘=>’ Type [nl]

No room for semicolon here. Why does it work for term match? We have:

MatchClause       ::=  ‘match’ <<< CaseClauses >>>
CaseClauses       ::=  CaseClause { CaseClause }
CaseClause        ::=  ‘case’ Pattern [Guard] ‘=>’ Block

Here it's the Block that can end in a semicolon, which you can verify by looking at the rules of https://docs.scala-lang.org/scala3/reference/syntax.html.

Generally, it's a good idea to consult that syntax before speculating whether something is a bug or not. That document is the final arbiter to decide whether something is correct or not. If we decide to change a parsing behavior we also have to change that document, and often suggested changes fail not because we could not hack the parser but because we cannot come up with a clear context free syntax to describe them.

The syntax document is not perfect, but the more eyes are looking at it the better it will get.

[odersky]

So the real question is: Should we change the language to allow this. Which means: Make the decision that this is the right thing to do, change the syntax document, and change the parser. I leave that to others as an exercise.

[odersky]

To come back to the example: The idiomatic style is to leave semicolons out in both cases. But maybe the "least surprises" strategy would be to allow it in both cases since we cannot forbid it in the term match case.

[neko-kai]

But maybe the "least surprises" strategy would be to allow it in both cases since we cannot forbid it in the term match case.

I agree, but I was also surprised because I've had an intuition that semicolons and newlines were usually interchangeable as separators. The semicolon in a term match being a property of the block and not a separator on its own wasn't something I thought about.

The idiomatic style is to leave semicolons out in both cases.

I agree, but the reason I've tried this is because that's the only way to write a match in one line in REPL.

[dwijnand]

Here it's the Block that can end in a semicolon, which you can verify by looking at the rules of docs.scala-lang.org/scala3/reference/syntax.html.

I actually can't verify that, as it looks like the last BlockResult can be preceded by semicolon-terminated statements, but the BlockResult can't end with a semicolon?

In my mental model a semi-colon is like an escaped newline, and clearly newlines are allowed between case clauses.

The idiomatic style is to leave semicolons out in both cases.

I'm surprised you consider it idiomatic because I think it's little known and used, and probably considered confusing and unreadable to not separate cases with a semicolon. Now that I know it's possible I can't help myself from using the short-hand in one-line REPL snippets, but I think I avoid doing so in committed code. I was even considering suggesting to avoid the confusing in match types..