Should unnecessary wildcard pattern be reported as a warning by all tools

[chloestefantsova]

Consider the following program.

test(dynamic x) => switch(x) { int y && _ => 1, _ => 0 };

Currently it's accepted by the CFE, and the Analyzer emits the following warning.

  warning • Unnecessary wildcard pattern. • /tmp/asdf.dart:1:41 • unnecessary_wildcard_pattern 

I wasn't able to find that warning message in the specification, but I want to bring more agreement into how the tools report the errors. I created a CL to make the CFE report the same warning: https://dart-review.googlesource.com/c/sdk/+/311221

Should I land my CL? Should the spec be updated?

[chloestefantsova]

/cc @johnniwinther @scheglov @eernstg

[eernstg]

Isn't it a general rule that the CFE only reports errors?

I'd expect warnings (and hints/lints/...) to be emitted by the analyzer, but the CFE would not have to do it.

It looks like the CFE doesn't actually emit any of the warnings that I can immediately come up with. For example 1?.isEven is a warning by the analyzer, but the CFE says nothing. Similarly for a couple of cases with dead code.

[johnniwinther]

We've never officially stated that warnings mandated by the spec are optional. The CFE still emits warnings about unnecessary ?., for instance. It would nice to state this, once and for all, so that we can have the clear story that only errors are required and warnings (including hints/lints/...) are only optional and can always be ignored using the // ignore: comments.

@leafpetersen

[eernstg]

On April 20th 2022, the language team expressed the tentative intention to remove all non-errors from the language specification. So warnings, hints, lints (and do we have even more kinds?) would be specified as, say, analyzer or linter specific behavior, and the CFE could choose to ignore all those situations rather than emitting anything.

The only missing part is a firm and explicit decision to proceed, and possibly sorting out some details.

[johnniwinther]

... and we are still (impatiently) awaiting this "firm and explicit decision"!

[eernstg]

I think it's a pretty safe bet to say that the analyzer produces compile-time errors and warnings (whatever they are or were called), and all other tools produce only compile-time errors. I know we do have a couple of situations which are characterized as a warning in a language specification document (a couple of feature specifications have this, I think). I added it to the language meeting agenda for this Wednesday.