Disallow default as a constant pattern.

[gafter]

The LDM has tentatively decided to disallow a default keyword (literal) as a constant pattern, in order to reduce confusion with "default" cases in switch statements and expressions. A programmer can always write null, 0, or '\0' instead. This is technically a breaking change.

In a switch statement, the case label
case default:
is proposed to produce an error something like
error CS8313: A default literal 'default' is not valid as a pattern. Use another literal (e.g. '0' or 'null') as appropriate. To match everything, use a discard pattern '_'.
We already produce approximately this diagnostic, but as a warning.

In an is-pattern expression and other places where patterns appear
e is default
is proposed to produce an error something like
error CS8405: A default literal 'default' is not valid as a pattern. Use another literal (e.g. '0' or 'null') as appropriate. To match everything, use a discard pattern '_'.

This will need review from the compatibility council.

[jcouv]

Just to clarify the proposal is to produce an error instead of the current warning: warning CS8313: Did you mean to use the default switch label (`default:`) rather than `case default:`? If you really mean to use the default literal, consider `case (default):` or another literal (`case 0:` or `case null:`) as appropriate.

[gafter]

Yes. Clarified above.

[jcouv]

Should we make this change in 15.6 or 15.7 instead, to reduce the impact of breaking change? Default expressions where introduced in C# 7.1.

[gafter]

That is a question for the compat council.

[jcouv]

I'm not on that thread ;-)
I think we should do this in 15.6. Thanks

[gafter]

For 15.6, the messages are modified as we do not have a discard pattern. For the switch, we recommend the default: label. For pattern matching we recommend var _.

[ericwj]

I disagree.

I see no reason why it is necessary to break my code after fixing case default: with case (default): based on the warning previously issued.

CS8313 re-introduces code edits that were happily a thing of the past changing the type of a thing.

And the remark If you intended to write the default label use 'default:' without 'case'. is totally lame. I don't mean default: I mean either case 0: or case null: or ... depending on the actual type.

You won't accidentally write case (default): when you meant default: imho.

[YairHalberstadt]

The suggested solution will of course not work for generic types without a class or a struct constraint

[alrz]

is it too late to also disallow goto case default?