RFR: 8268320: Better error recovery for broken patterns in switch

[Jan Lahoda]

On Thu, 10 Jun 2021 09:59:44 GMT, Rémi Forax <forax at openjdk.org> wrote:

>> Trying to improve error recovery in case the user wanted to use a type-pattern in the switch construct, but didn't put a binding variable name after the type name. E.g.:
>> 
>> $ cat test/langtools/tools/javac/diags/examples/PatternExpected.java
>> [snip]
>> class PatternSwitch {
>>     private void doSwitch(Object o) {
>>         switch (o) {
>>             case String: break;
>>             default: break;
>>         }
>>     }
>> }
>> $ javac --enable-preview -source 17  test/langtools/tools/javac/diags/examples/PatternExpected.java
>> test/langtools/tools/javac/diags/examples/PatternExpected.java:32: error: type pattern expected
>>             case String: break;
>>                  ^
>> Note: test/langtools/tools/javac/diags/examples/PatternExpected.java uses preview features of Java SE 17.
>> Note: Recompile with -Xlint:preview for details.
>> 1 error
>
> src/jdk.compiler/share/classes/com/sun/tools/javac/comp/Attr.java line 1748:
> 
>> 1746:                                     }
>> 1747:                                 } else if (!stringSwitch && !types.isAssignable(seltype, syms.intType)) {
>> 1748:                                     log.error(pat.pos(), Errors.ConstantLabelNotCompatible(pattype, seltype));
> 
> I wonder if in the long term it's not better instead of having several ad-hoc way to find the kind of switch it is,
> `types.isAssignable(seltype, syms.intType)`, `stringSwitch` or `pattype.constValue() == null && s.kind == TYP`
> to have an enum with 4 values INT_SWITCH, STRING_SWITCH, ENUM_SWITCH and PATTERN_SWITCH.

I apologize for a belated answer. It is my plan to try to categorize switches, but when considering error recovery and various desugaring stages, it is not trivial. Also, please note that some of the conditions are not a property of the switch, but rather a property of the case (like in this case - saying `case String:` instead of `case String s:` is more a property of the case, than the switch.

-------------

PR: https://git.openjdk.java.net/jdk/pull/4441