On Wed, 19 Apr 2023 16:45:06 GMT, Vicente Romero <vrom...@openjdk.org> wrote:

>> Jan Lahoda has updated the pull request incrementally with six additional 
>> commits since the last revision:
>> 
>>  - Fixing infinite loop where a binding pattern is replaced with a binding 
>> pattern for the same type.
>>  - Reflecting review comments.
>>  - Fixing exhaustiveness for unsealed supertype pattern.
>>  - No need to enable features after error reported.
>>  - SwitchBootstraps.typeSwitch should not initialize enum classes.
>>  - A prototype of avoiding enum initialization.
>
> src/java.base/share/classes/java/lang/runtime/SwitchBootstraps.java line 279:
> 
>> 277:     }
>> 278: 
>> 279:     private static int lazyDoEnumSwitch(Enum<?> target, int startIndex, 
>> Object[] labels, MethodHandles.Lookup lookup, Class<?> enumClass, 
>> MutableCallSite callSite) throws Throwable {
> 
> can `doEnumSwitch` be folded into `lazyDoEnumSwitch`? just a suggestion, I'm 
> OK with either way just that now it is not clear that we need two methods 
> here. Also in `doEnumSwitch` and out of curiosity what example of user code 
> could hit this section:
> 
> 
>             if (label instanceof Class<?> c) {
>                 if (c.isAssignableFrom(targetClass))
>                     return i;
>             }
> 
> EDIT: nvm I found one example

The intent here is to avoid eager initialization of enums - so, until the enum 
is initialized, `lazyDoEnumSwitch` will be used (which, by itself, won't 
compute any results, except the result for `null`), which is then replaced with 
`doEnumSwitch` using the enum constants directly.

The enum switches support patterns as well, so something like:

enum E {A;}
E e = ...;
switch (e) {
     case E ee -> {}
}

is valid, and should trigger the part with Class labels.

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

PR Review Comment: https://git.openjdk.org/jdk/pull/13074#discussion_r1172623143

Reply via email to