RFR: 8303374: Compiler Implementation for Primitive types in patterns, instanceof, and switch (Preview) [v3]

[Aggelos Biboudis]

On Thu, 5 Oct 2023 08:14:53 GMT, Aggelos Biboudis <abimpoudis at openjdk.org> wrote:

>> Aren't there cases here where the type of the expr is unconditional w.r.t. the type of the pattern (and so, no test is required) ? Perhaps the case I'm thinking of is dealt with one of the ifs at the beginning of this method.
>> 
>> I have to say I'm a bit confused by how this method is arranged. From a logical perspective I would expect first and foremost to test if the pattern type is a primitive, because that's the only case where Lower needs to do anything (right?).
>> 
>> Then you can have two cases: the expression has a reference type, in which case you need null check and unboxing. Or the expression has a primitive type.
>> 
>> In both cases you can be in a situation where the conversion is provably exact, so no need to call the exactness routine - or you can be in an inexact situation where a runtime test is required.
>> 
>> While I'm sure that is what happens in the above code, I think perhaps the various tests could be arranged in a way to make the above structure pop out a little more?
>
> Refactoring attempt No. 1 complete. I need to recheck the completeness of `checkUnconditionallyExact`. In a following commit.

`checkUnconditionallyExact` and dominance for constants were adjusted accordingly in https://github.com/openjdk/jdk/pull/15638/commits/c041d208a6fc6beaf785e8a400815fb4a7d08e78 🎉

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

PR Review Comment: https://git.openjdk.org/jdk/pull/15638#discussion_r1347473254