On Tue, 15 Oct 2024 17:34:53 GMT, Chen Liang <li...@openjdk.org> wrote:
> Currently, to efficiently check if a `ClassEntry`, such as one from an > `InvokeInstruction`, is of a particular class, we use such a pattern (as seen > in JEP 486 [Appendix](https://openjdk.org/jeps/486#Appendix)): > > inst.owner().name().equalsString("java/lang/System") > > This pattern has a few issues: > > 1. This is not straightforward. Users may be tempted to write > > inst.owner().asSymbol().equals(CD_System) > > unaware of the degraded performance from extra conversions. > > 2. We aim to reduce the use of "internal names" with encapsulation. Direct > use of `"java/lang/System"` goes against this, and we have no constants > offered to reduce such error-prone use of internal name. > > Thus, I propose a new API in `ClassEntry`: > > boolean equalsSymbol(ClassDesc) > > that can be used conveniently to check the identity of a class entry with a > clean syntax and better performance. src/java.base/share/classes/jdk/internal/classfile/impl/AbstractPoolEntry.java line 609: > 607: var mySym = this.sym; > 608: if (mySym != null) > 609: return mySym.equals(symbol); Note that `ClassEntry::asSymbol()` also caches array `ClassDesc`s in `ref1.typeSym` so that the symbol cache can be shared with `FieldRefEntry` and `ConstantDynamicEntry`. -------------------------------------------------------------------------------- I suggest adding the following check: if (ref1.typeSym instanceof ClassDesc ref1Sym && ref1Sym.isArray()) { return ref1Sym.equals(symbol); } ------------- PR Review Comment: https://git.openjdk.org/jdk/pull/21526#discussion_r1802394895
