Like CI, JVMCI has a mechanism for creating handles to objects during compilation (e.g. when folding the load of an initialized static final object field, the compiler can replace the load with a reference to field's object value). Unlike CI, JVMCI creates a new handle each time the compiler resolves an object value (e.g. 2 static fields referring to the same value). This puts pressure on the resource from which JVMCI handles are allocated. For example, there are more roots for a GC to scan.

This issue is to investigate making JVMCI do something more like CI. It would be a bonus if there could be some implementation sharing between CI and JVMCI in terms of handle management.