On Tue, 10 Oct 2023 17:17:00 GMT, Jorn Vernee <jver...@openjdk.org> wrote:
> Port of: https://github.com/openjdk/panama-foreign/pull/898 > > Original PR body: > >> >> This patch addresses some issues with `MemorySegment.Scope::equals`. More >> specifically, when two segments are created from the same array/buffer, >> their underlying scope should be equal, but currently it is not, which is >> confusing. >> >> This patch addesses that by adding a new subclass of `GlobalSession` >> (`HeapSession`) whose `equals`/`hashcode` methods are defined in terms of >> the heap reference it wraps. Note that the reference is deduped on >> construction - that is, we always try to find the "bottom" refrence that we >> should try to keep alive with the session, and we use that for comparisons >> (which avoid issues when comparing buffer slices and views). >> >> As part of the changes, this patch also adds more javadoc on >> `MemorySegment.Scope` to document how equality works. And, this patch also >> makes all zero-length memory segments associate with the same "native" >> scope. This is a scope that is always alive, but is not the global scope >> (not to confuse with segments created by the global arena). While what the >> implementation does today is technically correct (a new fresh scope is >> returned with each new ZLMS), it seems also overkill given that the scope >> associated with segments originated outside Java code is not very >> interesting - if users really care about lifetime of such segments they >> should override the scope with `MemorySegment::reinterpret`. This leads to >> some changes in the javadoc of other methods where I replaced `fresh scope >> that is always alive` with just `scope that is always alive`, to avoid >> suggesting that a new scope should be created each time. Note to reviewers: I will also need a reviewer on the CSR. If you've reviewed that, please add yourself as reviewer (on the CSR). Thanks ------------- PR Comment: https://git.openjdk.org/jdk/pull/16126#issuecomment-1760301526
