On Mon, 25 Mar 2024 18:43:33 GMT, Mandy Chung <mch...@openjdk.org> wrote:
> Do you expect if developers start to package shared objects in the format of
> archive objects? If so, it would be reasonable to support #1 for
> compatibility.
Yes. I agree. We should keep this compatibility for entire flow from
classloader to the hotspot code. And it has been in J9 for long time for the
exact same reason, which actually brought out the issue for dcstartup in
OpenJDK.
> For #2, `System::loadLibrary("foo(libfoo.so.1)"`,
> `System::load("/usr/lib/libfoo.so((libfoo.so.1)")` and
> `System::load("/usr/lib/libfoo.a((libfoo.so.1)")` are never supported. I
> think applications should use `SymbolLookup::libraryLookup` instead on Java
> 22 and later. I don't use `libclang` as the example here because that's
> related to jextract and has nothing to do with `System::load` as Maurizio
> said.
I just feel keeping load functionality is still applicable, which does not
expect a deterministic mapping (or does it ? ). if loadLibrary expects this to
be deterministic, then i agree, we should drop that.
-------------
PR Comment: https://git.openjdk.org/jdk/pull/17945#issuecomment-2018690683
PR Comment: https://git.openjdk.org/jdk/pull/17945#issuecomment-2018693188
