[foreign] RFR: struct/global functions stubs should attached to some scope

[Sundararajan Athijegannathan]

Looks good.

-Sundar

On 09/07/18, 11:36 PM, Maurizio Cimadamore wrote:
> Hi,
> there is a currently a problem with function pointers: when you assign 
> a functional interface object to a struct/global variable, a stub is 
> generated on the fly; the native stub has a weak link back to the 
> UpcallHandler object it originated from - but the UpcallHandler object 
> is not kept anywhere - which means the GC is free to collect it as 
> soon as possible (and this will also cause deallocation of the native 
> stub, via the cleaner).
>
> We need some way to attach stub resources to scopes (either the scope 
> of a resource or the recently-added global library scope). That's what 
> this patch does. When Scope::close is called, the list of stub is 
> null-ed, so that the GC can deallocate things as expected.
>
> Note that this is an incremental improvement while we await for a more 
> cohesive native resource management strategy/API.
>
> The test for function callbacks has been tweaked to feature some 
> System.gc - w/o this patch, these alterations cause the test to fail 
> pretty reliably.
>
> Webrev:
>
> http://cr.openjdk.java.net/~mcimadamore/panama/func-scope/
>
> Maurizio
>
>
>