Pinning of on-heap MemorySegment

[Maurizio Cimadamore]

On 28/08/2023 15:37, Gavin Ray wrote:
> Where this feels like it could potentially be an issue is when 
> consumers interact with a library/API that utilizes something like 
> pinning under the hood.
> Even if the user is aware of this, it seems like one thing that may 
> make sense is to make the ability to indefinitely pin segments gated 
> by a flag.
>
> That way the user has to knowingly opt-in to the potential of things 
> going awry?

Note that this is exactly what happens with FFM API's restricted 
methods. Some methods in the API are marked this way - e.g. 
MemorySegment::reinterpret, or Linker::downcallHandle.

This means that these methods are part of the Java SE API, but, if you 
call them you will get a warning.

To get rid of the warning, you have to explicitly opt-in the calling 
module into native access (which is done using the 
--enable-native-access command line flag).

In the future we plan to apply similar restrictions on JNI as well [1] 
and also, at some point, to turn warnings into errors.

All this to say that, *if* MemorySegment::pin is a restricted method, 
then I think the opt-in you talk about would already be in place.

[1] - https://mail.openjdk.org/pipermail/jdk-dev/2023-August/008061.html