Question about global SegmentScope

[Maurizio Cimadamore]

On 17/04/2023 15:23, 刘希晨 wrote:
> Thank you for the clarification. I am stuck with my mind because I 
> thought that if the SegmentScope() is global, then even if the 
> MemorySegment Object has been GCed, the underlying memory won't be 
> released, not until the JVM exit. (Is that correct? I think that's the 
> difference between the global scope and the auto scope).

This is not correct.

The segment you get back from a native library has an "unknown scope" 
(as said, in the Java 21, this is different from the "global" scope, 
which is guaranteed to be always alive until VM termination). Segments 
with "unknown" scopes can be accessed, but at your own risk, as the 
region of memory backing the segment could be freed (in the general 
case) at any time by the native library.

Now, in reality things are not as bleak as they sound - while _in 
principle_ the region backing the segment might disappear w/o notice, in 
practice a well-defined library will advertise who and when should 
release the regioin of memory associated with a pointer that has been 
returned by some function. But this is a detail that is typical buried 
in the documentation of the function you are interacting with - it 
cannot be expressed in the static type system of the C language and the 
FFM API has no knowledge of it.

Maurizio