dissecting memory sessions

[Maurizio Cimadamore]

On 10/11/2022 16:52, Pedro Lamarão wrote:
> IMHO, though the concern is clearly with reasoning about life-time, I 
> think the design is improving in another direction, the specification 
> of a clear, well defined, common memory allocator "service".

I'm glad you noticed. Really, one of the things that didn't quite click 
in the Java 19 API was to see how hard it was to construct a custom 
allocator implementing a custom policy. This should be, in my opinion, 
the bread and butter of the FFM API, so we need to make it easy(er) for 
libraries to come up with their own allocators. The fact that, in Java 
19, all custom closeable allocators need to protect themselves against 
"sneaky" close (e.g clients calling close on the segments they return) 
is a major source of problem - both conceptual (because all implementors 
of custom allocators will have to worry about creating non-closeable 
views and attaching segments there), and practical (because creating all 
these views has a non trivial performance cost, given that the very 
existence of session views makes everything less monomorphic).

Thanks
Maurizio