[PATCH 03 of 10] rust-cpython: mark PySharedState as Sync so &'PySharedState can be Send

[Raphaël Gomès]

Poisoning is the way to go in the long run. I haven't really thought 
about how much of a pain this could be.

On 10/5/19 5:03 PM, Yuya Nishihara wrote:
> On Sun, 22 Sep 2019 15:41:40 +0900, Yuya Nishihara wrote:
>> # HG changeset patch
>> # User Yuya Nishihara <yuya at tcha.org>
>> # Date 1568674765 -32400
>> #      Tue Sep 17 07:59:25 2019 +0900
>> # Node ID fcaf01804ac8198d52d141d642d5ef7c935360cc
>> # Parent  34a355c69f643f15ff46f4524c75137c08a8de3b
>> rust-cpython: mark PySharedState as Sync so &'PySharedState can be Send
>>
>> The goal is to store &'static PySharedState in $leaked struct, which allows
>> us to move the $leaked struct out of the macro. Currently, it depends on
>> inner.$data_member(py), which can't be generalized.
>>
>> PySharedState is Sync because any mutation or read operation is synchronized
>> by the Python GIL, py: Python<'a>, which should guarantee that &'PySharedState
>> can be sent to another thread.
>>
>> diff --git a/rust/hg-cpython/src/ref_sharing.rs b/rust/hg-cpython/src/ref_sharing.rs
>> --- a/rust/hg-cpython/src/ref_sharing.rs
>> +++ b/rust/hg-cpython/src/ref_sharing.rs
>> @@ -18,6 +18,10 @@ pub struct PySharedState {
>>       mutably_borrowed: Cell<bool>,
>>   }
>>   
>> +// &PySharedState can be Send because any access to inner cells is
>> +// synchronized by the GIL.
>> +unsafe impl Sync for PySharedState {}
> Perhaps, this unsafe impl can be removed later.
>
> If we take the generation-poisoning idea, PySharedState will be just
> { generation: AtomicUsize }, which is inherently Sync.