[External] : Re: MemorySegment.ofAddress(...).reinterpret(...)

[Brian S O'Neill]

I was confused by your earlier statement and thought that a restriction 
check is applied on each invocation, not just at the initial acquisition 
of the segment.

As for two versions of the copy method, I didn't necessarily think that 
a new public API is needed. Assuming that all the inlining magic works, 
then the copy could check if the segment is unrestricted or not and then 
those checks would disappear.


On 2023-07-07 12:26 PM, Jorn Vernee wrote:
> This is typically how we handle the module access check in the current 
> API. The check is done when obtaining a capability, so that the access 
> use can be check-free. e.g. we check module access when reinterpret is 
> called, but not when accessing the resulting segment. This doesn't 
> necessarily require using MethodHandles.
> 
> I think the greater issue is that e.g. having a separate version of 
> MS::copy that doesn't do any checks would create two versions of the 
> same API, one checked and one unchecked, for seemingly very little gain. 
> This doesn't seem worth it.
> 
> Jorn
> 
> On 07/07/2023 18:11, Brian S O'Neill wrote:
>> When I obtain a MethodHandle, a module access check is performed. When 
>> I invoke it, no check is performed, and a static final reference to a 
>> it gets fully optimized.
>>
>> It seems to me that if I were to obtain access to the copy method (and 
>> the get/set methods) via a MethodHandle, then the restricted method 
>> check only needs to be performed once.
>>
>> Granted, this approach is a little bit ugly from an API perspective, 
>> but MethodHandles are already a key component of the ffm API. A 
>> MethodHandle transformation could also be performed automatically by 
>> the compiler (at some level), but that's probably going too far.
>>
>> On 2023-07-07 08:55 AM, Jorn Vernee wrote:
>>> Well, the implementation of a restricted method doesn't know whether 
>>> the caller's module has already been checked :) That is tracked on a 
>>> per-module basis, but either way we have to walk the stack to get the 
>>> caller's module.
>>>
>>> Jorn
>>>
>>> On 07/07/2023 17:51, Brian S O'Neill wrote:
>>>> Is the restricted method check performed every single time? Why? The 
>>>> restriction applies to the module, so it only really needs to be 
>>>> checked once.
>>>>
>>>> On 2023-07-07 08:32 AM, Jorn Vernee wrote:
>>>>
>>>>> We can't just drop an array bounds check, even if we were to make 
>>>>> the method restricted. (Besides, a restricted method needs to check 
>>>>> if the caller has native access, which adds its own overhead)
>>>>>