[foreign-abi] RFR 8236004: Memory access var handles should support MemoryAddress carrier

[Jorn Vernee]

Hi,

Some comments:

AddressVarHandleGenerator.java:
- `erase` method seems to be indented too far.

X-VarHandleMemoryAddressView.java.template:
- In the long2addr method I'd prefer using invokeExact instead of 
invoke, to make sure the call always gets inlined.

LayoutPath.java:
- In dereferenceHandle, can the carrier check here be made to use Utils 
as well?

Rest looks good. (Casing over the ABI impls in Utils to get the 
POINTER_SIZE seems pretty unfortunate, but that was already on the todo 
list I believe).

Also, the patch does not apply cleanly (probably due to the manual merge 
with foreign-memaccess I just did, sorry), so you might want to rebase.

cheers,
Jorn

On 18/12/2019 12:03, Maurizio Cimadamore wrote:
> Here's a new revision which restores erased carrier and access types:
>
> http://cr.openjdk.java.net/~mcimadamore/panama/8236004_v3/
>
> Thanks for taking the time of retracing the steps and explaining how 
> linking is supposed to work; I feel I now have a much better 
> understanding of how things are connected.
>
> Maurizio
>
> On 17/12/2019 19:01, Paul Sandoz wrote:
>>
>>
>>> On Dec 17, 2019, at 10:47 AM, Maurizio Cimadamore 
>>> <maurizio.cimadamore at oracle.com 
>>> <mailto:maurizio.cimadamore at oracle.com>> wrote:
>>>>
>>>> Yes, it works fine, until you pass an instance of something that is 
>>>> not of MemoryAddressProxy e.g. pass in an instance of String, which 
>>>> will get spoofed as an instance of MemoryAddressProxy.  The guards 
>>>> will only check if the object is a reference type.
>>>
>>> Ok, I see what you mean, the problem is not that linking occurs 
>>> abnormally, the problem is that the underlying linkage is not sound 
>>> (in the absence of Java casts)
>>>
>> Yes, the linkage + invocation runtime checks are not sound.
>>
>>> I guess this means that same treatment should also be applied to the 
>>> MemoryAddressProxy parameter (e.g. of a VarHandle::set operation) 
>>> right?
>>>
>>
>> Yes, for all access modes where the arg of $type$ is 
>> of MemoryAddressProxy a cast check is required (which should optimize 
>> away).
>>
>> I think that can be performed by the addr2long method?
>>>
>> Paul.