[foreign-memaccess] RFR 8230449: Simplify creation of handles for memory slices

[Jorn Vernee]

Looks good!

Cheers,
Jorn

On 03/09/2019 20:16, Maurizio Cimadamore wrote:

> New webrev:
>
> http://cr.openjdk.java.net/~mcimadamore/panama/8230449_v2/
>
> Cheers
> Maurizio
>
> On 03/09/2019 15:55, Maurizio Cimadamore wrote:
>>
>> On 03/09/2019 15:25, Jorn Vernee wrote:
>>>
>>> Hi,
>>>
>>> Some comments:
>>> - MemoryLayout.java; The javadoc should be updated from "if the step 
>>> factor is < 0" -> "is 0", since it can also be negative.
>>> - LayoutPathImpl.java; Could use a helper method to do the sequence 
>>> index check (code is duplicated between 2 `sequenceElement` 
>>> methods). Also, the `step == 0` check is done in the 
>>> PathElement::sequenceElement already.
>>>
>> Well spotted - will fix
>>>
>>> Lastly, I'm not so sure about the last check for invalid step 
>>> factor. It would still be valid to access a single element when the 
>>> coordinate passed to the VH is 0, but this check blocks that.
>>>
>> The idea is that if you pass a step factor that is bigger than the 
>> selected sequence slice.
>>
>> It's true that, in general, we provide no guarantee for out-of-bound 
>> indices in array-like VH - but I guess I saw this as a user-mistake 
>> and wanted to prevent it.
>>
>> Now that I think more, I guess I see that, by removing that check, we 
>> would actually be able to produce VH which are capable of accessing 
>> exactly ONE element, which I guess can come in handy in certain cases.
>>
>> Maurizio
>>
>>> Rest looks good.
>>>
>>> Cheers,
>>> Jorn
>>>
>>> On 02/09/2019 17:27, Maurizio Cimadamore wrote:
>>>> Hi,
>>>> the foreign memory API is pretty good at expressing access to 
>>>> multi-dimensional data, e.g. by allowing to define custom strides 
>>>> and offsets. There are however some limitations that limit the 
>>>> expressiveness of the API when it comes to denote complex 
>>>> VarHandle-based views of contiguous chunks of memory:
>>>>
>>>> 1) It is not possible to use negative strides (which makes it 
>>>> impossible e.g. to create a VarHandle which returns the contents of 
>>>> the memory region "in reverse order")
>>>> 2) there's no way in the PathElement API to select a 'slice' of a 
>>>> sequence layout (e.g. from start given start index, with given step)
>>>>
>>>> The following webrev aims at closing this gap:
>>>>
>>>> http://cr.openjdk.java.net/~mcimadamore/panama/8230449/
>>>>
>>>> P.S.
>>>> To allow negative strides I had to remove a validation of the 
>>>> stride in the combinator API which ensured that the stride was 
>>>> greater than the offset + carrierSize. I believe this restriction 
>>>> to be a tad obscure and not a fundamental one (this was also 
>>>> pointed out by Jorn when I added the restriction in the first 
>>>> place), so I'm ok to drop it in name for more expressiveness.
>>>>
>>>> Maurizio
>>>>