[foreign-memaccess] API refinements

[Maurizio Cimadamore]

On 06/01/2020 23:28, Paul Sandoz wrote:
> The trick I was trying to pull off, which does not seem to be working :-), is to define temporal confinement as well as thread confinement.  In that sense a segment returned by “acquire” is a thread-confined *and* temporally-confined view, be it owned by a different thread or otherwise.

Heh - too clever :-)

Jokes aside, the problem is that we use the C-word in many place in the 
javadoc to say that segments are thread-confined. So I think 
"overriding" the C-word in this way might lead to confusion?

But you are on the right path when you talk about temporal confinement - 
in the sense that this is part of what the method does - a part that the 
A-word does not capture at all, sadly.

Maurizio


>
> Status quo is ok by me.
>
> Paul.
>
>> On Jan 6, 2020, at 3:07 PM, Maurizio Cimadamore <maurizio.cimadamore at oracle.com> wrote:
>>>> * MemorySegment::acquire() -- replace with MemorySegment::fork?
>>>>
>>>> I came up with that - and I think I still like it. Acquire seems to suggest that it has something to do with threads and ownership - which it does, but in reality you can acquire a segment even from same owner. Fork seems a more neutral term, and, to me helps better visualizing what's really going on (e.g. a tree of segments is being created).
>>>>
>>> Not sure about fork, it reminds me of processes or fork-join, something that is task oriented, overloaded term… but I think it could grow on me.
>>>
>>> Re-reading the JavaDoc, we could refer to it as a confined view and call the method asConfined and s/acquired memory segment/confined view memory segment”.
>> Not too sure. For two reasons: (i) it seems to suggest that the segment it came from it wasn't confined (they all are, in a way) and, more importantly (ii) because the 'asXYZ' naming convention implies a 'dumb' view which, if closed, closes everything, which is not the case here (this is why slice() has been renamed to asSlice()).
>>
>> So, while we could call it confine(), the problem is that this method is effectively doing two things:
>>
>> 1) create a new sub-segment whose life-cycle controls that of the parent
>> 2) give a new confinement owner to the new sub-segment
>>
>> Seems to me than names like acquire() and confine() suggest a string bias for (1), whereas fork() is probably biased for (2).
>>
>> The problem with bias is that names like acquire() or confine() make perfect sense in a share context, but makes little sense when used in a single thread context, where you might want to use the acquire mechanism as a way to e.g. make sure that you can't close the parent segment before all the slices have been closed. In those case a name like fork() works better.
>>
>> I'm fine with keeping acquire() if nobody comes up with a better alternative.
>>