Feasibility of a Span class in java

[forax at univ-mlv.fr]

> From: "Red IO" <redio.development at gmail.com>
> To: "Remi Forax" <forax at univ-mlv.fr>
> Cc: "panama-dev" <panama-dev at openjdk.org>
> Sent: Monday, November 7, 2022 6:54:42 PM
> Subject: Re: Feasibility of a Span class in java

> This looks like it has a pretty large overhead when accessing multiple values. I
> think it would be more beneficial to have 1 object which represents the whole
> range like a list. And is able to subdivide like a memory segment. It would
> practically be a user friendly wrapper for memory segments. The biggest
> challenge with an api like this is
> 1. Primitives inability to be efficiently generic and 2. Integration with the
> whole jdk. The first could be solved in in 2 ways: Valhalla delivers generics
> over primitives (still in development) or old school with a ton of duplicate
> code.
> The second problem would be a group effort. To effectively use this kind of api
> it would need integration in strings, buffers, arrays, lists, streams and
> everywhere you might find a continuous block of data. This would require it to
> become an agreed goal to create it. Otherwise the effort would be wasted.

It should be efficient given that all span objects are value types, and only exist on stack, so can be completly unboxed by a Valhalla VM. 
There is a third options if you want to implement specialized generics now, use bytecode generation / method handles / Lookup.defineHiddenClass. 

Here is an example of how to have the API of a list/map of records but the implementation of a record of arrays. 
[ https://github.com/forax/struct-of-array | https://github.com/forax/struct-of-array ] 

The integration with strings and buffers can be done through implementations of CharSequence. Arrays and Map can be implemented by bytecode generation and for stream too (but that last part will be painful). 

Rémi 

> On Mon, Nov 7, 2022, 18:34 Remi Forax < [ mailto:forax at univ-mlv.fr |
> forax at univ-mlv.fr ] > wrote:

>> It's on my TODO list :)

>> You create an interface
>> interface MutableSpan {
>> int value();
>> void value(int value);
>> }

>> and you can map it to any Panama memory because this is implemented as value
>> class with two fields, a memory segment (or whatever the actual name) and an
>> index.
>> Then you need to see a contiguous space of memory as a List of span objects
>> where the value class instances representing the value are created on the fly
>> when asked.

>> Rémi

>>> From: "Red IO" < [ mailto:redio.development at gmail.com |
>>> redio.development at gmail.com ] >
>>> To: "panama-dev" < [ mailto:panama-dev at openjdk.org | panama-dev at openjdk.org ] >
>>> Sent: Monday, November 7, 2022 5:59:28 PM
>>> Subject: Feasibility of a Span class in java

>>> Following the development and plans for both Panama's memory access api and the
>>> required changes provided by Valhalla while looking at c#'s directions for the
>>> last few years I was wondering if writing a class that serves as a universal
>>> and easy-to-use interface for accessing memory of all sorts of "value based"
>>> types inspired by c#' s Span struct would be beneficial for Java. Of course
>>> this assumes the completion of the memory api (panama) , the generic
>>> unification (Valhalla) and custom value based types (Valhalla). The idea is to
>>> speed up array and string manipulation by using short lived slicing objects
>>> (Span). I already wrote a poorly implemented slicer for strings and saw
>>> incredible performance gains in comon operations like substring and split.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/panama-dev/attachments/20221107/0a0ef031/attachment.htm>