Integrated: 8318538: Add a way to obtain a strided var handle from a layout

Maurizio Cimadamore mcimadamore at openjdk.org
Fri Oct 20 10:38:01 UTC 2023


On Thu, 19 Oct 2023 16:52:53 GMT, Maurizio Cimadamore <mcimadamore at openjdk.org> wrote:

> The 21 iteration of the FFM API used to contain a method, namely `ValueLayout::arrayElementVarHandle`. This var handle factory was useful when accessing elements in a flat array of variable size.
> 
> In 22, this method has been removed because:
> * It was not general enough (it only worked with value layouts)
> * A more useful/general version could be obtained with the following code:
> 
> 
> MethodHandles.collectCoordinates(varHandle(elements), 1, scaleHandle());
> 
> 
> Given that accessing variable-length array is rather common, and given that combining var handles is perceived as hard, we noticed that developers often tried to workaround by creating a sequence layout of maximal size, and then deriving a var handle from it. This feels suboptimal (as a new layout is created just so that we can select from it, and using this layout for anything else will likely lead to exceptions, because of its size).
> 
> For these reasons, we would like to expose this functionality as a new method in the `MemoryLayout` interface, namely `MemoryLayout::arrayElementVarHandle`. This method generalizes the old method in the 21 API, and is implemented in the obvious way, using a combinator (see above).
> 
> This PR also adds a lot of narrative text to describe how to deal with variable-length arrays, which should clarify the role of the new method.

This pull request has now been integrated.

Changeset: 848ecc16
Author:    Maurizio Cimadamore <mcimadamore at openjdk.org>
URL:       https://git.openjdk.org/jdk/commit/848ecc1621c347ab12dd3f421af82cb55c71e075
Stats:     163 lines in 6 files changed: 119 ins; 21 del; 23 mod

8318538: Add a way to obtain a strided var handle from a layout

Reviewed-by: jvernee, pminborg

-------------

PR: https://git.openjdk.org/jdk/pull/16272


More information about the core-libs-dev mailing list