[foreign-memaccess+abi] RFR: 8291826: Rework MemoryLayout Sealed Hierarchy [v5]

Paul Sandoz psandoz at openjdk.org
Fri Aug 19 17:15:50 UTC 2022


On Fri, 19 Aug 2022 12:12:19 GMT, Per Minborg <duke at openjdk.org> wrote:

>> This PR is a significant refactor of the `MemoryLayout` sealed hierarchy and whose purpose is described more in detail in  https://bugs.openjdk.org/browse/JDK-8291826. In short, the main purpose is to allow pattern matching.
>> 
>> The objective of the PR is to fix the issues described above while retaining as much API compatibility as possible with the pre-PR API. No changes were needed in any of the existing tests for example.
>> 
>> The old sealed hierarchy prevented pattern matching with totality from being used and implied a number of other problems as illustrated in the picture below:
>> 
>> ![graphviz (33)](https://user-images.githubusercontent.com/7457876/185412743-febeb3cf-aebc-4e2e-b78f-3b8ef87b0510.png)
>> 
>> Red nodes cannot be used in pattern matching. Light red nodes are implementations that are visible in the API. White nodes are interfaces.
>> 
>> After this PR has been integrated, an improved sealed hierarchy will exist as depicted here:
>> 
>> ![graphviz (34)](https://user-images.githubusercontent.com/7457876/185413448-8435b23f-5ac4-4999-8b0a-0d6051a0d9be.png)
>> 
>> Gray "Impl" nodes are internal and not exported by the `java.base` module. White nodes are interfaces.
>> 
>> Below, the "Principal Totalities" for the new API are outlined:
>> 
>> 
>> var v0 = switch (memoryLayout) {
>>   case MemoryLayout ml -> 0;
>> };
>> var v1 = switch (memoryLayout) {
>>   case GroupLayout gl -> 0;
>>   case PaddingLayout pl -> 0; // leaf
>>   case SequenceLayout sl -> 0; // leaf
>>   case ValueLayout vl -> 0;
>> };
>> var v2 = switch (memoryLayout) {
>>   case PaddingLayout pl -> 0; // leaf
>>   case SequenceLayout sl -> 0; // leaf
>>   case ValueLayout vl -> 0;
>>   case StructLayout sl -> 0; // leaf
>>   case UnionLayout ul -> 0; // leaf
>> };
>> var v3 = switch (memoryLayout) {
>>   case PaddingLayout pl -> 0; // leaf
>>   case SequenceLayout sl -> 0; // leaf
>>   case StructLayout sl -> 0; // leaf
>>   case UnionLayout ul -> 0; // leaf
>>   case OfAddress oa -> 0; // leaf
>>   case OfBoolean ob -> 0; // leaf
>>   case OfByte ob -> 0; // leaf
>>   case OfChar oc -> 0; // leaf
>>   case OfDouble od -> 0; // leaf
>>   case OfFloat of -> 0; // leaf
>>   case OfInt oi -> 0; // leaf
>>   case OfLong ol -> 0; // leaf
>>   case OfShort os -> 0; // leaf
>> };
>> var v4 = switch (memoryLayout) {
>>   case GroupLayout gl -> 0;
>>   case PaddingLayout pl -> 0; // leaf
>>   case SequenceLayout sl -> 0; // leaf
>>   case OfAddress oa -> 0; // leaf
>>   case OfBoolean ob -> 0; // leaf
>>   case OfByte ob -> 0; // leaf
>>   case OfChar oc -> 0; // leaf
>>   case OfDouble od -> 0; // leaf
>>   case OfFloat of -> 0; // leaf
>>   case OfInt oi -> 0; // leaf
>>   case OfLong ol -> 0; // leaf
>>   case OfShort os -> 0; // leaf
>> };
>> 
>> 
>> 
>>  Additional improvements can be made later in a separate PR, including removing/improving some of the internal abstract classes.
>
> Per Minborg has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Fix problems in static initializers

src/java.base/share/classes/jdk/internal/foreign/layout/AbstractGroupLayout.java line 110:

> 108:      */
> 109:     public List<MemoryLayout> memberLayouts() {
> 110:         return Collections.unmodifiableList(elements);

`elements` should already be unmodifable. The constructor performs `List.copyOf()`, and that's a noop if the list passed was produced from that method or `List.of`.

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

PR: https://git.openjdk.org/panama-foreign/pull/710


More information about the panama-dev mailing list