RFR: 8342486: Implement JEP draft: Structured Concurrency (Fourth Preview)

[Alan Bateman]

On Wed, 6 Nov 2024 15:47:55 GMT, Alan Bateman <alanb at openjdk.org> wrote:

> Changes for [JEP draft: Structured Concurrency (Fourth Preview)](https://openjdk.org/jeps/8340343). The JEP isn't on the technical roadmap yet. The proposal is to re-preview the API with some changes, specifically:
> 
> - A [StructuredTaskScope](https://download.java.net/java/early_access/loom/docs/api/java.base/java/util/concurrent/StructuredTaskScope.html) is now opened with a static factory method instead of a constructor.  Once opened, the API usage is unchanged: fork subtasks individually, join them as a unit, process outcome, and close.
> - In conjunction with moving to using a static open method, policy and desired outcome is now selected by specifying a Joiner to the open method rather than extending STS. A Joiner handles subtask completion and produces the result for join to return. Joiner.onComplete is the equivalent of overriding handleComplete previously. This change means that the subclasses ShutdownOnFailure and ShutdownOnSuccess are removed, replaced by factory methods on Joiner to get an equivalent Joiner.
> - The join method is changed to return the result or throw STS.FailedException, replacing the need for an API in subclasses to obtain the outcome. This removes the hazard that was forgetting to call throwIfFailed to propagate exceptions.
> - Configuration that was provided with parameters for the constructor is changed so that can be provided by a configuration function.
> - joinUntil is replaced by allowing a timeout be configured by the configuration function. This allows the timeout to apply the scope rather than the join method.
>  
> The underlying implementation is unchanged except that ThreadFlock.shutdown and wakeup methods are no longer confined. The STS API implementation moves to non-public StructuedTaskScopeImpl because STS is now an interface. A non-public Joiners class is added with the built-in Joiner implementations.

src/java.base/share/classes/java/util/concurrent/StructuredTaskScope.java line 122:

> 120:  * {@link #close() close} method always waits for threads executing subtasks to finish,
> 121:  * even if the scope is cancelled, so execution cannot continue beyond the {@code close}
> 122:  * method until the interrupted threads finish.

Viktor has suggested that we move this paragraph up or at least make it more prominent.

src/java.base/share/classes/java/util/concurrent/StructuredTaskScope.java line 337:

> 335:  * <a href="{@docRoot}/java.base/java/util/concurrent/package-summary.html#MemoryVisibility">
> 336:  * <i>happen-before</i></a> any actions taken by that subtask, which in turn
> 337:  * <i>happen-before</i> the subtask result is {@linkplain Subtask#get() retrieved}.

Chen has suggested that this be expanded to specify a HB between onComplete and join. Viktor and I have discussed extending the existing text to deal with subtask outcome that "contributes" to the scope result but did not agree wording (note that this is a pre-existing short-coming in the spec, not a new issue).

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

PR Review Comment: https://git.openjdk.org/jdk/pull/21934#discussion_r1832464575
PR Review Comment: https://git.openjdk.org/jdk/pull/21934#discussion_r1832468226