On Fri, 10 Apr 2026 05:20:15 GMT, Alan Bateman <[email protected]> wrote:

>> ServiceLoader specifies its iterator methods to throw 
>> ServiceConfigurationError (SCE) when there is an error locating, loading or 
>> instantiating a service provider.
>> 
>> There are linkage error cases where the LinkageError is propagated rather 
>> than throwing SCE with the LinkageError as cause. These typically arise in 
>> "bad environments" such as a mismatch between compile-time and run-time, 
>> missing dependencies, or maybe instrumentation/injection that has modified 
>> classes with references to classes that are not visible at run-time.
>> 
>> The behavior for these cases has varied a bit over time.  Since JDK 9 (when 
>> ServiceLoader was significantly re-implemented), a LinkageError thrown by 
>> Class.forName when attempting to load a service provider class is propagated 
>> without throwing SCE. Since JDK 24 (when support for the security manager 
>> was removed), if getting the no-arg the constructor triggers class loading 
>> and a LinkageError is thrown then is propagated without throwing SCE.
>> 
>> The proposal here is to change the implementation to throw SCE consistently 
>> for these cases. This would be observable change so there is a CSR to review 
>> the impact.
>> 
>> One of the existing tests for "bad" annotation processors needed to be 
>> updated to handle the ServiceConfigurationError as that test had worked 
>> around the issue and was handling the cause instead.
>
> Alan Bateman has updated the pull request incrementally with one additional 
> commit since the last revision:
> 
>   Move test sources into text blocks

Thanks very much for the test scenario simplification.

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

Marked as reviewed by liach (Reviewer).

PR Review: https://git.openjdk.org/jdk/pull/30405#pullrequestreview-4087649841

Reply via email to