RFR: JDK-8267936: PreserveAllAnnotations option isn't tested

[Peter Levart]

On 01/06/2021 22:28, John Rose wrote:
> On Jun 1, 2021, at 7:02 AM, Brian Goetz <brian.goetz at oracle.com<mailto:brian.goetz at oracle.com>> wrote:
>
> As Alan's archaeology discovered, this flag appears to be a leftover from the original implementation, and I could find no signs of its usage.
>
> Note to self and other reviewers of future versions
> of the JVMS: When you see proposed language like
> the “unless…” of JVMS 4.7.17, remember today's
> conversation and try to avoid putting dark corners
> like that into the JVMS.
>
> The RuntimeInvisibleAnnotations attribute is similar to the RuntimeVisibleAnnotations attribute (§4.7.16), except that the annotations represented by a RuntimeInvisibleAnnotations attribute must not be made available for return by reflective APIs, [[WE SHOULD HAVE STOPPED HERE]] unless the Java Virtual Machine has been instructed to retain these annotations via some implementation-specific mechanism such as a command line flag. In the absence of such instructions, the Java Virtual Machine ignores this attribute.
> https://docs.oracle.com/javase/specs/jvms/se16/html/jvms-4.html#jvms-4.7.17


Hi,

I do think this option, in current form, is useful as an escape hatch 
for exactly the case that Jaroslav has. Java/JVM as a late-binding 
runtime does a very good job of keeping the information from source 
files local in direct compilation artefacts which facilitates separate 
compilation where changes to and recompilation of one source file have 
immediate effect on the whole app. There are exceptions to the rule such 
as compile-time constants, ... and also annotation uses where the 
information from one source file (annotation retention) is baked into 
compilation artefacts of other source files (RuntimeVisibleAnnotations 
vs. RuntimeInvisibleAnnotations). PreserveAllAnnotations option helps to 
overcome the situation where the annotation has been updated but classes 
where such annotation is used have not been recompiled (yet). I see the 
two class file attributes merely as a runtime optimization while both 
kinds of annotations could be kept in a single attribute. The JVM option 
just disables this optimization. So I would still keep the option.

Regards, Peter