RFR: 8333854: IllegalAccessError with proxies after JDK-8332457 [v2]

[Claes Redestad]

On Mon, 10 Jun 2024 04:08:41 GMT, Chen Liang <liach at openjdk.org> wrote:

>> Please review this patch that fixes a critical issue that breaks some Proxy usages.
>> 
>> CONSTANT_Class and CONSTANT_MethodType must fail resolution for inaccessible package-private types per JVMS 5.4.3.1 and 5.4.3.5, yet such types can appear in method descriptors in the same class. The proposed way to bypass is to store the MethodType as its descriptor string in the bootstrap method arguments, and use MethodType.fromMethodDescriptorString to restore the arguments, which does not have this restriction and does not eagerly load the parameter classes. This case isn't covered by existing tests, so a new test has been added.
>
> Chen Liang has updated the pull request incrementally with one additional commit since the last revision:
> 
>   Obtain classloader in security manager friendly code path

Agree with Roger that static imports should be used sparingly (possibly never with .*). 

If other reviewers are happy with the semantic shift in when/where class loading can/will happen and which class loader will be used then I won't object. I'm just not enough of an expert in this area to tell if there might be unintended side-effects we need to account for.

The added cost of spinning and calling these clinits means we might be better off reverting from a startup overhead perspective. JDK-8332457 was already a regression for some usage patterns.

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

PR Comment: https://git.openjdk.org/jdk/pull/19615#issuecomment-2163123616