Fwd: What about Array.newInstance for restricted value-types?

[Brian Goetz]

Received on the -comments list.

Yes, this issue is well understood; I believe John has written about 
this on the -experts list already, or if not he will soon.  Reflection 
frequently has to bake back in access control checks that otherwise 
would be done by the VM during linkage, and many reflective methods are 
caller-sensitive for exactly this reason.

Do you  have any reason to believe that the standard techniques of 
securing reflection would result in problems?


-------- Forwarded Message --------
Subject: 	What about Array.newInstance for restricted value-types?
Date: 	Tue, 26 Jul 2022 14:38:05 +0200
From: 	Gernot Neppert <mcnepp02 at googlemail.com>
To: 	valhalla-spec-comments at openjdk.java.net




There is an ongoing discussion about how to prevent the creation of 
values with non-sensible zero-defaults.

Currently, the consensus seems to be that restricting access to a 
value-type's "companion-type" will achieve this.

However, I found this Anomaly:

The method java.lang.reflect.Array.newInstance(Class<?> 
componentType,int length) currently does not take the accessibility of 
the "componentType" into account!

So, it looks as if one could still create flat, zero-initialized arrays 
of private companion-types from anywhere in the codebase:

package org.secrets;

public value class Bar {

     private value companion Bar.val;

}

package com.external;

import java.lang.reflect.Array;

public class Main {

      Bar[] bars = 
(Bar[])Array.newInstance(Class.forName("org.secrets.Bar.val"), 42);

}






-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://mail.openjdk.org/pipermail/valhalla-dev/attachments/20220727/80e2ca9a/attachment.htm>