196017 - 1.5R4 regression: script can not find classes on some versions of JDK

Status: VERIFIED FIXED

(Rhino Graveyard :: Core, defect)

Description

[Igor Bukanov]

New class loader code for generated classes in 1.5R4 to fix bug 166530 causes a
regressions when used with some JDK 1.4 releases as was reported by Anreas Rozek:

After having "installed" Rhino 1.5r4 on my system, many of my old
Rhino scripts fail to run any longer. Reason: Java class lookup seems
to be broken!

Here are some facts:
 - Java classes and interfaces within JAR files or directory trees
   (within my "ClassPath") are no longer found - although they are
   definitely present
 - unpacking JAR files (or moving directory tree contents) into a
   folder such as $JavaHome/jre/lib/classes solves the problem -
   however, I would even have to unpack the Rhino library (js.jar) 
 - going back to Rhino 1.5r3 immediately solves the problem as well
 - I'm currently running Sun JDK 1.4.0_03 - but the problem exists
   as well under 1.4.0_02, 1.4.0_01 and 1.3.?_06 (don't remember the
   version number) I already had a similar problem with "BeanShell"
   (www.beanshell.org) and JDK 1.4.1 a few weeks ago - I then deci-
   ded to go back to 1.4.0_03 as this solved most of the problems.
   Later, a colleague ran into similar class loading problems and
   had to go back to 1.4.0 as well...(very strange)
 - I'm running Win98SE, she was running WinXP Professional

Comment 1

[Igor Bukanov]

It seems the regression is caused by a JDK bug on Win98 when it returns for
Rhino classes loaded from CLASSPATH variable a class loader that can only load
classes from Rhino jar or system locations and can not find classes from other
components of CLASSPATH. On the other hand a class loader returned by
Thread.getContextClassLoader() is able to find classes from all CLASSPATH
components and since Rhino 1.5R3 uses that for class loading, the problem only
appeared in 1.5R4.

Comment 2

[Igor Bukanov]

Attachment: Use Thread.getContextClassLoader as the last resort to find classes

The patch does not restore 1.5R3 since it used Thread.getContextClassLoader()
initially and only then it tried a class loader for Rhino classes leading to
various problems. The patch instead uses Thread.getContextClassLoader() as a
last attempt to find the required class thus preserving child-parent class
loader semantics between generated and Rhino classes.

Comment 3

[Igor Bukanov]

I commited the patch, for verification I asked Anreas Rozek to check it, let see
what he will report.

Comment 4

[Phil Schwartau]

Can we mark this one Verified? Have we heard from Anreas?

Comment 5

[Igor Bukanov]

Marking as verified assuming that silence of Anreas Rozek means it is OK for him...

Comment 6

[Igor Bukanov]

Targeting as resolved against 1.5R4.1