On Fri, 9 Feb 2024 12:48:48 GMT, Daniel Fuchs <dfu...@openjdk.org> wrote:
>> Currently these tests ignore vm flags, In most cases I've updated them to
>> use `ProcessTools.createTestJavaProcessBuilder` this usually required some
>> cleanup also.
>>
>> `test/jdk/java/net/ServerSocket/AcceptCauseFileDescriptorLeak.java` and
>> `test/jdk/java/net/URLConnection/6212146/TestDriver.java` have been set to
>> use `@require vm.flagless` because they both use `sh` commands.
>>
>> I've ran these changes against tiers 1-3 and everything seems stable
>
> test/jdk/java/net/URLClassLoader/getresourceasstream/TestDriver.java line 63:
>
>> 61: ),
>> 62: ProcessTools.createTestJavaProcessBuilder(
>> 63: "-cp", ".",
>
> If I'm not mistaken, this will result in the `-cp` option being passed twice,
> with different values, to the `java` sub process. I believe that
> `ProcessTools.createTestJavaProcessBuilder` will end up passing "-cp
> <java.class.path>" to the subprocess, unless -Dtest.noclasspath=false is
> defined in the parent process.
> Tough I could not find what would happen if -cp is passed twice in `man java`
> - it appears that the long standing behaviour is that the last one wins - so
> I guess that's OK, as I doubt this behaviour (last one wins) could be
> changed...
>
> An alternative could be to create a new method in ProcessTools that would
> allow to pass a flag to prevent the addition of -cp <java.class.path> when
> that's not desiravle, and call that here - and in all other places where -cp
> is passed to `ProcessTools.createTestJavaProcessBuilder`.
Would it be possible to change the `createJavaProcessBuilder` method in
ProcessTools to include something to check if -cp is already an argument being
passed?
maybe something like this?
if (!noCP && !args.contains("-cp")) {
args.add("-cp");
args.add(System.getProperty("java.class.path"));
}
-------------
PR Review Comment: https://git.openjdk.org/jdk/pull/17787#discussion_r1486023544
