2016/01/25

DISCLAIMER: These notes are from the defunct k8 project which precedes SquirrelJME. The notes for SquirrelJME start on 2016/02/26! The k8 project was effectively a Java SE 8 operating system and as such all of the notes are in the context of that scope. That project is no longer my goal as SquirrelJME is the spiritual successor to it.

11:16

Currently at the hospital. In either case, must solve the classpath issue when using URLClassLoader. I am on my oftly used netbook which is lacking in some departments such as a lack of free graphics acceleration. Right now I am working offline completely, there are wireless access points but I am not sure of which one would be the most secure. At least with fossil I still have access to all of my commits, which is good.

11:22

So the initialization of my HairballClassLoader does not trigger it.

11:41

Probably the best thing to do would be to just load the classes as I can and just avoid the issue without actually fixing the issue.

11:44

I should perhaps just instead add a compliance.FIXME as some kind of note of sorts for issues I am stuck on and just have to avoid. Although thinking about it, it is possible that the initialization of the compiler is chaning the context class loader perhaps so that after compiling my URLClassLoader which did work just gets wiped away.

11:47

So to figure this out, I will have to just NOT call the compiler.

11:49

Ok that is not it.

11:50

Actually forget that, I never rebuilt. Ok, so it may be linked to the compiler which may be messing with the stuff. So I was right on that so far by throwing an Error before the compiler is initialized.

11:54

Ok, so before I call "build" on the compiler it works and deletes. So how about just BEFORE the compiler task is initialized? And for safety it works before the task is initialized. Now to check after initialize but before the actual compilation.

11:57

Ok, so initialization is fine, I am now going to drop it AFTER the running of it, and it should trigger then. It should hit then, if not I have a very odd heisenbug. And I got failure as expected. So I need to determine what the compiler is doing before and after. It is possible it is changing the system class loader or similar.

12:02

Ok, so the class loaders I can easily know about work the same. It is possible that my JARs are being removed by the compiler. I am going to check that.

12:08

Ok so, my URLs are the same, the compiler is just doing something it should not be doing so that it causes the classes to not be found.

12:10

So the main issue is to either see the cause of it, guess, or try a work around by perhaps seeing if launching another thread does not cause said issue.

12:20

The irony of portals. The WAP at the hospital asks whether to agree or to decline the terms and conditions before using it, however selecting the terms and conditions ends up going back to the portal page.

12:33

Ok so running it in another thread does cause the issue. So the JavaCompiler does something which changes things. The JavaCompiler would be initialized by the class loader used by the system, while the one for hairball is used for it lower in the chain. So this has never happened before, I have changed my classes around. It also is possible that this has always existed. It is also possible that this has been fixed in a future version. However later versions do not run on my PowerPC systems at least. Also if it is fixed anyway, it would not be fixed anyway.

12:39

So it is quite possible that there is a kind of cross class loader with exceptions.

12:44

The code is in a finally block, so an exception was thrown, it just has not been logged because it is lost, it is most likely however a TODO.

12:48

It is possible that the exception wrapping my exception thrown by the Java compiler causes the issue.

at com.sun.tools.javac.main.Main.compile(Main.java:559)

So if I implement the code in the file manager which causes the unrolling may just fix it.

20:05

I believe I know the cause. I just need to check if it happens with zero. Anyway, basically UnknownError which is part of the system classloader is being used. Also, the JavaCompiler is also part of the system classloader. Hairball being run though is in its own URLClassLoader initialized by the bootstrap. And the DPC is part of the URLCL. So JamVM at least appears to run the exception handlers in the method that threw the exception and using its class loader, when it should be using the classloader of the class the handling method is in. So I will see what Zero does. If it has this issue too, then I will probably want to avoid duplicating this behavior since it seems like a rather bad bug. And Zero does it too. I just hope fixing the exceptions due to the incomplete JavaFileManager using PackageContents can fix such things. So hopefully by implementing the stuff I can avoid it, this will be a rather big gotcha. I could make a test case for this actually, a small set of code which can duplicate the behavior.