GNU Classpath Linuxtag meeting minutes

[Mark Wielaard]

GNU Classpath meeting minutes
LinuxTag, 12 July 2003, Karlsruhe (Germany)

Present
- Dalibor Topic (Kaffe)
- Sascha Brawer (GNU Classpath)
- Mark Wielaard (GNU Classpath)
- Andy Wolter (Aicas)
- James Hunt (Aicas)
- Ingo Proetel (Aicas)

Sascha was the most organised and said someone should actually make
notes, so here they are. Some of the notes were made in the Biergarten
so take them with a grain of salt... I have surely forgotten something
important. And sometimes I couldn't clearly remember the context of my
notes. My applogies. If you were present and have more details please
follow up with clarifications and additions.

- Sascha has been working on font reading and rendering, see his very
  detailed architecture document, which describes all issues involved
  <http://www.dandelis.ch/development/fonts/>.

  Tim Tyler his public domain, pure java font rendering program was
  mentioned <http://fonteditor.org/>
  Adam Megacz of <http://www.xwt.org/> also needs this
  <http://gcc.gnu.org/ml/java/2003-07/msg00036.html>.

- For the above and AWT/Swing in general the issue of
  accessing/mapping frame/display buffers came up multiple times. What
  interface should be defined for mapping these (java.nio JNI
  additions?).

- OpenGL seems to have all the 2D drawing primitives that you ever
  want and implementations are widely available.

- Aicas wanted to know who are interested in the Realtime specs being
  discussed in the OpenGroup and the jconsortium
  <http://www.opengroup.org/rtforum/rt_java/>
  <http://www.jconsortium.org/> Thet were also looking for
  participants in an open European standards project (which name I
  forgot to write down).

- Aicas is working on some RMI fixups. During the discussion it came
  up that since object streams use serialization and serialization
  uses reflection, this only will work if you get all the details of
  reflection right. It was pointed out that reflection has some
  underspecified issues/bugs, for example see this note from Kissme
  (Class.getModifiers):

  // Arrays are always final and never an interface.
  // And also abstract (but that is actually a bug
  // - see JDC Bug: 4208179)
  // It matters for serial version UID calculations :{

  There was much complaining about the broken design of the RMI
  protocol. It was suggested to look at how RMI can work on IIOP which
  is cleaner.

- Sascha suggested to create a list of such issues and publish a list
  for GNU Classpath users/VM implementors of
  tricky/underspecified/buggy specs.

- Mark wants to extend the VM interface with more VMState package local
  final classes like to JRVM people just did for a couple of java.lang
  classes to make it easier to use GNU Classpath with non-JNI/Posix
  based VMs.

- The plan for character encodings is to move to the java.nio.charset
  interface. We already have required encodings for this. But GNU
  Classpath and gcj both still also have their old implementations
  (which are actually used in most places). gcj also has a libiconv
  provider (but not as java.nio.charset provider). A java.nio.charset
  libiconv provider would be nice to have for those systems that
  have that library.

- Mark and Dalibor wanted to name next month Mauvegustus and clean up
  Mauve (it still contains some bogus/broken tests) and import as much
  tests from the different projects that have "non-standard" core
  library tests so they can be more easily shared between the
  different projects.  Ingo dared Mark to create a Mauve plugin for
  Eclipse now that we have it running like the standard JUnit
  plugin. (The JUnit framework uses lots of reflection which we don't
  want to use for Mauve.) Other ideas were to create a standard JNI
  testsuite and maybe security issues/holes tests.

- Acias has property files which lists which class/method needs which
  other class/method. This is nice when you are use smart or static
  linking of classes.

- Kaffe has split up their library in different profiles (core, awt,
  xml, sound, rmi, etc) it would be nice if GNU Classpath also
  provided this. We could deliver different jar files.

- We need to update/document the state of each package. It is not
  really clear how much has been implemented, signature complete
  (japitools), tested (has mauve tests), how compliant with which
  spec, etc. packages/classes are. Problem is that we don't have
  gatekeepers anymore per package (we had this in the past, but
  developers on the project changed and the number of supported
  packages has actually exploded).

- The above is also important to show what people can do to help the
  project. We really need to setup a more visible task list then the
  current Savannah one with task people could do for the project which
  are annotated with how hard it would be it (simple -> follow
  existing example, like image provider or new character encoding,
  medium -> add something to an existing framework, like jar file
  verification for ZipFile, hard -> add complete new package or
  infrastructure).

- We should have a standard NotYetImplementedException. Just returning
  something random (like null) from stubs is really not
  acceptable. Kaffe for example has one.

- We really have a complete framework now to run significatly large
  applications. Getting semi-free application written in the java
  language (but which currently uses a proprietary runtime
  environment) running with a free VM is a great way to test and/or
  expand our classes and keeps us focused on the needs of free software
  developers.

  It would be really useful to have a "closure tool" which someone that
  has developed an application that he or she wants to get working with
  a GNU Classpath based VM can run and that tells possible problem
  areas/missing methods/classes that can then be used as a request for
  help on implementing the missing pieces.

- How well does Antlr work and can we use it to generate parsers for
  things? (There was some talk about Antlr on the RHUG
  <http://sources.redhat.com/rhug/> mailinglist recently
  <http://sources.redhat.com/ml/rhug-rhats/2003-07/msg00001.html>).

- There was some talk about sharing different things then just the
  core libraries such as jitters or a common shared compiled library
  format between VMs. Can be discussed on the GNU Classpath list at
  first, but will need a separate mailinglist if there is to much off
  topic (non-core classes) discussion.

- The GNU Classpath homepage contains a list of free (non-)core
  libraries that are useful for developers and/or VM implementors. Is
  something missing?

- Everybody had a really good time and looked forward to meet again in
  the future. Dalibor suggested the Linux-Kongress 2003 which will be
  from October 14 to October 16, 2003 in Saarbreucken, Germany, where
  he lives.