Importing GNU Classpath 0.90

Fri Mar 10 21:15:00 GMT 2006

Hi,
On Fri, 2006年03月10日 at 15:14 -0500, Bryce McKinlay wrote:
> You'll probably find that, in addition to the VMCompiler vs. MD5 thing

Sorry for not recognizing that part. I made a bc-compiled eclipse and
made sure that it works (and loads the bc-compiled classes) with the
merge.
> there are a couple more issues with an import right now:
>> - GCJ bug resulting in a link error in Swing code - 
> http://gcc.gnu.org/PR26390

Yes. I tried making that package bc compiled but that triggered a bug in
the verifier. I added the issue to the report. And I have ignored this
package for now (it is the not-really-functional-yet swing-awt-peers).
> - Dependence on Java 1.5 Character implementation. Anthony Balkissoon is 
> currently working on updating libgcj's Character, paralleling his work 
> on Classpath's implementation.

Yes, I talked to Tony on irc to make sure I could use the patch that Tom
made for the 4.1 branch to get the regex stuff working. And that was OK
(and makes the new regex stuff compile).
> These are all pretty easy to work around, of course.

They were. There was one more issue with the new Math/VMMath split which
wasn't completely trivial to get right. So I just made the old Math.java
into an override file. We need to add support for VMMath and the new
Math methods later.
Here is the final result:
2006年03月09日 Mark Wielaard <mark@klomp.org>
 Imported GNU Classpath 0.90
 * scripts/makemake.tcl: Set gnu/java/awt/peer/swing to ignore.
 * gnu/classpath/jdwp/VMFrame.java (SIZE): New constant.
 * java/lang/VMCompiler.java: Use gnu.java.security.hash.MD5.
 * java/lang/Math.java: New override file.
 * java/lang/Character.java: Merged from Classpath.
 (start, end): Now 'int's.
 (canonicalName): New field.
 (CANONICAL_NAME, NO_SPACES_NAME, CONSTANT_NAME): New constants.
 (UnicodeBlock): Added argument.
 (of): New overload.
 (forName): New method.
 Updated unicode blocks.
 (sets): Updated.
 * sources.am: Regenerated.
 * Makefile.in: Likewise.
OK to commit?
I tested against the libgcj testsuite of course and mauve. I also made
sure eclipse runs interpreted and bc-compiled with this. And that
MegaMek starts up properly :)
There is one new issue however. I had hoped that this would give us out
of the box https support. But there seems to be some bug that
miscompiles something in the new code:
http://gcc.gnu.org/bugzilla/show_bug.cgi?id=26638
This isn't a regression of course since before we didn't even have this
code.
Full diff (including the classpath sources and generated files) can be
found at: http://www.klomp.org/mark/classpath/libgcj-0.90.diff.gz
If someone wants to test it out. All my testing was done on
i686-pc-linux-gnu.
Cheers,
Mark
-------------- next part --------------
A non-text attachment was scrubbed...
Name: 0.90-merge.diff
Type: text/x-patch
Size: 84081 bytes
Desc: not available
URL: <http://gcc.gnu.org/pipermail/java/attachments/20060310/67dc4fd7/attachment.bin>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: signature.asc
Type: application/pgp-signature
Size: 189 bytes
Desc: This is a digitally signed message part
URL: <http://gcc.gnu.org/pipermail/java/attachments/20060310/67dc4fd7/attachment.sig>