java/decl.c patch for clear_binding_level

Zack Weinberg zackw@Stanford.EDU
Sat Mar 17 23:44:00 GMT 2001

On Sun, Mar 18, 2001 at 01:20:26AM -0500, Jeff Sturm wrote:
> On 17 Mar 2001, Per Bothner wrote:
> > I just did a libjava 'make check' and am finding that almost
> > all bytecode compilation tests are failing, with the compiler
> > crashing.
> Yep.  The bytecode compiler is completely broken on the trunk.  This seems
> to fix it:

Um, I see why you'd want to do that - it seems odd to be initializing
the start_pc field with LARGEST_PC and the end_pc with 0 - but when I
make that change, or swap start_pc and end_pc in struct binding_level,
I get *more* failures.

Current trunk + Alexandre's -c -o fix + --enable-threads:

                === libjava Summary ===
# of expected passes            1384
# of unexpected failures        78
# of unexpected successes       12
# of expected failures          246

Same plus your patch:

# of expected passes            919
# of unexpected failures        262
# of unexpected successes       89
# of expected failures          450

Now we did get a bunch more unexpected successes, too, so maybe your
patch is correct but exposes a bug somewhere else...

All the unexpected failures in both test runs are 'compilation from
bytecode' or 'byte compilation'.  Most of the unexpected successes are
'compilation from source'.


More information about the Java-patches mailing list