This is the mail archive of the gcc-patches@gcc.gnu.org mailing list for the GCC project.


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: [tree-ssa] merge to 20030923


On Sat, 27 Sep 2003, Richard Henderson wrote:
> (2) I couldn't merge Java at all.  The changes for unit-at-a-time
>     from mainline conflicted heavily with the gimplification changes
>     on the branch.

I'm finished with the conflicts and am about to commit a Java merge.
There is a new test failure:

FAIL: assignment_2 compilation from bytecode

The ICE doesn't mean much given the woeful state of the bytecode compiler
on tree-ssa.  However there is a regression in the .java->.class compiler
that must also be present on mainline.  For assignment_2.java, the
premerge gcj would create bytecodes:

Method name:"foo" public Signature: 11=()int
Attribute "Code", length:16, max_stack:1, max_locals:2, code_length:4
  0: iconst_0
  1: istore_1
  2: iload_1
  3: ireturn

Postmerge gcj generates:

Method name:"foo" public Signature: 11=()int
Attribute "Code", length:41, max_stack:1, max_locals:4, code_length:21
  0: iconst_0
  1: istore_1
  2: jsr 14
  5: goto 17
  8: astore_3
  9: jsr 14
 12: aload_3
 13: athrow
 14: astore_2
 15: ret 2
 17: iload_1
 18: istore_2
 19: iload_2
 20: ireturn
Exceptions (count: 1):
  start: 0, end: 2, handler: 8, type: 0 /* finally */

I.e. the useless finally block is no longer killed as before.

Jeff


Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]