Re: [RFA] dwarf2_physname FINAL

["Ulrich Weigand" <uweigand at de dot ibm dot com>]

Keith Seitz wrote:

> There is an apparent regression in jprint.exp. Printing static class 
> variables (java only) is "broken." This is a gcc bug (gcc/43260): we 
> don't get any location info from gcc about where the static variable 
> lives, so we report it as "optimized out." NOTE: This "works" on CVS 
> HEAD because we fallback to using the linkage name (from 
> DW_AT_MIPS_linkage_name) to search the minimal symbol table, using that 
> to grab the address of the variable. Obviously we won't be able to do 
> that any more.

Would it be possible to KFAIL that?

I'm also seeing a second Java-related regression:
FAIL: gdb.java/jmisc.exp: ptype jmisc

This is because "ptype jmisc" now shows:

type = class jmisc  extends java.lang.Object {

    void main(java.lang.String[])void;
    void <init>(void)void;
}

instead of

type = class jmisc  extends java.lang.Object {

    void main(java.lang.String[])void;
    jmisc();
}

This is apparently caused by Java using DW_AT_MIPS_linkage_name to hide
the <init> magic name for the constructor:

 <2><223>: Abbrev Number: 6 (DW_TAG_subprogram)
     DW_AT_external    : 1
     DW_AT_name        : <init>
     DW_AT_MIPS_linkage_name: _ZN5jmiscC1Ev
     DW_AT_artificial  : 1
     DW_AT_declaration : 1

Is this to be expected?  Or should this be handled by the new code 
somewhere?


Also, I'm now seeing three failures in the new cpexprs.exp:
FAIL: gdb.cp/cpexprs.exp: list base::overload(void)
FAIL: gdb.cp/cpexprs.exp: setting breakpoint at base::overload(void)
FAIL: gdb.cp/cpexprs.exp: continue to base::overload(void)
apparently related to whether "const" needs to be specified when
identifying an overloaded method.  I seem to recall some reference
to this problem in one of the dwarf2_physname threads -- is this
still an expected problem with the final patch?
 
This is all on a RHEL5.4 with the system GCC 4.1.

Bye,
Ulrich

-- 
  Dr. Ulrich Weigand
  GNU Toolchain for Linux on System z and Cell BE
  Ulrich.Weigand@de.ibm.com