[egcs 1.0.1] i386-pc-solaris2.6 and -g

[egcs 1.0.1] i386-pc-solaris2.6 and -g

[Jeremiah W. James]

It appears that the -g option is not working quite correctly on my
i386-pc-solaris2.6 build of egcs 1.0.1.  I configured it with the
following options:

 --enable-shared --enable-threads --with-gnu-as

since binutils 2.8.1 was available.  I did not set BOOT_CFLAGS, or do
anything else out of the ordinary during the build, just a "make
bootstrap" followed by a "make install".

Programs compiled with -g cause the binutils 2.8.1 addr2line and objcopy
(and hence strip) to dump core.  The problem seems to be that a lookup
of the filename symbol doesn't succeed, returning a null pointer for the
filename.  Addr2line, in particular, then tries to print the filename,
causing it to seg fault in libc.  The strange thing is that nm shows the
filename symbols in there.  Has anybody else seen this?  If not, any
clues on what might be happening?  Is my egcs somehow corrupted?  Note
that programs compiled -g with Solaris cc and gcc 2.7.2 cause no
problems.

Thanks for any answers,
-- 
Jerry James
Email: jerry@cs.ucsb.edu
WWW:   http://www.cs.ucsb.edu/~jerry/

Re: [egcs 1.0.1] i386-pc-solaris2.6 and -g

[Jeffrey A Law]

  In message < tpr3eigdu1u.fsf@cs.ucsb.edu >you write:
  > It appears that the -g option is not working quite correctly on my
  > i386-pc-solaris2.6 build of egcs 1.0.1.  I configured it with the
  > following options:
I believe we use dwarf1 on that system by default (since that's what
the system compiler defaults to).

I highly suspect addr2line, objcopy and friends don't understand dwarf1
debug symbols as well as they need to.

jeff

Re: [egcs 1.0.1] i386-pc-solaris2.6 and -g

[Joe Buck]

[ after building egcs ]
> Programs compiled with -g cause the binutils 2.8.1 addr2line and objcopy
> (and hence strip) to dump core.  The problem seems to be that a lookup
> of the filename symbol doesn't succeed, returning a null pointer for the
> filename.  Addr2line, in particular, then tries to print the filename,
> causing it to seg fault in libc.

Even if this is an egcs bug, it is also a binutils bug.  Binutils programs
should not drop core even if presented with corrupt object files.

> 

Re: [egcs 1.0.1] i386-pc-solaris2.6 and -g

[Gavin Koch]

Jeffrey A Law writes:
 > I highly suspect addr2line, objcopy and friends don't understand dwarf1
 > debug symbols as well as they need to.

No they don't; it's on my to-do-when-I-get-a-chance list; 
but I wouldn't be crushed if someone else did it.

                                -gavin...