Multi-Arch symbol read warning message

Multi-Arch symbol read warning message

[mrcompiler]

I am building a multi-arch version of GDB for handling ELF format core files 
for the ARM and PowerPC processors running on vxWorks. In order to get 
everything to build I combined the information from 
config/powerpc/vxworks.mt and config/arm/embed.mt and made the appropriate 
changes necessary to compile with GDB_MULTI_ARCH=GDB_MULTI_ARCH_PURE. 

The top level config is: 

./configure --target=arm-vxworks --enable-targets=strongarm-coff, 
powerpc-elf, strongarm-elf 

I was not able to get it working with --enable-targets=all since bfd 
couldn't resolve the appropriate architecture, but that's a secondary issue. 

I had to set the target to arm-vxworks instead of powerpc-vxworks in order 
to get the appropriate remote stuff to link. Otherwise it was off to hack 
the configure.in and related files which seemed like an extraordinary amount 
of unecessary work. 

Everything seems to work fine except for one, probably cosmetic, issue. When 
I load a PowerPC symbol-file I get the warning 

"unknown symbol descriptor `;`" which doesn't appear if I build a "pure" 
powerpc-vxworks target. 

Using gdb on gdb it seems like this is related somehow to the native target. 

I have two questions: 

1) does this indicate a significant issue? 

2) If not is there a solution other than suppressing the warning in 
dbxread.c? 

A prompt response would be appreciated greatly as my boss is breathing down 
my neck to make this available. I'd have to have my team debugging with a 
faulty debugger. 

Thanks,
Gene 

Re: Multi-Arch symbol read warning message

[Jim Blandy]

mrcompiler@mrcompiler.com writes:
> When I load a PowerPC symbol-file I get the warning "unknown symbol
> descriptor `;`" which doesn't appear if I build a "pure"
> powerpc-vxworks target.

This message means that the compiler has emitted debugging information
GDB doesn't know how to read.  (Or the compiler could be broken, or
GDB could be misreading correct debug info.)  If you set a breakpoint
on the appropriate 'complaint' call in dbxread.c:read_dbx_symtab and
'print namestring' when you get there, what does it say?

Re: Multi-Arch symbol read warning message

[mrcompiler]

Jim Blandy writes: 

> If you set a breakpoint
> on the appropriate 'complaint' call in dbxread.c:read_dbx_symtab and
> 'print namestring' when you get there, what does it say?

Jim, 

Thanks for the prompt response, the namestring value is below. 

Breakpoint 3, read_dbx_symtab (objfile=0x2b3c38) at dbxread.c:2024
2024                complain (&unknown_symchar_complaint, p[1]);
(top-gdb) print namestring
$1 = 0xbd089f "(10,7)=##(10,6);:;2A.;;" 

It complains about the ';' symbol, I believe its the one in front of 2A. 

Its kind of interesting that this only occurs when I specify the arm-vxworks 
target, if I build a powerpc-vxworks target it doesn't happen. 

It turns out that if I set the stop-whining value higher there are a number 
of other messages (several thousand, most of which seem to dups with 
different memory locations, possible the result of multiple library links or 
something). 

I picked some at random and included an example of our build flags (we're 
using gcc from Wind River). I don't see any problems from this, but that may 
be because we use this version of GDB only for our core files. 

It may also be significant that the host is Solaris. 

Here's the supported architecture info: 

(gdb) info architectures
Supported architectures are:
[arm] [armv2] [armv2a] [armv3] [armv3m] [armv4] [armv4t] [armv5] [armv5t]
[armv5te] [xscale] [powerpc:common] [rs6000:6000] [rs6000:rs1] [rs6000:rsc]
[rs6000:rs2] 

Here's an example of our build flags: 

/vob/TORNADO/host/sun4-solaris2/bin/ccppc -pipe -ansi -I. -DCPU= PPCEC603   
 -mcpu=PPCEC603 -mstrict-align -msoft-float -DVXWORKS -D__vxworks__=1 
 -fno-builtin -nostdinc -nostdlib -fvolatile -fno-defer-pop -g -O2 
 -Wpointer-arit h  -I/vob/ip/include  -I/vob/infra/include 
 -I/vob/TORNADO/target/h -Wp,-MD,.deps/TORNADO/target/src/wv/mumble.d 
 -fno-builtin   -c /vob/TORNADO/target/src/wv/mumble.c -o 
TORNADO/target/src/wv/mumble.o 

Here's some other example of "problem" namestring values, 

1: "(100,39)=##(100,38);:;2A.;;"
1: "(100,46)=##(100,45);:;2A.;;"
1: "(100,52)=##(100,51);:;2A.;;"
1: "(101,
1: "(101,12)=##(101,11);:;2A.;;"
1: 
"(101,12)=#(101,6),(101,9),(101,9),(0,20);:__Q310radix_node4$_834$_84;2A.;;, 
0,96;\\"
1:
"(101,19)=#(101,13),(101,16),(101,16),(0,20);:__Q310radix_node4$_834$_85;2A. 
;
;,0,96;\\" 

Re: Multi-Arch symbol read warning message

[Daniel Jacobowitz]

On Fri, Feb 21, 2003 at 07:16:49PM -0800, mrcompiler@mrcompiler.com wrote:
> Jim Blandy writes: 
> 
> >If you set a breakpoint
> >on the appropriate 'complaint' call in dbxread.c:read_dbx_symtab and
> >'print namestring' when you get there, what does it say?
> 
> Jim, 
> 
> Thanks for the prompt response, the namestring value is below. 
> 
> Breakpoint 3, read_dbx_symtab (objfile=0x2b3c38) at dbxread.c:2024
> 2024                complain (&unknown_symchar_complaint, p[1]);
> (top-gdb) print namestring
> $1 = 0xbd089f "(10,7)=##(10,6);:;2A.;;" 
> 
> It complains about the ';' symbol, I believe its the one in front of 2A. 
> 
> Its kind of interesting that this only occurs when I specify the 
> arm-vxworks target, if I build a powerpc-vxworks target it doesn't happen. 

At an off-the-wall guess, the problem is that ARM tools use
continuation characters in stabs and PowerPC tools don't.  Are you sure
a plain arm-vxworks debugger doesn't show the same thing?

-- 
Daniel Jacobowitz
MontaVista Software                         Debian GNU/Linux Developer

Re: Multi-Arch symbol read warning message

[Andrew Cagney]

> I am building a multi-arch version of GDB for handling ELF format core files for the ARM and PowerPC processors running on vxWorks. In order to get everything to build I combined the information from config/powerpc/vxworks.mt and config/arm/embed.mt and made the appropriate changes necessary to compile with GDB_MULTI_ARCH=GDB_MULTI_ARCH_PURE.

Hmm, cute.

I posted a WIP patch for this several years ago but never got around to 
integrating it (there wasn't sufficient immediate needed).  Now that 
most architectures are multi-arch, it is definitly time to integrate 
such a change.

Can you contribute this stuff?

Andrew

Re: Multi-Arch symbol read warning message

[mrcompiler]

Andrew Cagney writes: 

> 
>> I am building a multi-arch version of GDB for handling ELF format core 
>> files for the ARM and PowerPC processors running on vxWorks. In order to 
>> get everything to build I combined the information from
> 
> Hmm, cute.

This sort of sounds like I might using multi-arch in a different from 
intended ( or normal to date ). I did notice that most of the other stuff 
was supporting a single processor class across multiple OS environments. 

Is there any reason this shouldn't work? 

I also had an ambiguity problem with --enable-targets=all in bfd and had to
specifically build those with --enable-targets=strongarm-coff, powerpc-elf, 
strongarm-elf ( possibly arm-elf, I don't have access to the build tree at 
this instant). 

> I posted a WIP patch for this several years ago but never got around to 
> integrating it (there wasn't sufficient immediate needed).  Now that most 
> architectures are multi-arch, it is definitly time to integrate such a 
> change.

You mean a patch for my symbol read warnings? Can you point me to it? 

> 
> Can you contribute this stuff? 
> 

Yes, I was planning on talking to you folks about this once I got it 
working. Obviously I don't want to make my first contribution a 
brain-damaged hack. Not that I'm necessary opposed to brain-damaged hacks as 
anyone who has downloaded my stuff from freshmeat can attest.... 

Gene