Does gdb support weak/normal symbols?

Does gdb support weak/normal symbols?

[H. J. Lu]

[-- Attachment #1: Type: text/plain, Size: 700 bytes --]

I got

[hjl@gnu-10 weak]$ make
cc -o foo -O -g main.c foo.c weak.c
for f in foo; do ./$f; done
Normal foo
[hjl@gnu-10 weak]$ gdb foo
GNU gdb 6.4.50.20060115-cvs
...
(gdb) b foo
Breakpoint 1 at 0x804839a: file weak.c, line 8.
(gdb) r
Starting program: /export/home/hjl/bugs/gdb/weak/foo
Normal foo

Program exited normally.
(gdb) b *0x08048380
Breakpoint 2 at 0x8048380: file foo.c, line 5.
(gdb) r
Starting program: /export/home/hjl/bugs/gdb/weak/foo
Breakpoint 1 at 0x8048386: file foo.c, line 6.

Breakpoint 2, foo () at foo.c:5
5       {
(gdb)

foo in weak.c isn't in the symbol table. But gdb sets the breakpoint
of foo on it. Shouldn't gdb be clever enough to know to check the
other one?

H.J.

[-- Attachment #2: bug.tar.gz --]
[-- Type: application/x-gzip, Size: 466 bytes --]

Re: Does gdb support weak/normal symbols?

[Daniel Jacobowitz]

On Sun, Jan 15, 2006 at 04:48:31PM -0800, H. J. Lu wrote:
> foo in weak.c isn't in the symbol table. But gdb sets the breakpoint
> of foo on it. Shouldn't gdb be clever enough to know to check the
> other one?

GDB uses the debug information in preference to the ELF symbol table. 
That one says this is a definition of foo.

Eventually "break foo" will breakpoint both of them.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Does gdb support weak/normal symbols?

[H. J. Lu]

On Sun, Jan 15, 2006 at 08:49:04PM -0500, Daniel Jacobowitz wrote:
> On Sun, Jan 15, 2006 at 04:48:31PM -0800, H. J. Lu wrote:
> > foo in weak.c isn't in the symbol table. But gdb sets the breakpoint
> > of foo on it. Shouldn't gdb be clever enough to know to check the
> > other one?
> 
> GDB uses the debug information in preference to the ELF symbol table. 
> That one says this is a definition of foo.
> 
> Eventually "break foo" will breakpoint both of them.

What did you mean by "Eventually"? Were you saying that it was a bug
and would be fixed in the future? Is anyone working on it? I'd like
to help.


H.J.

Re: Does gdb support weak/normal symbols?

[Daniel Jacobowitz]

On Mon, Jan 16, 2006 at 07:47:54AM -0800, H. J. Lu wrote:
> On Sun, Jan 15, 2006 at 08:49:04PM -0500, Daniel Jacobowitz wrote:
> > On Sun, Jan 15, 2006 at 04:48:31PM -0800, H. J. Lu wrote:
> > > foo in weak.c isn't in the symbol table. But gdb sets the breakpoint
> > > of foo on it. Shouldn't gdb be clever enough to know to check the
> > > other one?
> > 
> > GDB uses the debug information in preference to the ELF symbol table. 
> > That one says this is a definition of foo.
> > 
> > Eventually "break foo" will breakpoint both of them.
> 
> What did you mean by "Eventually"? Were you saying that it was a bug
> and would be fixed in the future? Is anyone working on it? I'd like
> to help.

Please read the archives for the last few days.  This is covered by the
same general problem description as "break FILE:LINE" - you have asked
for an ambiguous breakpoint, and instead of properly placing multiple
breakpoints or prompting GDB has picked one at random.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Does gdb support weak/normal symbols?

[H. J. Lu]

On Mon, Jan 16, 2006 at 10:58:46AM -0500, Daniel Jacobowitz wrote:
> On Mon, Jan 16, 2006 at 07:47:54AM -0800, H. J. Lu wrote:
> > On Sun, Jan 15, 2006 at 08:49:04PM -0500, Daniel Jacobowitz wrote:
> > > On Sun, Jan 15, 2006 at 04:48:31PM -0800, H. J. Lu wrote:
> > > > foo in weak.c isn't in the symbol table. But gdb sets the breakpoint
> > > > of foo on it. Shouldn't gdb be clever enough to know to check the
> > > > other one?
> > > 
> > > GDB uses the debug information in preference to the ELF symbol table. 
> > > That one says this is a definition of foo.
> > > 
> > > Eventually "break foo" will breakpoint both of them.
> > 
> > What did you mean by "Eventually"? Were you saying that it was a bug
> > and would be fixed in the future? Is anyone working on it? I'd like
> > to help.
> 
> Please read the archives for the last few days.  This is covered by the
> same general problem description as "break FILE:LINE" - you have asked
> for an ambiguous breakpoint, and instead of properly placing multiple
> breakpoints or prompting GDB has picked one at random.
> 

"break foo" may be slightly different from "break FILE:LINE". For the
global symbol, foo, there will be only one definition. It is a
matter of picking the right foo for "break foo". There is no need to
set multiple breakpoints nor prompt user.


H.J.

Re: Does gdb support weak/normal symbols?

[Daniel Jacobowitz]

On Mon, Jan 16, 2006 at 10:11:31AM -0800, H. J. Lu wrote:
> "break foo" may be slightly different from "break FILE:LINE". For the
> global symbol, foo, there will be only one definition. It is a
> matter of picking the right foo for "break foo". There is no need to
> set multiple breakpoints nor prompt user.

This is wrong.  You're still thinking about a compiler and a linker,
not about a debugger.  There may also be a dozen static functions named
"foo", and one of them may happen to be in the current file and the
user may just have been looking at it.

Since we can't read the user's mind, we have to do our best to silently
accomodate the user's potential intentions.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Does gdb support weak/normal symbols?

[H. J. Lu]

On Mon, Jan 16, 2006 at 01:30:52PM -0500, Daniel Jacobowitz wrote:
> On Mon, Jan 16, 2006 at 10:11:31AM -0800, H. J. Lu wrote:
> > "break foo" may be slightly different from "break FILE:LINE". For the
> > global symbol, foo, there will be only one definition. It is a
> > matter of picking the right foo for "break foo". There is no need to
> > set multiple breakpoints nor prompt user.
> 
> This is wrong.  You're still thinking about a compiler and a linker,
> not about a debugger.  There may also be a dozen static functions named
> "foo", and one of them may happen to be in the current file and the

How can a static function be global?



H.J.

Re: Does gdb support weak/normal symbols?

[Daniel Jacobowitz]

On Mon, Jan 16, 2006 at 10:32:13AM -0800, H. J. Lu wrote:
> On Mon, Jan 16, 2006 at 01:30:52PM -0500, Daniel Jacobowitz wrote:
> > On Mon, Jan 16, 2006 at 10:11:31AM -0800, H. J. Lu wrote:
> > > "break foo" may be slightly different from "break FILE:LINE". For the
> > > global symbol, foo, there will be only one definition. It is a
> > > matter of picking the right foo for "break foo". There is no need to
> > > set multiple breakpoints nor prompt user.
> > 
> > This is wrong.  You're still thinking about a compiler and a linker,
> > not about a debugger.  There may also be a dozen static functions named
> > "foo", and one of them may happen to be in the current file and the
> 
> How can a static function be global?

This is not C++.  There's no One Definition Rule.  The existence of a
global function "foo" somewhere in the executable does not preclude the
existence of a static function "foo" somewhere else, in the same
binary, in a dynamically loaded library, et cetera.

In fact it's passably likely.

-- 
Daniel Jacobowitz
CodeSourcery

Re: Does gdb support weak/normal symbols?

[Daniel Jacobowitz]

On Mon, Jan 16, 2006 at 10:32:13AM -0800, H. J. Lu wrote:
> On Mon, Jan 16, 2006 at 01:30:52PM -0500, Daniel Jacobowitz wrote:
> > On Mon, Jan 16, 2006 at 10:11:31AM -0800, H. J. Lu wrote:
> > > "break foo" may be slightly different from "break FILE:LINE". For the
> > > global symbol, foo, there will be only one definition. It is a
> > > matter of picking the right foo for "break foo". There is no need to
> > > set multiple breakpoints nor prompt user.
> > 
> > This is wrong.  You're still thinking about a compiler and a linker,
> > not about a debugger.  There may also be a dozen static functions named
> > "foo", and one of them may happen to be in the current file and the
> 
> How can a static function be global?

This is not C++.  There's no One Definition Rule.  The existence of a
global function "foo" somewhere in the executable does not preclude the
existence of a static function "foo" somewhere else, in the same
binary, in a dynamically loaded library, et cetera.

In fact it's passably likely.

-- 
Daniel Jacobowitz
CodeSourcery