Getting gdb to find definitions

Getting gdb to find definitions

[Paul Koning]

There's a problem I'm seeing increasingly often with GDB 6.1 --
especially now that our compilers are switching to Dwarf-2 format.  It
seems that gdb doesn't load definitions of types until you reference
some variable that causes it to load the symbols for a particular
source file -- or something like that.

For example, I might do this:

(gdb) file netbsd.gdb
Reading symbols from netbsd.gdb...done.
(gdb) ptype struct user
No struct type named user.
(gdb) p curproc
$1 = (struct proc *) 0x8282e9e0
(gdb) ptype struct user
type = struct user {
    struct pcb u_pcb;
    struct pstats u_stats;
}

This is causing a lot of confusion among users, and it breaks various
scripts.

I would argue that gdb should read the definitions of types when they
are mentioned (as it does with the definitions of variables).  Either
that, or it shouldn't default to delayed loading, if it causes
malfunctions like this.

Is there a fix for this?  Alternatively, is there a way to at least
change the default symbol loading rule?

       paul

Re: Getting gdb to find definitions

[Daniel Jacobowitz]

On Tue, Jun 29, 2004 at 11:41:46AM -0400, Paul Koning wrote:
> There's a problem I'm seeing increasingly often with GDB 6.1 --
> especially now that our compilers are switching to Dwarf-2 format.  It
> seems that gdb doesn't load definitions of types until you reference
> some variable that causes it to load the symbols for a particular
> source file -- or something like that.
> 
> For example, I might do this:
> 
> (gdb) file netbsd.gdb
> Reading symbols from netbsd.gdb...done.
> (gdb) ptype struct user
> No struct type named user.
> (gdb) p curproc
> $1 = (struct proc *) 0x8282e9e0
> (gdb) ptype struct user
> type = struct user {
>     struct pcb u_pcb;
>     struct pstats u_stats;
> }
> 
> This is causing a lot of confusion among users, and it breaks various
> scripts.
> 
> I would argue that gdb should read the definitions of types when they
> are mentioned (as it does with the definitions of variables).  Either
> that, or it shouldn't default to delayed loading, if it causes
> malfunctions like this.

GDB is supposed to do this.  If you've found a case where it doesn't
work, please do provide a testcase.

> Is there a fix for this?  Alternatively, is there a way to at least
> change the default symbol loading rule?

-readnow.

-- 
Daniel Jacobowitz