Re: Intrusive GDB Symbol Lookup when debugging remotely

[Dave Trollope <trollope@lucent.com>]

Hello,

No-one responded to this, but I'm sure someone on this list understands 
how difficult it would be to implement a new handler for a signal to run 
gdb in a "Local" mode when connected to a gdbserver. I'd really 
appreciate an estimate of how much effort this would be, or where in the 
code to start looking.

Cheers
Dave
David Steven Trollope wrote:

> Hello!
>
> I'm trying to determine if there exists an alternative solution, or a 
> patch for the following problem, or if someone familiar with the GDB 
> source can estimate how much work it would be to implement a patch. 
> Any advice would be welcome and appreciated.
>
>
> Configuration:
>
> GDB running on a Sun Solaris machine.
> GDBServer running on a PowerPC Linux machine
> The application is running on the target in GDBServer and GDB is 
> connected to the target using "target remote x.x.x.x:y"
>
> Problem:
>
> While the target is running, I don't think it is possible to examine 
> the symbol table from within gdb without sending an interrupt signal 
> to GDB. This is a problem because users (and frontends) don't seem to 
> be able to look up symbols in the symbol table in GDB without 
> interrupting the target.
>
> In theory, it would be possible to stop the interrupt being propogated 
> to the server and hence the target could remain running without 
> interruption. One possible way of solving this would be to use a 
> different signal as a "Non-intrusive interrupt" or "Local Interrupt". 
> The symbol lookup can occur, and then gdb resumes. This would imply 
> gdb would need to know if its running in "Local" or "Remote" mode to 
> prevent operations trying to access the target.
>
> An alternative mechanism could be to have a separate thread that runs 
> and is signalled for non-intrusive operations.
>
> You might ask why not just run a separate gdb instance to do this kind 
> of look up? However, I am dealing with large executables where loading 
> separate instances for the sake of a symbol lookup is neither 
> efficient, timely or practical.
>
> I would very much appreciate discussion on this topic with regard to 
> feasibility, if its been done already or how much effort it would be 
> to implement this type of functionality.
>
> Cheers
> Dave Trollope
>