Re: [PATCH v2] Add gstack script

[Keith Seitz <keiths@redhat.com>]

On 12/13/24 11:45 AM, Tom Tromey wrote:
>>>>>> "Keith" == Keith Seitz <keiths@redhat.com> writes:
> 
> Keith> The script includes suggestions from gdb/32325 to include
> Keith> versioning and help. [If this approach to gdb/32325 is acceptable,
> Keith> I could propagate the solution to gcore/gdb-add-index.]
> 
> This would be super.

I will get that on my TODO list, then.

> 
> Keith> +function print_try_help() {
> Keith> +    echo "Try '$0 --help' for more information."
> 
> This should probably print to stderr.
> 
> Keith> +function print_help() {
> Keith> +    print_usage
> 
> Here, print_usage should print to stdout.  So maybe this later spot:
> 
> Keith> +# The sole remaining argument should be the PID of the process
> Keith> +# whose backtrace is desired.
> Keith> +if [ $# -ne 1 ]; then
> Keith> +    print_usage
> 
> ... should redirect to stderr.

Yeah, I will audit/adjust this.

> 
> Keith> +define attach-bt
> Keith> +attach \$arg0
> Keith> +echo "ATTACHED"
> Keith> +thread apply all bt
> Keith> +end
> Keith> +attach-bt $PID
> 
> I'm curious to understand why this is wrapped in a 'define'.

By wrapping in a command like this, we can abort doing the backtrace
if there was a problem attaching to a PID. Otherwise, if the attach
fails, using a series of "-ex" options won't abort, and the terminal
will be "spammed" with not-too-helpful error messages. Example:

$ gdb -batch -ex "attach 1234" -ex "bt"
ptrace: No such process.
No stack.

Admittedly, using "thread apply all bt full" does mitigate some of
this, but IIRC, I ran into other use cases where we would see
a additional, unhelpful error messages like demonstrated above.

Motivation:
Fedora's original version of gstack checked /proc/$PID and output an
error message if that didn't exist. This, of course, only works on
systems that support procfs. This is an attempt to include more than
just those systems.

Furthermore, (again with Fedora's original implementation) if gdb could
not attach to a valid PID (e.g., 1), Fedora's gstack simply exited
silently. This should fix that use case, too.

It's almost certainly not going to work everywhere, but it works pretty
well for Linux, and I figure that's a start.

Keith