Is it possible to preserve hardware watchpoint across re-runs in the same gdb session?

Question

Suppose I start a gdb session, and create a breakpoint and run.

After I break, I create a watchpoint based on the memory address of a symbol in the current execution, and delete the original breakpoint.

Some time later, I interrupt the program with Control-C, still inside gdb, I issue the run command to restart the program from the beginning.

However, I would like to preserve the hardware watchpoint across the restart of the debugged process.

Is there a gdb setting that allows me to preserve hardware watchpoints across reruns?

Update: Here is an example to reproduce the problem.

int main(){
    int NeverGoOutOfScope = 0;
    NeverGoOutOfScope = 7;
    while (1);
}

Here are the sequence of gdb commands.

break 3
run
watch NeverGoOutOfScope
info watch 
run 
# After this point, the breakpoint remains but the watchpoint is gone.
info watch

Is it possible to preserve the watchpoint without having to recreate it?

Nope. The watchpoints disappear. The breakpoints may be preserved. — merlin2011
I'm getting the warning Error in re-setting breakpoint NUMBER: No symbol "variable_name" in current context, but it leaves the watchpoint in the breakpoints list: NUMBER hw watchpoint keep y variable_name and its value is getting printed in the next run... — W.F.

Tom Tromey Tom Tromey · Accepted Answer · 2016-05-04T20:00:01

It should work with two conditions: first, that you disable address space randomization; and second, that you use watch -location.

To disable randomization, use:

(gdb) set disable-randomization on

I think this is the default, so you might be able to skip this.

Next, use:

(gdb) watch -location NeverGoOutOfScope

This will record the address of the variable in the watchpoint. This is almost always what you actually want -- gdb has some fancy code to re-evaluate watchpoint expressions when a variable goes out of scope, but in practice I think this is more confusing than helpful.

This approach works for me on your test, though I have to ignore a few hits of the watchpoint during process initialization.

Is it possible to preserve hardware watchpoint across re-runs in the same gdb session?

1 Answers