Disable saving traces

[shamefulCake1]

I don't seem to be able to find a way to disable traces in the debugger.

Is there a way to do this?

It takes a lot of time when debugging a process remotely (on Android).

[d-millar]

The short answer is no, but, having investigated this in the past, it's unlikely that recording the trace is actually the thing slowing you down. The volume of data passing over the wire is much more likely to be the time sink. Depending on what the actual cause is, this can be controlled.

As a test, I would "log enable gdb-remote packets" and see if you can tell what packets are being sent and when. The costliest transfers are modules and regions, but those should be triggered only when there is a change. Those events, if I remember correctly, can be disabled, possibly from the GUI (will have to verify this) but certainly from the command line. It's also obviously possible to modify the "hooks.py" file to minimize the requests set as part of "on_stop". Information in the GUI will be stale, but you can force a refresh on things you need when you need them.

[shamefulCake1]

It saves a trace when I click "disconnect".

This usually happens when the target has already crashed, and I want to re-connect to a new process. The saved trace of the old process stops being very useful at that time.

[d-millar]

You could maybe spell out in a little more detail how you re-connected to the new process and what you are trying to do with the old trace?

[shamefulCake1]

I am debugging an Android app, which I am using the "lldb attach" feature of the debugger.

Since the PID changes every time, I have tweaked lldbsetuputils.sh a little bit, but not much:

        target_pid=$(adb -s a740e4ea shell 'pgrep -f endor.qti.camera.provider-service')
	args+=("$OPT_LLDB_PATH")
	add-lldb-init-args
	args+=(-o "platform select '$target_type'")
	args+=(-o "platform connect '$target_url'" -o 'settings set plugin.jit-loader.gdb.enable off')
	add-lldb-pid "$target_pid"
	add-lldb-connect-and-sync "$rmi_address"
	add-lldb-extra-cmds
	add-lldb-tail-args
        args+=(-o "process handle --pass true --stop false --notify false SIGSEGV" -o "process handle --pass true --stop false --notify false SIGBUS" -o "settings set plugin.jit-loader.gdb.enable off" -o 'settings set plugin.jit-loader.gdb.enable off'   -o 'b ChiOfflineIPEUpscale::PostProcess'   -o 'process handle --pass true --stop true --notify true SIGCHLD' -o 'settings set plugin.jit-loader.gdb.enable off' -o 'process status') 

After this I am just pressing the "Re-launch ... using lldb Android (attach)".

The process stops at ChiOfflineIPEUpscale::PostProcess, I investigate it a bit, by examining the registers and stepping the instructions. Eventually the process crashes and the process is receiving SIGABRT.
After this point there is not much I can do with the old process/image, so I am either typing kill in the debugger and pressing the "Close the connection to the debugging agent" button, or right-clicking on the connection in the "Connections" pane and selecting "disconnect".

At this point Ghidra tries to save a trace, which is not useful and takes a long time (I presume, to fetch the full image from the device over USB).

[d-millar]

Done a little more digging, and I'm pretty sure (not 100%, but....) that this is not what's happening. The existing logic in hooks.py terminates the event loop when the process is no longer alive. There was a reason for this, but I think I have a better solution. If you want to try it out before it makes master:

• comment out lines 327-326

            if proc is not None and not proc.is_alive:
                break

• add the following after line 301:

        rc = listener.StartListeningForEventClass(
            util.get_debugger(), lldb.SBTarget.GetBroadcasterClassName(), ALL_EVENTS)
        if not rc:
            print("add listener for targets failed")
        rc = listener.StartListeningForEventClass(
            util.get_debugger(), lldb.SBProcess.GetBroadcasterClassName(), ALL_EVENTS)
        if not rc:
            print("add listener for processes failed")

• replace lines 496-497 with:

        enable_current_process()
        commands.put_processes()

• run gradle assemblePyPackage

When the event loop terminates, nothing happens afterwards. Correct me if I'm wrong - I assume the wait was not long but infiinite.

Also, if you think of it, let me know what version of lldb you're using. 21 has issues.