LLDB breaks on adding a breakpoint using python

[abocquet]

Hello,

I wanted a LLDB breakpoint callback to add another breakpoint on the next frame in the callstack, but LLDB raises an exception on running the next breakpoint. Here is a minimal working example.

I have a file trace_mvp.py with the following code:

import lldb

def syscall_callback(frame: lldb.SBFrame, bp_loc, internal_dict):
    print("[+] Entering syscall_callback")
    thread = frame.GetThread()

    # print backtrace for debug
    for frame in thread.get_thread_frames():
        print(frame)

    # address of the next frame
    next_frame_pc = thread.get_thread_frames()[1].GetPC()
   
    print("[+] Add breakpoint at : ", hex(next_frame_pc))
    target = thread.GetProcess().GetTarget()
    bp = target.BreakpointCreateByAddress(next_frame_pc)
    bp.SetScriptCallbackFunction("trace_mvp.bp_callback")
    print(bp)


def bp_callback(frame: lldb.SBFrame, bp_loc, internal_dict):
    print("[+] Hello from bp_callback")

def __lldb_init_module(debugger, internal_dict):
    debugger.HandleCommand('breakpoint set -n write')
    debugger.HandleCommand('breakpoint command add -s python -F trace_mvp.syscall_callback')
    debugger.HandleCommand('run')  

This code adds a breakpoint on all write syscall. Then, at each stop it is supposed to add another breakpoint on the frame above in the stack, and print a message when it reaches it.

On the side I have a small Python file with a single print

print("Hello world")

In the terminal I run

# in bash
lldb python hello_world.py
# inside LLDB
command script import trace_mvp.py

On running the syscall_callback the breakpoint is added to the right address (here 0x101328b74) but LLDB still raises an error

Breakpoint 1: 137 locations.
Process 8128 launched: '/Users/macm1/.venv/bin/python' (arm64)
14 locations added to breakpoint 1
frame #0: 0x0000000188f758ec libsystem_kernel.dylib`write
frame #1: 0x0000000101328b74 libpython3.11.dylib`_Py_write_impl + 128
[rest of the stack]
[+] Add breakpoint at :  0x101328b74
SBBreakpoint: id = 2, address = libpython3.11.dylib[0x0000000000174b74], locations = 1
Traceback (most recent call last):
  File "<string>", line 8, in lldb_autogen_python_bp_callback_func__0
KeyError: 'lldb_autogen_python_bp_callback_func__1'
Process 8128 stopped

When I set the wrong function name on purpose, the error is not raised.
I'm running on macOS 15.1 with LLDB lldb-1600.0.39.109

This is a repost from https://stackoverflow.com/questions/79520679/lldb-breaks-on-adding-a-breakpoint-using-python, where Jim Ingham told me this was likely a bug and that I should open an issue.

[llvmbot]

@llvm/issue-subscribers-lldb

Author: Adrien Bocquet (abocquet)

Hello,

I wanted a LLDB breakpoint callback to add another breakpoint on the next frame in the callstack, but LLDB raises an exception on running the next breakpoint. Here is a minimal working example.

I have a file trace_mvp.py with the following code:

import lldb

def syscall_callback(frame: lldb.SBFrame, bp_loc, internal_dict):
    print("[+] Entering syscall_callback")
    thread = frame.GetThread()

    # print backtrace for debug
    for frame in thread.get_thread_frames():
        print(frame)

    # address of the next frame
    next_frame_pc = thread.get_thread_frames()[1].GetPC()
   
    print("[+] Add breakpoint at : ", hex(next_frame_pc))
    target = thread.GetProcess().GetTarget()
    bp = target.BreakpointCreateByAddress(next_frame_pc)
    bp.SetScriptCallbackFunction("trace_mvp.bp_callback")
    print(bp)


def bp_callback(frame: lldb.SBFrame, bp_loc, internal_dict):
    print("[+] Hello from bp_callback")

def __lldb_init_module(debugger, internal_dict):
    debugger.HandleCommand('breakpoint set -n write')
    debugger.HandleCommand('breakpoint command add -s python -F trace_mvp.syscall_callback')
    debugger.HandleCommand('run')  

This code adds a breakpoint on all write syscall. Then, at each stop it is supposed to add another breakpoint on the frame above in the stack, and print a message when it reaches it.

On the side I have a small Python file with a single print

print("Hello world")

In the terminal I run

# in bash
lldb python hello_world.py
# inside LLDB
command script import trace_mvp.py

On running the syscall_callback the breakpoint is added to the right address (here 0x101328b74) but LLDB still raises an error

Breakpoint 1: 137 locations.
Process 8128 launched: '/Users/macm1/.venv/bin/python' (arm64)
14 locations added to breakpoint 1
frame #<!-- -->0: 0x0000000188f758ec libsystem_kernel.dylib`write
frame #<!-- -->1: 0x0000000101328b74 libpython3.11.dylib`_Py_write_impl + 128
[rest of the stack]
[+] Add breakpoint at :  0x101328b74
SBBreakpoint: id = 2, address = libpython3.11.dylib[0x0000000000174b74], locations = 1
Traceback (most recent call last):
  File "&lt;string&gt;", line 8, in lldb_autogen_python_bp_callback_func__0
KeyError: 'lldb_autogen_python_bp_callback_func__1'
Process 8128 stopped

When I set the wrong function name on purpose, the error is not raised.
I'm running on macOS 15.1 with LLDB lldb-1600.0.39.109

This is a repost from https://stackoverflow.com/questions/79520679/lldb-breaks-on-adding-a-breakpoint-using-python, where Jim Ingham told me this was likely a bug and that I should open an issue.

[jimingham]

I have a fix for this, I just have to write a test case.

[jimingham]

#132846

[abocquet]

Thank you @jimingham!

[EugeneZelenko]

Pull request is not merged yet.

[jimingham]

Merged...