rr-debug-logic

Debug Logic Bug with rr

Use this skill when a program produces wrong output, returns an incorrect value, or behaves incorrectly without crashing. Unlike crashes (which have a clear failure point), logic bugs require you to first identify WHERE the wrong value becomes observable, then work backwards to find WHERE the computation went wrong.

Workflow

1. Record the Program

rr_record(command=$ARGUMENTS, trace_dir="<project>/rr-trace-<random>")

The program will run to completion (it doesn't crash). rr captures everything.

2. Start Replay

rr_ps(trace_dir="<trace>")
rr_replay_start(trace_dir="<trace>", pid=<pid>)

3. Set Breakpoints at the Observable Bug

You need to stop execution at the point where the wrong behavior is visible. Strategies:

If you know the function that returns the wrong value:

rr_breakpoint_set("compute_result")
rr_continue()

Step through and find where the return value is wrong.

If you know the line that produces wrong output:

rr_breakpoint_set("output.cpp:42")
rr_continue()

If you know the condition that should hold but doesn't:

rr_breakpoint_set("process_data", condition="result < 0")
rr_continue()

If the bug is in iteration N of a loop:

rr_breakpoint_set("loop_body.cpp:100", condition="i == 42")
rr_continue()

4. Confirm the Wrong Value

At the breakpoint, examine the state:

rr_locals()
rr_evaluate("result")
rr_evaluate("expected_value")

Verify that the value is indeed wrong at this point. Save a checkpoint:

rr_checkpoint_save()

5. Trace the Value Backwards

Now work backwards to find where the wrong value originated.

For a simple variable:

rr_watchpoint_set("result")
rr_continue(reverse=True)

This stops at the exact write that set result to its current (wrong) value.

For a computed value (e.g., a + b is wrong): Check which input is wrong:

rr_evaluate("a")
rr_evaluate("b")

Then watchpoint the wrong input:

rr_watchpoint_set("a")
rr_continue(reverse=True)