What is this?
rr is a lightweight tool that records the execution of an application — entire trees of processes and threads — and lets you replay and debug that recording.
rr is a lightweight tool that records the execution of an application — entire trees of processes and threads — and lets you replay and debug that recording.So do I debug with rr instead of gdb, or alongside it?Alongside: rr extends gdb, so you record a run once and then debug the replay in gdb, which gains very efficient reverse-execution on top of standard features.
Alongside: rr extends gdb, so you record a run once and then debug the replay in gdb, which gains very efficient reverse-execution on top of standard features.What kind of machine and OS does it need to run?Linux with kernel 4.7 or newer and a supported CPU: Intel Nehalem (2010) or later, certain AMD Zen or later processors, or certain AArch64 microarchitectures.
Linux with kernel 4.7 or newer and a supported CPU: Intel Nehalem (2010) or later, certain AMD Zen or later processors, or certain AArch64 microarchitectures.Can I use it on macOS or Windows too?No, rr is Linux-only. Even the Apple Silicon M-series chips mentioned in the README are supported as AArch64 hardware under rr's Linux requirements, not on macOS.
No, rr is Linux-only. Even the Apple Silicon M-series chips mentioned in the README are supported as AArch64 hardware under rr's Linux requirements, not on macOS.Can I run it inside a VM or a cloud instance?Yes, as a VM guest, as long as the VM virtualizes hardware performance counters: some VMware, KVM, AWS and GCP configurations are known to work, while no Xen setup is.
Yes, as a VM guest, as long as the VM virtualizes hardware performance counters: some VMware, KVM, AWS and GCP configurations are known to work, while no Xen setup is.What does reverse-execution actually give me when chasing a bug?You replay the recording in gdb and can run backwards to reach the cause of a failure, while hardware data watchpoints show where memory changed — the README says this combination makes debugging much more fun.
You replay the recording in gdb and can run backwards to reach the cause of a failure, while hardware data watchpoints show where memory changed — the README says this combination makes debugging much more fun.Does it work with multithreaded programs?Yes — rr records and replays whole trees of processes and threads, so multithreaded applications are exactly what it is built for.
Yes — rr records and replays whole trees of processes and threads, so multithreaded applications are exactly what it is built for.Is there a deeper technical writeup on how rr works?Yes — the README points to the paper 'Engineering Record And Replay For Deployability: Extended Technical Report' as the best technical overview of the project.
Yes — the README points to the paper 'Engineering Record And Replay For Deployability: Extended Technical Report' as the best technical overview of the project.What are similar projects?pwndbg, zenc, ir, gdbgui, open-gpu-kernel-modules, lytlnyblOS, TurboBench, NoGraphicsAPI
pwndbg, zenc, ir, gdbgui, open-gpu-kernel-modules, lytlnyblOS, TurboBench, NoGraphicsAPI