Question about JIT-based record and replay for rr Debugger

[IamHuskar]

Can rr Debugger avoid using CPU hardware features and instead use JIT trace tools like pintools/dynamorio to provide the data needed for recording and replay?

It seems that pinplay and undodb implement timeless debugger using a JIT approach, and there are also some implementations using qemu.

[rocallahan]

There's a version of rr like that here: https://github.com/sidkshatriya/rr.soft
But rr was designed to avoid that, and we're trying to stick to avoiding that.

[IamHuskar]

There's a version of rr like that here: https://github.com/sidkshatriya/rr.soft But rr was designed to avoid that, and we're trying to stick to avoiding that.

thanks @rocallahan

Another question: Does rr currently support, or are there plans to support, starting record and replay from an arbitrary checkpoint (e.g., a saved program state), rather than being required to start execution from the very beginning (the program's entry point) for every recording? (like intel pinpoints https://www.intel.com/content/www/us/en/developer/articles/tool/pin-a-binary-instrumentation-tool-pinpoints.html )

The motivation is that for very large and complex software, the overhead of recording and replaying from the absolute start can be significant or impractical.

[dzaima]

See #3406

[IamHuskar]

See #3406

It looks like this pull request has been open for three years without being merged; I wonder what the reason is?

This is exactly the feature I’ve been looking for. Since Intel PinPlay doesn’t support ARM architecture and is closed-source, I’m wondering if rr could add checkpoint functionality, with recording and replay data provided by JIT tracing tools like Pin or DynamoRIO.

This way, rr could run on virtual machines, cloud hosts, and even on platforms like Android/Windows.