GSoC 2025 - Feedback Crash Profiler Proposal

[Kumkum]

Hi all,

I'm a GSoC 2025 aspirant currently working on a crash-aware feedback profiling tool for V8, based on a real crash caused by the flag `--turbo-collect-feedback-in-generic-lowering`. I’ve submitted a GitHub demo with a reproducible case. I would love to get feedback on whether this direction could benefit V8 development or embedders. Thanks!

Kumkum Kaushik
GitHub: https://github.com/justttkumkum/v8-startup-profiler

[Jakob Kummerow]

I think this project needs more focus.

I'm not sure what "crash-aware feedback profiling" means or in what way it would be useful. I'm also not sure what connection you see between type feedback collection and profiling. Generally speaking, crashes need to be fixed, not accommodated.

That said, experiments may sometimes be crashy, and that may be okay, depending on the state and anticipated future of the experiment: if it's served its purpose and will be discarded, then fixing it first isn't worth the time. I don't know the current state of --turbo-collect-feedback-in-generic-lowering.

When you talk about "development of a lightweight profiling [...] framework", it's also not clear in which way that would be different from the existing profiling frameworks we have.

So I'd recommend to focus on one problem you'd like to solve, and then be more specific in how you want to go about solving it.

(I can't really comment on the GSoC aspect as I'm not involved with the planning. I'm not aware of any V8-specific GSoC projects, but I might just not know.)

--
--
v8-dev mailing list
v8-...@googlegroups.com
http://groups.google.com/group/v8-dev
---
You received this message because you are subscribed to the Google Groups "v8-dev" group.
To unsubscribe from this group and stop receiving emails from it, send an email to v8-dev+un...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/v8-dev/f0b881d1-fb93-4994-ac8d-ad47badd7ebbn%40googlegroups.com.

[Kumkum]

Thank you so much for your detailed feedback.

 I really appreciate you taking the time to go through my proposal!

You're right, the scope and terminology could be clearer. Based on your suggestions, I’m planning to refocus the project on a more concrete and useful problem: building a deterministic crash reproduction harness for experimental flags in V8. This tool would help identify and reproduce crashes more reliably when using flags like --turbo-collect-feedback-in-generic-lowering.

The idea is not to accommodate crashes but to make them easier to diagnose and fix, especially for contributors exploring internal flags. I’ll also remove the loosely defined parts like "feedback profiling" and make the goals more specific.

Thanks again for the guidance — it means a lot!

Best regards,
Kumkum Kaushik

To view this discussion visit https://groups.google.com/d/msgid/v8-dev/CAKSzg3TqLHm4YVrQjFipWF7Kw48AdXogR_vSSSvSyS9BDLn2sg%40mail.gmail.com.