Bojun-Seo wrote:

> I think the consensus in [this 
> thread](https://discourse.llvm.org/t/rfc-introduction-of-doublefreesanitizer-dsan/91363/16)
>  was having a dedicated tool that is a subset or very closely overlapping of 
> a range of other tools (ASan, Scudo, GWP-ASan, LSan, etc.) in compiler-rt 
> doesn't make much sense. You've evaluated against the no-instrument-ASan 
> option, but did you manage to prototype the **out-of-tree option**? Why does 
> it need to be part of compiler-rt? Because you do not need compiler 
> instrumentation, the flag `-fsanitize=doublefree` is a convenience and not 
> technically needed, and a pure runtime tool can be standalone out-of-tree.

Following @vitalybuka 's request, I am no longer planning to add this as a 
dedicated tool in compiler-rt. Instead, I am implementing double-free detection 
as an extension to LSan.

Your points are very reasonable. While there are trade-offs, I also agree that 
keeping this functionality out-of-tree is a viable and good option.

I would like to gather more community feedback on the RFC. Between the two 
approaches — extending LSan or maintaining it out-of-tree — which one do people 
prefer? I will follow the consensus.

https://github.com/llvm/llvm-project/pull/213846
_______________________________________________
cfe-commits mailing list
[email protected]
https://lists.llvm.org/cgi-bin/mailman/listinfo/cfe-commits

Reply via email to