BorrowSanitizer

[tomassedovic]

Metadata
Contact	@icmccorm
Team(s)	compiler (@RalfJung), infra, lang (@tmandry), opsem (@RalfJung)
Goal document	2026/borrowsanitizer
Funding contact	Funding team

Summary

We are building BorrowSanitizer: an LLVM-based instrumentation tool for finding violations of Rust's aliasing model. In 2026, we want to make it feature-complete and useful in practice.

---

This issue is intended for status updates only.

For general questions or comments, please contact a goal owner or post in the Zulip topic.

[tomassedovic]

This issue is intended for status updates only.

For general questions or comments, please contact the owner(s) directly.

[icmccorm]

May has been busy for BorrowSanitizer! We just posted our monthly status update. Here's the TL;DR on what our team has been up to:

• Our retag function calls are starting to make their way upstream, with #156208 and #156210.

• We have expanded our test suite, replicating many of the cross-language Tree Borrows errors that we found in our prior work with MiriLLI.

• We have a new benchmarking dashboard for tracking the performance of the LLVM components of BorrowSanitizer, in a "no-op" mode. This inserts our instrumentation, but it does not execute any run-time checks. This was originally a summer project by @Gitter499.

• We are using Miri to test out optimizations that we hope to bring to BorrowSanitizer. This effort has been lead by @mojeanmac in collaboration with @shinhae-kim from Cornell. We created a new allocation strategy for Miri that leads to a 6.7% speedup. We are also looking into using metadata from Tree Borrows to determine when to run Miri's garbage collector. This has lead to considerable improvements in performance—up to a 171.3% speedup (63% time reduction) for one crate.

• We posted an RFC for BorrowSanitizer's LLVM components.

[icmccorm]

Here's our June status update! TL;DR:

• Added support for Tree Borrows' wildcard provenance, expanding our test suite to cover 93% of relevant tests from Miri's pass and fail directories, up from 85% last month.
• Implemented our initial deferred reference counting garbage collector.
• Upgraded our diagnostics to point to relevant user code.

We also updated our benchmarking dashboard to compare our full instrumentation against Miri, and our retag intrinsics are nearly finished!

[icmccorm]

Here's our July status update! TL;DR:

• We optimized our garbage collector, discovering new heuristics that we can use to improve Miri, too!
• We found two new cross-language aliasing bugs! BorrowSanitizer is ready to try out on real libraries.

We plan on submitting an MCP for upstreaming BorrowSanitizer as an experimental component in nightly between now and RustConf in September.

[icmccorm]

Instead of a longer blog post, we decided to prioritize preparing our presentation for RustConf next week. We're looking forward to seeing y'all in Montreal!

Here's everything else that's new:

• Our RFC for upstreaming BorrowSanitizer on the LLVM side has been accepted. We intend for BorrowSanitizer to become a general-purpose tool for validating provenance in LLVM. Our next step is an MCP for upstreaming BorrowSanitizer on the Rust side.

• Thanks to feedback from @cmazakas and @Wybxc, we found a couple of new false positives that show up in release mode. These have been related to packed structs (#315) and tail recursion elimination (#306). We're looking to have fixes ready by RustConf.

• @mojeanmac has been busy preparing several of our GC optimizations for upstreaming in Miri. She'll post a more detailed status update soon.