kache 0.7.0 improves Rust build cache hits on real codebases

kache 0.7.0 landed with a narrower ambition than a flashy release usually gets: it wanted fewer wasted rebuilds on Firefox, LLVM, and other messy trees where Rust shares space with C and C++. The project already called itself a zero-copy, content-addressed build cache for Rust, C/C++ and more, but 0.7.0 pushed past the easy win of sharing cache entries across machines and checkouts, which arrived in 0.6.0, and focused on making those entries actually hit in real codebases.

The team did that work with a nightly benchmark harness instead of a toy workload. kache’s docs say the suite runs against Firefox, LLVM, and Substrate, then measures hit rate, speedup, storage reuse, key stability, passthrough reasons, and logs. Firefox is the harshest test in the set: a mixed Rust and C/C++ build that can run for tens of minutes to hours and needs roughly 50 GB of scratch space. The workflow builds one fresh checkout, then rebuilds a second checkout at a different absolute path against the populated cache, which is exactly the sort of path churn that trips up real CI systems and local branches.

That benchmark work also forced the cache to understand compiler behavior more precisely. The release notes say the team modeled fast-math flags, -fomit-frame-pointer, x86 ISA selection, and -fansi-escape-codes so those settings would not poison cache reuse. They also taught the probe to read gcc and g++ -### output, which means codegen flags can now cache on gcc as well as clang. That matters in mixed-language repositories, especially when Rust is only one part of a larger build graph and the C or C++ side has its own flag soup.

kache is also trying to be easy to adopt, not just technically clever. The 0.7.0 release ships through Homebrew, apt, and WinGet, and the repository showed about 284 stars and 11 forks as the release went live. The project’s benchmark docs also compare kache-backed runs with sccache-backed runs, which gives Rust teams a familiar baseline instead of asking them to trust a new cache on faith. For anyone spending time in Cargo-heavy workflows with C/C++ dependencies alongside Rust, 0.7.0 reads less like another point release and more like an attempt to stop rebuilds where they usually start: in the flags, the paths, and the compiler details that waste an afternoon.