1483775 - Investigate un-inlining some parts of encoding_rs::mem

Status: NEW

(Core :: XPCOM, enhancement)

Description

[Henri Sivonen (:hsivonen) (not reading bugmail until 2026-07-27)]

A 0.13% installer size increase was attributed to bug 1402247. See
https://treeherder.mozilla.org/perf.html#/alerts?id=15036

Currently the Rust FFI forces a non-inlining boundary and in order to reduce the total number of non-inline function calls, the functions in encoding_rs::mem permit cross-crate inlining.

Once we get cross-language inlining from LLVM ThinLTO, we should revisit the inlining status of encoding_rs::mem in the interest of code size: Once ThinLTO can inline the FFI stubs, it'll probably make sense move the non-inlining boundary from FFI one step towards encoding_rs::mem. At that point, we could also remove inlined C++-side ASCII handling from the C++ wrappers in nsReadableUtils.h.

(Do we have a tool that compiles Firefox with both optimizations and debug info enabled and looks at the debug info to count the number of copies of each function from encoding_rs::mem?)