Skip to content

Performance Notes

This section tracks performance-oriented algorithm evolution by version.


Current Version

v0.9.6 sends every keyed match (dense val<N> literals, lit(...) literals and std::variant alternatives, on borrowed and owned subjects) through one switch and always inlines the whole path, so these matches compile to the code of a hand-written switch on GCC as well as Clang. It also cuts GCC's debug information about 8×. See v0.9.6 performance note for details.

v0.9.5 lets the bare spellings is<T> and alt<I> reach the static variant dispatch fast path. See v0.9.5 performance note for details.

v0.9.3 tuned the variant inline dispatch threshold for common workloads. See v0.9.3 performance note for details.

v0.8.2 established the lowering engine baseline. See v0.8.2 performance note for benchmark data and algorithm details.

Releases from v0.8.3 to v0.9.2, and v0.9.4, focused on guard fixes, API ergonomics, and compiler compatibility. No new dispatch algorithms were introduced.


Version Index

  • v0.9.6 (one switch for every keyed match; owned subjects; GCC debug size)
  • v0.9.5 (bare is<T> / alt<I> reach the variant fast path)
  • v0.9.3 (variant dispatch threshold tuning)
  • v0.8.2 (lowering engine baseline)
  • v0.8.0 (tiered variant dispatch)

Scope

Performance notes focus on:

  • dispatch strategy evolution
  • fast-path and cold-path behavior
  • algorithm-level tradeoffs and applicability

For API compatibility and migration notes, see: