P2996 - compilation time benchmarks

[kris-jusiak]

Some initial exploration of the P2996 meta-programming model in terms of compilation times and comparison to different approaches such as boost.mp11, circle, etc. Thought it could be beneficial to share it here.

Libraries

• boost.mp11

Proposals

• P2996 - Reflection for C++26
• P1858 - Generalized pack declaration and usage

Compilers

• clang-p2996
• circle

Results

• https://boost-ext.github.io/mp

Notes

• Since the meta-programming is based on consteval the performance of abstract machine has a huge impact.
• Meta-programming with P2996 can be done either with ranges or without. The former is much faster to compile.
• Compiler support is still in the early stages, and while compilation is likely to improve, at this point, there is a huge difference between P2996 and, for example, boost.mp11.

Repository

• https://github.com/boost-ext/mp/tree/benchmark

Missing

• https://cppmeta.codereckons.com/ (in progress)
• EDG-p2996 implementation (@daveedvdv is there a way to run EDG-reflection locally somehow?)