nanobind 2.11.0 introduced some internal changes which break MPI code.I am yet uncertain what the underlying cause is, but nanobind added immutability to internal classes, which trigger certain fast paths in CPython and might cause issues with MPI.
I have bisected the nanobind upstream repository and traced the i386 unit test regression to https://github.com/wjakob/nanobind/commit/ac63d3a63dfec1be5c1129fa6be3ee35d5ac13ed
[Implicit nb::none() annotation for std::optional<T>-typed arguments] -- ⢀⣴⠾⠻⢶⣦⠀ ╭────────────────────────────────────────────────────╮ ⣾⠁⢠⠒⠀⣿⡁ │ Timo Röhling │ ⢿⡄⠘⠷⠚⠋⠀ │ 9B03 EBB9 8300 DF97 C2B1 23BF CC8C 6BDD 1403 F4CA │ ⠈⠳⣄⠀⠀⠀⠀ ╰────────────────────────────────────────────────────╯
signature.asc
Description: PGP signature
