> Could you share a more complete recipe for the benchmark? I wonder how it
> compares to Gerd's weak vector/freelist in scratch/igc.

Put it into the `benchmark` subdir of `elisp-benchmarks` and then do

    emacs --batch -Q -l .../elisp-benchmarks/elisp-benchmarks-autoloads.el \
          --eval '(elisp-benchmarks-run "bytechar")'


- Stefan