Benchmark performance of shaping individual words.
Don't cache results of main function unless processed (neglected in previous commit).
Tell Haskell that its cheaper to recompute the data than store 40x the input text.
Ensure I can pass commandline args to Criterion.
Found some more performance optimizations.
Use dupablePerformIO where relevant, & it doesn't incur excessive allocation.
Ensure amounts of output are dereferenced as quickly as large amounts are.
Fix multi-threaded benchmark, see if its any faster.
Turns out no, the degree varies with free cores.
Add benchmark, against Dracula by Bram Stoker.
Release Harfbuzz-Pure 1.0.3.2 (got version-number wrong last time)
Merge branch 'main' of git.argonaut-constellation.org:~alcinnz/harfbuzz-pure
Expose exception & update metadata.
Release harfbuzz-pure 1.0.2.1
Attempt fix regarding fontscale bitsizes.
Add contribution instructions.
Implement more typeclasses & ensure empty input don't throw errors.
Expose HarfBuzz 3.3 APIs, attempt 2.