To an outsider, it seems a bit more like NIH than performance. Relying on GHC internals also seems like a regression and a liability more than an improvement.
To an outsider, it seems a bit more like NIH than performance
yamlwasn’t written by me
libyaml(the C library) wasn’t written by me
streamlywasn’t written by me
99% of the code was written by others
Looks like there is already a Pure Haskell YAML Parser.
Yes, it’s mentioned in the very intro section of the blog and it’s 10 times slower than
yaml, which caused a noticeable slowdown in ghcup and issues in pandoc.