Nixpkgs diff

Can I ask how you're doing the benchmarking? I don't have a lot of experience doing that with Haskell

Can I ask how you're doing the benchmarking? I don't have a lot of experience doing that with Haskell

Heap Profiling: compile the program with profiling (e.g. by using the --enable-profiling with cabal), then run the program with heap profiling enabled in the RTS, e.g. to break down by cost centre and output to an eventlog, +RTS -hc -l-au, then use haskellPackages.eventlog2html to visualize the heap profile.

Time profiling: almost the same, except you run with +RTS -p -l-au , then you can convert the eventlog to a json with haskellPackaes.hs-speedscope and load it into e.g. https://speedscope.app.

The GHC manual also has a section about profiling haskell programs:
https://downloads.haskell.org/ghc/latest/docs/users_guide/profiling.html

If you happen to have space leaks, you often can bring down max residency and run time at once, if you are lucky enough to find the space leak

lazy state is a huge red flag though, so that wasn't too hard, thankfully.

Very useful, thanks!

I have gotten nixfmt fast enough to feasibly run on hacke-packges.nix and even though the pretty printer is much faster than the parser, it still got a pretty bad space leak

The big brown triangle is labeled with Int, which means that it's probably some arithmetic that is not forced.

and this is how it looks when it's gone.

(yeah, the space leak of the parser is still pretty bad, I just now fixed set to force its arguments...)

DocE uses lists, which is a really bad datatype for these sorts of things, so there's another space leak (orange triangle in the bottom right)