Skip to content

Latest commit

 

History

History
48 lines (31 loc) · 1.87 KB

README.md

File metadata and controls

48 lines (31 loc) · 1.87 KB

Repo for semi automatic benchmarking of ghc branches making use of hackage-head

Usage: The various (supported) libraries can be found under lib.

Each library has a driver script (driver.sh) which will try to setup and run the packages benchmarks. This is an eternal catchup game so often some will be broken, PRs welcome!

Running the benchmarks

All the drivers take a bunch of flags:

driver.sh [<ghc-exec> [<ghc-alias> [<ghc-flags> [<flag-alias>]]]]

  • ghc-exec is the compiler used
  • ghc-alias is a name to reference it by in file names and the like
  • ghc-flags are compilation options to be passed to ghc if any
  • flag-alias is again a name to reference these with.

All of these have defaults which are:

  • ghc-exec: ghc
  • ghc-alias: vanilla
  • ghc-flags: ""
  • ghc-flags: <ghc-alias>

The benchmark results will be written in the criterion-created csv format under lib//benchResults/..csv

Other important information

Some of the packages use Symlinks, so if you use windows make sure you set up your system correctly. See here for details: https://www.joshkel.com/2018/01/18/symlinks-in-windows/

In general things should work at the bleeding edge but I don't actively maintain this. If you want to benchmark against a snapshot over a longer period of time then I suggest to adjust the driver to clone a specific commit instead.

The drivers also instruct cabal to use different caches per ghc-alias. This is important since two compilers built with different settings might end up using the same cache otherwise.

Other stuff.

If you end up using this in some fashion let me know. Knowing people actually use it would go a long way towards me actively maintaining this.

I also welcome contributions if they don't fundamentally change how the setup works, or maybe even if depending on the change.