Do you want to know why it takes so long for your Haskell project to compile?

YES

Use this shell script in a stack project

In this repo there's a shell script stack-passes that reads from your .stack-work directory.

Dependencies?

You'll need textql. You can get it from sudo apt install textql in Debian or Ubuntu.

Ok, how do I make it go?

Change into a directory where you've built a stack project and run stack-passes.

That script will create a new file ghc-passes-grouped.csv. Copy that file into your clone of this project, run a web server in this directory, and load it in your browser.

For example, the Haskell warp server will serve the folder on http://localhost:3000/ or python3 -m http.server behaves similarly, if you have python installed.

How do I turn this information into actions to reduce my compile time?

For incremental builds, try to make your module structure as tree shaped as possible. When you change a module that all other modules use, all modules must be recompiled. The more tree shaped your package dependencies, the less time spent doing incremental compiles when making changes.

GHC has many flags to get more information about what goes on in particular phases. The graphs generated by stack-passes can be used as a first step, to find out which phases and modules dominate your compile time, and then you get more verbose output for those sections.

Example Output

When run on the lens package, stack-passes produces the following graph:

If you compile lens with stack --fast (equivalent to GHC's -O0), you instead get the graph below. Note that the Simplifier bars which dominated the graph above have shrunk to nothing. The Simplifier phase is where most of GHC's optimization work occurs.

When you load a module in ghci, it is type checked, but no code is generated until you actually want to run it. So the CodeGen bars, which dominate the second graph, also can be skipped.