th-module-data is a library to allow template haskell authors to store state in a data log which can span multiple modules. In order to accomplish this, it outputs data files which mirror the haskell modules in which the relevant template haskell code is called. th-module-data came about as the result of this thread on the libraries mailing list.

The Env module provides the highest level interface, exporting a few TH functions which generate envGet_<namespace> and envPut_<namespace> functions. These function add an environment to the Q monad which can be used to store types or kinds or whatever else you might wish for.

The DataLog module provides a slightly lower level interface, allowing you to directly append to a log (represented with a Vector).

In order to properly import all the modules you depend on it is required to call dataLogInit_<namespace> before any of the other functions are invoked. Fortunatly, this funciton is idempotent, so you can call it as many times as you want without being worried that it might trample on any state. All the work is done the first time, and after that it just immediatly aborts.

The code itself has haddock comments, which are a good place to look for more information.

If there is a way to add rm -rf .stack-work/th-module-data to the stack clean command, the README should recomend doing that. I should add simmilar advice for cabal.

Efficent append? Mutable vectors? Should I just give up and use a list?

What is the right name for all these modules? Language.Haskell.TH.DataLog, Language.Haskell.TH.Env, and Language.Haskell.TH.File? The last one at least seems a little presumptuous.