See Demo.stt for notes on usage, goals and benchmarks.
See krakow-pres.pdf for a presentation I previously gave about this system.
NOTE: as of november 2020, the implementation here is obsolete in some respects. The evaluator can be greatly simplified along the lines of this. Here's also a video where I talk about conversion checking using the simplified glued evaluator. I might update this repo at some point to reflect newer best practices.
Some benchmarks
Benchmarks are informal and unfortunately not yet automated or easily reproducible. Loading Demo.stt causes smalltt benchmark times (but not memory usage) to be displayed, but there's no equivalent for Agda and Coq now.
Setup:
smalltt is compiled with GHC 8.6.2, with -O2 -fllvm
GHC 8.6.2 is likewise used for GHC/GHCI benchmarks.
Coq 8.8.2 for Coq, with -type-in-type
Agda 2.6.0-master, built in 2018 september, is used for Agda,
again with --type-in-type
System: Core i7 3770, 8GB RAM, Ubuntu 16.04
Timings and memory usage measured wth "time" util and process monitoring, or
by using a stopwatch in the case of Agda normalization.