Work log 2016-08-12

Evaluation has slowed due to something being built at eval time :(

formatted was reading in clustered at eval time, so I’ve changed it.

Eval time cut from 1000+ seconds (> 30 mins) to ~120 seconds :)

Easy way to see what’s causing expensive building: everything expensive (pretty much) should be benchmarked. Put an exit 1 at the top of the benchmarking scripts and see what fails to evaluate.

Turns out failed was forcing some stuff.

TIP benchmarks are now generated at eval time; but that seems reasonable. I’ve moved its tests to a separate directory to ensure derivation so we don’t have to run them every time (they’re more like integration tests than unit tests).

Adding PDFs to Hydra too.