The cost of using the `concurrency` abstractions

As a follow up of Testing concurrent code using DejaFu, I spent some time investigating whether/how the use of the concurrency abstractions impacts performance vs. using plain IO and STM. As somewhat expected, there’s an impact, but specialization recovers native performance. There’s some use of the inspection-testing library as well for good measure.

See concurrency and Performance »

Let this be a plea for authors of libraries exposing STM datatypes etc. to use the concurrency abstractions, so programs using these libraries can still use DejaFu in testing :wink:


Great to see inspection-testing in action!