I just watched a thought-provoking video by Dave Thomas:
The Best OO Language is a Functional One
The talk brings together ideas that Dave Thomas has been working on for a while. It touches on functional and object-oriented programming, hardware speed and power consumption, state, time, transformations, etc. Quite interesting and tasty…
I’m particularly pleased by the continuing evolution of Dave’s tooling, eg:
- Diet supports architecting programs as sets of transformation recipes.
- Diet.Debug allows time-traveling debugging over trees of execution paths.
- Jeeves wraps (sets of) GenServers in a tasty OO-like DSL.
I also watched a (predictably) interesting talk by José Valim:
Introducing HDD: Hughes Driven Development
This talk brings together the topics of code formatting and property-based testing (aka generative testing). I like the fact that José (like Dave) encourages folks to forage through odd corners of computer science lore, pulling out cool and useful ideas (and giving credit freely and generously to the lore’s originators).
The implementation of these ideas in Elixir is already raising the level of discourse and practice; it will continue to do so in the future. As it becomes easier to create and manage distributed sets of processes, we’ll find more need for features such as back pressure (ie, GenStage), truth maintenance (ie, Phoenix Presence), etc.
FWIW, Diet and Jeeves look like a nice way to solve a problem I’ve been considering. I’m putting together a system (AxAp) that tries to make various kinds of documents more accessible. For example, it currently handles EPUB3 and some text file formats. Obviously, this involves a lot of transformations. More to the point, many of the transformations can be used in multiple processing chains.
Assuming that the necessary transformations have been encoded, it should be possible to perform dependency analysis and generate a recipe for any producible output. Of course, DIET and Jeeves would be on hand to handle the ugly details…