The Duality of Transducers
I finally got around to re-recording and posting this talk on Clojure’s transducers that I gave last year to the Austin Clojure Meetup:
The talk walks through what transducers are, their benefits, where they make sense to use (and where they don’t), and how to implement them from scratch.
The title refers to an idea that really helped make transducers click for me: namely that there are two different conceptual models of transducers that I needed to apply (to different contexts) to really get it. (This reminded me a lot of the wave-particle duality of light in physics, which describes a single underlying reality in two different ways, with each way tending to prove more practical in analyzing particular scenarios).
The two models, then, are:
The encapsulated transformation model, where the transducer is an opaque representation of a (likely-compound) transformation of a collection, which merges with other transformations via the mechanical application of
comp. And…The constructive model, where we’re dealing in the underlying machinery of transducers (say, implementing a transducible process), and it’s helpful to conceptualize a transducer simply as a function from reducing-function to reducing-function (
rf -> rf).
I hope that comes through clearly in the presentation.
If you don’t use transducers in your Clojure code today, I highly suggest you do—you will see benefits. And I’m convinced that the best way to get them to really click is to implement them from scratch, which the video will walk you through.
For your convenience, here are the slides for the presentation. The links are not clickable there (sorry), but are all included in the video description on YouTube.
If you have any questions, feedback, or need mentorship, feel free to reach out on Bluesky, Mastodon, or X and I’d be happy to help.