Clojurians Log v2

Hi! I want to discuss how to use Clerk to gain visibility into stateful, external processes (and whether it’s a good idea). I feel like what I’m asking for is a bit muddy. But I’d rather ask possibly vague questions than hold my silence. PROBLEM STATEMENT: MENTAL MODEL FOR STATEFUL RUNNING PROCESSES & CLERK I’m torn about something. 1. I find Clerk to be very effective to work with deep, composite values. For example, using the table viewer to explore what a database really contains. a. I recently used Clerk as my “database browser” in a project where I previously would have used https://www.postgresql.org/docs/current/app-psql.html + https://tableplus.com/. I liked that I gained control (Clojure at my fingertips), and explanatory power (previous “what’s in the database?” questions were answered in notebooks). 2. I struggle with how I should architect my code to play nicely with Clerk when I have an external stateful resource. a. A recent struggle example of this was me learning to use Playwright from Clojure with java interop. At first, I tried doing it with Clerk. Then I realized that i had to “start a playwright”, then “start a browser”, then “create a page”. All actions with side effects, and return values I needed to keep track of. I ended up stopping Clerk and continuing in a normal REPL. i. Just a normal REPL felt quite great here, in contrast to a compile-rerun-see-what happens cycle, asking 3 browsers to all show one page, controlled from the REPL! ii. The current state of my code a mess of functions and defs that have to be run in a specific order. I’d love to be able to pull out a Clerk-style nice literate explanation, but I’m not sure how to get there. b. My current diagnosis of my struggles is that (A) there is a good solution to this, (B) I’m just currently not “holding Clerk right” yet. I recall having heard @mkvlr comment on core.async as being opaque. In contrast, in Erlang, one can query the system for the current running processes. I feel like this is the problem I encountered with Playwright. The running system state was left invisible. I wasn’t sure when to def, when to defn, and when to (def ,,, (atom ,,,)). I felt like I ended up with a bit of a mess. Two specific things I’m interested in: • Whether I should be leaning into using Clerk when working with stateful systems at all, or just use the REPL. ◦ Since Clerk can be configured to run only when the user asks for Clerk to run (using the function nextjournal.clerk/show!), my hunch is: ▪︎ (A) yes, clerk is suited, ▪︎ (B) avoid using clerk/show! on namespaces that aren’t written to play nicely with Clerk’s caching behavior, and def-ed values could require resource cleanup. • Thoughts on whether Clerk is a tool that can help make “external processes” visible, or if the caching & literal features make that hard. ◦ I realize that Smalltalk-like object browsers may perhaps be a better suited tool for this, as they (per my understanding) don’t usually rely on caching and immutability for their representation. ▪︎ Then again, I do generally prefer the immutable approach when one is possible. ◦ After all, Clerk’s tagline is Moldable Live Programming for Clojure, not Literate Programming For The Cases When Everything Is Immutable! (😸)