Well, okay Shriram said the term "homoiconicity" is fuzzy, but the underlying thing (the language of data and the language of programs being the same representation) is real and worth taking seriously. Citing him to wave away the whole concept is pretty wild misuse of the citation. They are refining the claim, not negating it.
> doesn't gain anything semantically
Syntactic properties create semantic affordances. The reason "code as data" matters isn't that the parentheses look a certain way - it's that the AST is a first-class citizen of the runtime data model. Those are semantic capabilities - they determine what programs can express and compute, especially at the meta level. The syntactic uniformity is the mechanism; the ability to write programs that construct and transform other programs, using the same tools as everything else, is the payoff.
Homoiconicity doesn't make Lisp programs mean something different, but it gives you a more uniform and powerful meta-programming model.
I don't disagree with this. Benjamin Pierce defines type-checking in the opening pages of Types and Programming Languages as an operation over syntax, for example.
My point was that the parent comment just kind of threw out "homiconicity" when somebody talked about writing properties about a language in that language, and those are entirely separate things. I was addressing a conflation of terms. The property that people generally refer to as "homoiconicity" is useful for things like writing macros, but it does not directly grant you access to any kind of property-checking capabilitiy. I mean, Rust's macro system is far from homoiconic (it's just an AST), but it gives you semantic capabilities. You know?
What a thought-terminating cliché. It sounds reasonable while saying nothing. "Useful ideas, but..." what? "But it's not popular"? Neither was Python before it become one. But it doesn't have a big ecosystem? You're describing a consequence of adoption and presenting it as a cause.
Like I said: Clojure runs significant chunks of Walmart's infrastructure, Nubank's entire banking stack serving several hundred million customers - for a second, Nubank is the biggest digital bank in the world; Apple uses it; Cisco uses it. Emacs Lisp runs one of the most enduring pieces of software in computing history. You you ever used Grammarly - it's powered on Lisp; This very site runs on Lisp.
"I don't encounter Lisp in my work and that feels meaningful to me" (there's no other way to interpret your words) is just another, personal opinion that has no practical, objective angle. "The Curse of Lisp" opinion, at least back in the day had some objective relevance that no longer holds true.
I think you're confusing newcomers landscape of "which library do I pick", (which is real) with stretched evidence for fundamentally fragmented ecosystem.
Let's walk through your examples:
SQL libraries: Korma is dead, YeSQL is dead, HugSQL is in maintenance mode. The community converged on next.jdbc + HoneySQL. That's not fragmentation - that's exploration followed by selection.
Build tools: Boot had its moment, it's essentially gone. Leiningen is still around but new projects overwhelmingly use deps.edn. That's not fragmentation either - it's a transition, and a remarkably clean one.
Spec vs. Schema vs. Malli is the most interesting case because the convergence is still in progress. Schema is effectively in maintenance mode. Spec has stalled (spec2 never really shipped). Malli is clearly gaining momentum as the community choice for new work, largely because it treats schemas as data — which is very Clojure. But even here, these aren't competing implementations of the same thing the way, say, Javascript build tools are. They have meaningfully different design philosophies, and the community is converging on the one that best fits Clojure's values.
Compare it to JS, where we have Webpack, Vite, esbuild, Parcel, Rollup, Turbopack, and the "winner" changes every 18 months. Or Python's packaging story - pip, pipenv, poetry, conda, PDM, uv - which is still a mess after decades with a vastly larger community. Fragmentation in large ecosystems often doesn't self-correct because there's enough mass to sustain competing approaches indefinitely.
The small community size that you frame as a weakness is actually why the self-correction works. There aren't enough people to sustain five competing SQL libraries, so the best one wins and everyone moves on. The tight community, the shared values around simplicity and data orientation, and the fact that people actually talk to each other - these create selection pressure that larger, more diffuse ecosystems lack.
The productivity gains aren't speculatory - from the most pragmatic angles, I see time and again - they are real and palpable.
> doesn't gain anything semantically
Syntactic properties create semantic affordances. The reason "code as data" matters isn't that the parentheses look a certain way - it's that the AST is a first-class citizen of the runtime data model. Those are semantic capabilities - they determine what programs can express and compute, especially at the meta level. The syntactic uniformity is the mechanism; the ability to write programs that construct and transform other programs, using the same tools as everything else, is the payoff.
Homoiconicity doesn't make Lisp programs mean something different, but it gives you a more uniform and powerful meta-programming model.
reply