[chipdale]

I can understand why developers might not care for DSLs, especially when they force a choice between entirely different toolsets and toolchains.

However, I feel people often miss the real value of a good DSL: it's not about the syntax, but about providing hardened semantics that can bolster or guarantee desired qualities. Elm, for instance, provides value insofar as it makes producing runtime exceptions significantly more difficult.

Personally, I hope languages like Lean, which provides exceptional support for creating DSLs within the language, renew interest in semantically sound DSLs, especially if we insist on using LLMs.

[lioeters]

I recently learned about that too, how Lean provides ways to create new language syntax and DSLs, which is (on the surface) similar to Lisps like Racket. It's like operator overloading but way more flexible and general purpose. I'm wary of such language features, convenient for specific purposes (like working with vectors or matrices) but I'm afraid it's too powerful for normal usage, if everyone starts building their own DSLs and syntaxes, it would likely make the typical codebase difficult to read and understand. For example, the C++ template metaprogramming madness, it can be used responsibly but in my opinion it was detrimental to the language ecosystem.

[rupertlssmith]

And then there are sealed DSLs...

Byte encoder/decoder in Elm are opaque, so their implementation is hidden. A smart compiler, such as the one I am writing, can take advantage of that with compiler intrinsics that replace the implementation with something that the compiler itself optimizes.