I do find the pattern interesting and powerful.

But at the same time, something feels off about it (just conceptually, not trying to knock your money-making endeavor, godspeed). Some of the issues that all of these hit is:

- No printf debugging. Sometimes you want things to be eager so you can immediately see what's happening. If you print and what you see is <RPCResultTracingObject> that's not very helpful. But that's what you'll get when you're in a "tracing" context, i.e. you're treating the code as data at that point, so you just see the code as data. One way of getting around this is to make the tracing completely lazy, so no tracing context at all, but instead you just chain as you go, and something like `print(thing)` or `thing.execute()` actually then ships everything off. This seems like how much of Cap'n Web works except for the part where they embed the DSL, and then you're in a fundamentally different context.

- No "natural" control flow in the DSL/tracing context. You have to use special if/while/for/etc so that the object/context "sees" them. Though that's only the case if the control flow is data-dependent; if it's based on config values that's fine, as long as the context builder is aware.

- No side effects in the DSL/tracing context because that's not a real "running" context, it's only run once to build the AST and then never run again.

Of the various flavors of this I've seen, it's the ML usage I think that's pushed it the furthest out of necessity (for example, jax.jit https://docs.jax.dev/en/latest/_autosummary/jax.jit.html, note the "static*" arguments).

Is this all just necessary complexity? Or is it because we're missing something, not quite seeing it right?

I think this kind of tracing-caused complexity only arises when the language doesn't let you easily represent and manipulate code as data, or when the language doesn't have static type information.

Python does let you mess around with the AST, however, there is no static typing, and let's just say that the ML ecosystem will <witty example of extreme act> before they adopt static typing. So it's not possible to build these graphs without doing this kind of hacky nonsense.

For another example, torch.compile() works at the python bytecode level. It basically monkey patches the PyEval_EvalFrame function evaluator of Cpython for all torch.compile decorated functions. Inside that, it will check for any operators e.g BINARY_MULTIPLY involving torch tensors, and it records that. Any if conditions in the path get translated to guards in the resulting graph. Later, when said guard fails, it recomputes the subgraph with the complementary condition (and any additional conditions) and stores this as an alternative JIT path, and muxes these in the future depending on the two guards in place now.

Jax works by making the function arguments proxies and recording the operations like you mentioned. However, you cannot use normal `if`, you use lax.cond(), lax.while(), etc,. As a result, it doesn't recompute graph when different branches are encountered, it only computes the graph once.

In a language such as C#, Rust, or a statically typed lisp, you wouldn't need to do any of this monkey business. There's probably already a way in the rust toolchain to interject at the MIR stage and have your own backend convert these to some Tensor IR.

Really nice summary of the core challenges with this DSL/code-as-data pattern.

I've spent a lot of time thinking about this in the database context:

> No printf debugging

Yeah, spot on. The solutions here would be something like a `toSQL` that let's you inspect the compiled output at any step in the AST construction.

Also, if the backend supports it, you could compile a `printf` function all the way to the backend (this isn't supported in SQL though)

> No "natural" control flow in the DSL/tracing context

Agreed -- that can be a source of confusion and subtle bugs.

You could have a build rule that actually compile `if`/`while`/`for` into your AST (instead of evaluate them in the frontend DSL). Or you could have custom lint rules to forbid them in the DSL.

At the same time -- part of what makes query builders so powerful is the ability to dynamically construct queries. Runtime conditionals is what makes that possible.

> No side effects in the DSL/tracing context because that's not a real "running" context

Agreed -- similar to the above: this is something that needs to be forbidden (e.g., by a lint rule) or clearly understood before using it.

> Is this all just necessary complexity? Or is it because we're missing something, not quite seeing it right?

My take is that, at least in the SQL case: 100% the complexity is justified.

Big reasons why: 1. A *huge* impediment to productive engineering is context switching. A DSL in the same language as your app (i.e., an ORM) makes the bridge to your application code also seamless. (This is similar to the argument of having your entire stack be a single language) 2. The additional layer of indirection (building an AST) allows you to dynamically construct expressions in a way that isn't possible in SQL. This is effectively adding a (very useful) macro system on top of SQL. 3. In the case of Typescript, because its type-system is so flexible, you can have stronger typing on your DSL than the backend target.

tl;dr is these DSLs can enable better ergonomics in practice and the indirection can unlock powerful new primitives