Naming the Gate
There’s a moment in design work where you realise the name you’ve been using for something is technically correct but conceptually wrong — it describes when something happens rather than what it does. That moment came today with JoinNSecond.
The class family has always had two halves: the one you work with most of the time, holding filter(), var(), and index() — things you call repeatedly, before you’re done accumulating facts — and the one holding join(), fn(), and the scope methods, the operations that advance the chain or terminate it. I’d named them First and Second. First for the initial work; Second for what comes after.
That’s defensible, but it implies a sequence. Claude pushed back: “Second implies temporal order, not capability.” I explained what I’d meant — First is what the DSL author does first, repeatedly; Second is where they go after. As I was explaining it, the better name surfaced: Gate. Not second chronologically — the gateway into the next stage of the chain.
JoinNGate says something. JoinNSecond just counts.
The rename was one line of sed in each repo.
While reviewing the design, three other things surfaced.
The first was the END phantom type. Every generated class carries it to enable typed scope chain-back — not().join().filter().end(), where end() returns the outer builder with the correct arity restored. I’d built this deliberately.
Claude’s question was simple: is END actually in vol2?
I checked. Join2<CTX, B, C>. Three type parameters. No END. The phantom type was a sandbox invention to support a scope pattern that vol2 doesn’t have — and since we’ve already agreed that not() and exists() should take a lambda rather than chain through end(), there’s no reason to keep it. That’s queued for the next session: signatures go from Join2First<END, DS, A, B> to Join2First<DS, A, B>. Cleaner, and more faithful to the real codebase.
The second thing was vol2 itself. I went to apply the Gate rename and found it further along than I’d assumed. Consumer2 already has no ctx parameter — accept(A a, B b), not accept(DS ctx, A a, B b). The RuleBuilder is already a Permuplate template, generating up to arity 10, living in a src/main/permuplate/ directory alongside RuleExtendsPoint.java, RuleOOPathBuilder.java, and BaseTuple.java.
Vol2 has been building the real version while the sandbox was proving the concept. The Gate rename there was one file, no test updates.
Three things from one design review: a name that finally says what it means, a phantom type to remove, and a real codebase that’s further ahead than I thought. Not bad.