spec: schema-camelcase-fix — paramTypes/retType → param-types/ret-type (closes #30)

Brainstorm output. Single-iteration milestone: rename the two
camelCase JSON tags on `Term::Lam` — the only camelCase outliers
in the AST schema — to kebab-case, matching the convention every
other compound-key tag (`reuse-as`, etc.) already follows.

Scope is verified-tiny: ast.rs (2 serde-rename strings) +
workspace.rs in-source test JSON-literal + 2 `.ail.json` fixtures.
None of the five hash-pinned `.ail` modules contains a lambda,
so the milestone refreshes zero hash pins. Form-A is untouched —
the surface uses `(params (typed ...))` / `(ret ...)`, never the
camelCase tags.

Feature-acceptance gate: passes weakly-but-honestly. Clause 1 is
indirect (schema-internal consistency reduces the LLM author's
"compound-tag-is-kebab" generalisation failure rate); there is no
empirical preference measurement for this axis. Clause 2 holds
(one fewer special case to memorise). Clause 3 vacuous (no
semantic surface touched).

Grounding-check (Step 7.5) PASS: all four load-bearing claims
ratified by currently-green tests — round_trip.rs (Form-A
invariance), hash_pin.rs (no lambda in pinned modules),
workspace.rs in-source ct1_validator test (exhaustive call-site
list), design_schema_drift.rs (kebab convention pin).

Not bundled with #27 (arith-rename) per the spec preamble: #27
carries its own four open brainstorm questions and is a separate
milestone; each gets one re-pin wave with its own rationale, no
churn savings from bundling.
This commit is contained in:
2026-05-21 12:37:45 +02:00
parent 26fb3459d8
commit 7d086e69ce
@@ -0,0 +1,325 @@
# schema-camelcase-fix — Design Spec
**Date:** 2026-05-21
**Status:** Draft — awaiting user spec review
**Authors:** Brummel (orchestrator) + Claude
**Closes:** Gitea #30
## Goal
Replace the two camelCase JSON tags `paramTypes` and `retType` on
`Term::Lam` with the kebab-case `param-types` and `ret-type`. These
are the only camelCase outliers in the entire AST schema; every
other compound-key tag (`reuse-as`, `drop-iterative`, etc.) is
kebab-case and every atomic tag (`type`, `fn`, `else`, `in`, `as`)
is a single lowercase word. The fix is a pure schema-tag rename in
the canonical JSON layer.
## Architecture
This is a schema-only milestone. Three layers are affected:
1. **Canonical JSON tag layer.** Two `#[serde(rename = "...")]`
attributes on `Term::Lam` in `crates/ailang-core/src/ast.rs:492,494`
change literal value: `"paramTypes" → "param-types"`,
`"retType" → "ret-type"`. The Rust field names
(`param_tys`, `ret_ty`) are unaffected.
2. **Surface (Form-A) layer.** **Unchanged.** Form-A prints lambdas
as `(lam (params (typed NAME TYPE)*) (ret TYPE) (effects ...)
(body TERM))`. There is no Form-A keyword `paramTypes` or
`retType` to migrate — the print/parse code at
`crates/ailang-surface/src/print.rs:502-528` and
`crates/ailang-surface/src/parse.rs:1430-1468` uses positional
sub-forms `(typed ...)` and `(ret ...)`, both of which already
match the kebab/single-word convention.
3. **JSON source files.** Two `.ail.json` fixtures and one inline
JSON literal in a Rust unit test embed the old tags directly
and must be updated:
- `examples/test_loop_binder_captured_by_lambda.ail.json`
- `experiments/2026-05-12-cross-model-authoring/master/examples/fn_with_lambda.ail.json`
- `crates/ailang-core/src/workspace.rs:1925-1926`
(Verified via `grep -rn '"paramTypes"\|"retType"' --include="*.rs"
--include="*.json"` — exhaustive list.)
**Hash blast radius is zero on pinned modules.** None of the
`.ail` modules pinned in `crates/ailang-core/tests/hash_pin.rs`
(`sum`, `list`, `ordering_match`, `test_22b1_dup_a`,
`test_22b1_dup_classmod`) contains a `(lam ...)` term. Verified
via `grep -c "(lam " examples/<each>.ail` — all zero. The two
`.ail.json` fixtures listed above are not hash-pinned. Therefore
the milestone does not refresh a single hash pin and no Honesty-Rule
hash-comment migration is required.
## Concrete code shapes
### The AILang program (Form-A — unchanged)
The Form-A author surface is unchanged by this milestone. Authors
write lambdas exactly as before. Worked example, lifted from
`examples/closure.ail`:
```
(fn main
(doc "captures `n` from the enclosing let, returns 42")
(type (fn-type (params) (ret (con Unit)) (effects IO)))
(params)
(body (let n 3
(app print
(app apply
(lam (params (typed x (con Int)))
(ret (con Int))
(body (app + x n)))
39)))))
```
The lambda sub-form `(lam (params (typed x (con Int))) (ret (con
Int)) (body (app + x n)))` reads consistently with the surrounding
`(fn-type (params ...) (ret ...))` — Form-A was never the locus of
the inconsistency. The Form-A round-trip
(`crates/ailang-surface/tests/round_trip.rs`) must stay GREEN with
zero diff to the printer/parser.
### The delivered code shape (canonical JSON — before → after)
The same lambda's canonical JSON before this milestone:
```json
{
"t": "lam",
"params": ["x"],
"paramTypes": [{ "k": "con", "name": "Int" }],
"retType": { "k": "con", "name": "Int" },
"effects": [],
"body": { "t": "app",
"callee": { "t": "var", "name": "+" },
"args": [{ "t": "var", "name": "x" },
{ "t": "var", "name": "n" }] }
}
```
After:
```json
{
"t": "lam",
"params": ["x"],
"param-types": [{ "k": "con", "name": "Int" }],
"ret-type": { "k": "con", "name": "Int" },
"effects": [],
"body": { "t": "app",
"callee": { "t": "var", "name": "+" },
"args": [{ "t": "var", "name": "x" },
{ "t": "var", "name": "n" }] }
}
```
That is the entirety of the observable change at the canonical-JSON
layer. After this milestone the JSON schema is regular: kebab-case
for every multi-word tag, single-word for every atomic tag, with
no exceptions.
### The Rust implementation shape (secondary)
In `crates/ailang-core/src/ast.rs`:
```rust
// before
Lam {
params: Vec<String>,
#[serde(rename = "paramTypes")]
param_tys: Vec<Type>,
#[serde(rename = "retType")]
ret_ty: Box<Type>,
#[serde(default)]
effects: Vec<String>,
body: Box<Term>,
}
// after
Lam {
params: Vec<String>,
#[serde(rename = "param-types")]
param_tys: Vec<Type>,
#[serde(rename = "ret-type")]
ret_ty: Box<Type>,
#[serde(default)]
effects: Vec<String>,
body: Box<Term>,
}
```
`crates/ailang-core/src/workspace.rs:1925-1926` — JSON-literal in a
test that synthesises a Lam to drive the bare-cross-module-type-ref
rejection path — updates the embedded keys verbatim:
```rust
// before
"body": {
"t": "lam",
"params": ["x"],
"paramTypes": [{ "k": "con", "name": "Ordering" }],
"retType": { "k": "con", "name": "Unit" },
...
// after
"body": {
"t": "lam",
"params": ["x"],
"param-types": [{ "k": "con", "name": "Ordering" }],
"ret-type": { "k": "con", "name": "Unit" },
...
```
The two `.ail.json` fixtures get the same key-rename, byte-for-byte;
the surrounding fixture content is untouched.
## Components
| Component | Change |
|-----------|--------|
| `crates/ailang-core/src/ast.rs:492,494` | Two `#[serde(rename)]` strings: `"paramTypes" → "param-types"`, `"retType" → "ret-type"`. |
| `crates/ailang-core/src/workspace.rs:1925-1926` | In-source JSON-literal in a test: same two key renames. |
| `examples/test_loop_binder_captured_by_lambda.ail.json` | Same two key renames (verified 3 occurrences combined of the two tags via `grep -c`). |
| `experiments/2026-05-12-cross-model-authoring/master/examples/fn_with_lambda.ail.json` | Same two key renames (verified 3 occurrences combined). |
| `crates/ailang-surface/src/{print,parse}.rs` | **No change.** Form-A never used these tag names. |
| `crates/ailang-core/tests/hash_pin.rs` | **No change.** No pinned `.ail` module contains `(lam ...)` (verified per-file). |
Historic documents (`docs/plans/2026-05-10-23.3-...`,
`docs/plans/2026-05-09-22b3-monomorphisation.md`,
`docs/plans/2026-05-09-22c-user-class-e2e.md`,
`docs/plans/24.2.md`) embed the old tag names inside example JSON
blocks. Per the Honesty Rule
(`design/contracts/honesty-rule.md`), historic plan documents
describe the state at the time of writing — not the present state —
and are NOT migrated.
## Data flow
The canonical-JSON ↔ AST bridge is `serde_json` with the
`#[serde(tag = "t")]` discriminator on `Term`. The `rename`
attributes are the only point where the JSON key differs from the
Rust field name. Changing the rename string changes (a) what
`serde_json::to_string` emits for the field, and (b) what
`serde_json::from_str` accepts for the field. Both directions move
together; there is no two-key transitional window — a single commit
ships the rename plus all three call-sites (workspace test +
two fixtures) so no `.ail.json` is observably inconsistent at any
intermediate point.
The Form-A path (`.ail` author surface → `parse_lam` → AST → canonical
serialisation → hash) does not touch the renamed tags as text: the
parser maps the Form-A keywords `params`/`typed`/`ret` directly to
the Rust fields by struct-position, and `serde_json` then re-emits
them with the new rename. The same module hashes the same way before
and after at the AST level; only the *canonical JSON byte stream*
differs, which is what the renaming targets.
## Error handling
Three failure modes possible during the migration:
1. **A `.ail.json` fixture is missed.** Surfaces as a deserialisation
error from `ailang-core` when the workspace loader hits an
unknown field. Caught by `cargo test --workspace` (every fixture
participates in at least one test). Resolution: the grep used to
build the Components table was exhaustive across `examples/`,
`experiments/`, and `crates/`; if a future fixture is added with
the old tag, the deserialiser rejects it.
2. **The `workspace.rs` JSON literal is missed.** Same as above,
caught by the same test crate — specifically the
`validate_canonical_type_names` test that owns the synthesised
module.
3. **A historic `docs/plans/*.md` is "helpfully" updated.** Not a
failure mode, but a rule-violation against the Honesty Rule.
Historic plans are frozen documents; updating them retroactively
would falsify the recorded design intent at the iter close. The
spec explicitly forbids this (see Components above) and the
ratifying test
(`crates/ailang-core/tests/docs_honesty_pin.rs`) does not
police plan files anyway — the rule is editorial discipline.
There is no compatibility shim and no transitional accept-both
behaviour. The JSON tag rename is atomic in one commit; old fixtures
not migrated by this commit cannot deserialise after this commit.
Any future encounter with a stray `"paramTypes"` is a stale fixture,
not a backward-compat concern.
## Testing strategy
The milestone is GREEN if:
- `cargo test --workspace` passes (this hits every `.ail.json`
fixture transitively and the workspace.rs test directly).
- `cargo test -p ailang-surface --test round_trip` passes — confirms
Form-A is untouched.
- A single targeted assertion is added: a fresh test in
`crates/ailang-core/tests/schema_coverage.rs` (or equivalent
existing schema-shape test) verifies that
`serde_json::to_value(&Term::Lam{...})` emits the keys
`"param-types"` and `"ret-type"` (and NOT the old camelCase). The
test pins the new schema layer concretely so a future regression
cannot silently flip the rename string back.
- `cargo build --workspace` produces no new clippy warnings.
No bench-regression check is required: the rename is a string
literal swap at compile time with no runtime cost path.
## Acceptance criteria
1. Tag-rename change in `crates/ailang-core/src/ast.rs:492,494` is
landed and observable in serialisation output.
2. Three call-sites (`workspace.rs` + two `.ail.json` fixtures)
carry the new tag names verbatim.
3. `cargo test --workspace` is GREEN.
4. `cargo test -p ailang-surface --test round_trip` is GREEN
(Form-A unchanged invariant).
5. A new schema-shape assertion test fires GREEN with the new tag
names and would fire RED with the old ones.
6. The commit body references `closes #30`. No `docs/plans/*.md`
files are mutated.
## Feature-acceptance gate
Per `design/contracts/feature-acceptance.md`, three clauses must
hold. This milestone's gate evaluation:
- **Clause 1 — an LLM author naturally produces code that uses the
feature.** Direct empirical evidence for *kebab-case preference
inside this schema* is absent — the 2026-05-21 naming-A/B run
measured a different naming axis (apply vs call vs classic head
tag, not field-key case). The argument here is indirect but
concrete: a future LLM author derives the lambda's JSON form by
*generalising from the rest of the schema* it has just read.
Every other compound tag in the AST is kebab-case (verified by
exhaustive `grep -n "serde(rename" crates/ailang-core/src/ast.rs`
— 17 sites, only the two on `Term::Lam` use camelCase). The
generalisation an LLM forms reads "compound keys are kebab" — and
it will get `paramTypes` wrong on first attempt with non-trivial
probability. The fix eliminates the one site where the
generalisation diverges from reality. Clause 1 holds **weakly but
honestly**: the gain is consistency-induced lower error
probability, not a measured preference for kebab.
- **Clause 2 — measurably improves correctness or removes
redundancy.** Redundancy reduction is direct: one less special
case in the "JSON tag spelling convention" the LLM author has to
memorise. The "measurable" side is the LLM-mistake probability
reduction sketched in Clause 1; even without a number it is a
monotonic improvement (the new rule is strictly simpler than the
old). Holds.
- **Clause 3 — reintroduces no bug class the core constraint exists
to eliminate.** Pure tag rename; no semantic, type-system,
uniqueness-mode, or codegen surface touched. The Form-A round-trip
invariant is structurally untouched (parser doesn't see these
tags). Holds vacuously.
The gate passes. The honesty here is to name Clause 1 as the
weakest of the three, supported by *schema-internal consistency*
rather than by an experiment. A future schema-touching milestone
(#27 arithmetic-rename) might collect more direct evidence; this
fix does not block on that and does not pretend to.