All 176 files in the four accumulating directories now use a zero-padded 4-digit counter prefix that reflects creation order (`NNNN-slug.md`). The counter is assigned per directory in strict git-log creation order; ties broken alphabetically by original name. The old `YYYY-MM-DD-` prefix on docs/specs/ and docs/plans/ files is dropped — the date is recoverable from git log and the counter carries the ordering. A file's counter is stable for the life of the file: never reassigned, never reused, never compacted. Deleted files retire their counter; subsequent files do not fill the gap. This is the property that lets cross-references stay literal — refs use the full filename including the counter (`design/contracts/0007-honesty-rule.md`) so they grep cleanly and resolve directly without a glob step. 313 cross-references updated across .md/.rs/.toml/.c/.json files (test pins, include_str! paths, design-INDEX entries, baseline notes, runtime C comments, inter-contract markdown links incl. bare basename and `../models/foo.md` forms). CLAUDE.md gets a new "File-naming convention" section spelling out the rule and rationale. skills/brainstorm/SKILL.md and skills/planner/SKILL.md updated so new spec/plan creation produces counter-prefixed names from the start. The full test suite (cargo test --workspace) passes.
13 KiB
22c — User-class e2e fixture
Parent spec:
docs/specs/0002-22-typeclasses.mdFor agentic workers: REQUIRED SUB-SKILL: use
skills/implementto run this plan. Steps use- [ ]checkboxes for tracking.
Goal: ship one .ail.json fixture exercising the typeclass
machinery end-to-end through user-defined code (no Prelude
involvement). Running ail build on the fixture and executing the
binary prints 42.
Architecture: the fixture defines a user class Foo a with one
method, a user ADT IntBox, and an instance Foo IntBox whose body
matches on the ctor and returns the wrapped int. main calls
foo (MkIntBox 42) and prints the result through io/print_int. The
22b.3 monomorphisation pass synthesises foo__<typehash> for
Foo IntBox, rewrites the call site, and the existing match-lowering
emits the body as ordinary code. No new compiler passes, no new
runtime primitives, no new surface syntax.
Tech Stack: examples/ (one new fixture), crates/ail/tests/
(one new e2e test file). No production-code crates touched.
Out of scope: Prelude class for primitives (deferred per spec
amendment 2026-05-09 §"22b.4b dropped"). Multiple instances per
class, multi-method classes (test_22b3_coherence_two_instances
and test_22b3_default_e2e already cover those).
Files this plan creates or modifies
- Create:
examples/test_22c_user_class_e2e.ail.json— the fixture. - Create:
crates/ail/tests/typeclass_22c.rs— the e2e test (cargo test -p ail --test typeclass_22c). Asserts the binary prints42\n. - Modify:
docs/JOURNAL.md— append iteration entry at iter close.
That is the entire iteration's surface. No code crate is modified, because 22b.3 already shipped every machine-code path the fixture exercises.
Cross-task invariants
- No production-code changes. If a task requires editing
anything under
crates/*/src/(other than tests), that is a sign of a hidden bug surfaced by the fixture — STOP and route throughdebugskill (RED-first), not through this plan. - No Prelude. The fixture defines its own class; do NOT import
any
examples/prelude*.ail.json(none exists yet anyway, but the discipline matters: 22c's property is "user-defined" end-to-end). - Bench gates green. Per spec acceptance criterion 2,
bench/check.py/bench/compile_check.py/bench/cross_lang.pymust all exit 0/0/0 at iter close.
Task 1: Author the fixture
Files:
- Create:
examples/test_22c_user_class_e2e.ail.json
Step 1.1 — Write the fixture
Hand-write the canonical JSON. Schema validation via ail check
catches any shape error in Step 1.2. Key shapes verified against
existing fixtures: examples/box.ail.json (data-type + match +
term-ctor), examples/test_22b3_default_e2e.ail.json (class +
instance with (t: "lam") body), examples/borrow_own_demo.ail.json
(param_modes: ["borrow"] snake_case).
{
"schema": "ailang/v0",
"name": "test_22c_user_class_e2e",
"imports": [],
"defs": [
{
"kind": "type",
"name": "IntBox",
"ctors": [
{ "name": "MkIntBox", "fields": [{ "k": "con", "name": "Int" }] }
]
},
{
"kind": "class",
"name": "Foo",
"param": "a",
"methods": [
{
"name": "foo",
"type": {
"k": "fn",
"params": [{ "k": "var", "name": "a" }],
"param_modes": ["borrow"],
"ret": { "k": "con", "name": "Int" },
"effects": []
}
}
]
},
{
"kind": "instance",
"class": "Foo",
"type": { "k": "con", "name": "IntBox" },
"methods": [
{
"name": "foo",
"body": {
"t": "lam",
"params": ["b"],
"paramTypes": [{ "k": "var", "name": "a" }],
"retType": { "k": "con", "name": "Int" },
"body": {
"t": "match",
"scrutinee": { "t": "var", "name": "b" },
"arms": [
{
"pat": {
"p": "ctor",
"ctor": "MkIntBox",
"fields": [{ "p": "var", "name": "v" }]
},
"body": { "t": "var", "name": "v" }
}
]
}
}
}
]
},
{
"kind": "fn",
"name": "main",
"type": {
"k": "fn",
"params": [],
"ret": { "k": "con", "name": "Unit" },
"effects": ["IO"]
},
"params": [],
"body": {
"t": "do",
"op": "io/print_int",
"args": [
{
"t": "app",
"fn": { "t": "var", "name": "foo" },
"args": [
{
"t": "ctor",
"type": "IntBox",
"ctor": "MkIntBox",
"args": [{ "t": "lit", "lit": { "kind": "int", "value": 42 } }]
}
]
}
]
}
}
]
}
Notes on canonical key shape:
data def: kind: "type"(NOT"data").Pattern: {"p": "ctor", ...}/{"p": "var", ...}— discriminator field is"p", not"pattern"or"kind". Pattern's child patterns are under"fields", not"args".term-ctor: {"t": "ctor", "type": "IntBox", "ctor": "MkIntBox", "args": [...]}— the type name and the ctor name are sibling fields; do NOT use"t": "term-ctor".- Instance method body's
paramTypesreferences the class param variablea(not the concrete typeIntBox). The mono pass substitutesa := IntBoxwhen synthesising the per-instance function. Mirror the existing 22b.3 fixtures (test_22b3_coherence_two_instancesshowsparamTypes: [{"k": "var", "name": "a"}]).
Step 1.2 — Verify the fixture loads
cargo run -q -p ail -- check examples/test_22c_user_class_e2e.ail.json
Expected: exit 0, no diagnostics.
If ail check reports an error like unknown method "foo" or
no-instance for Foo IntBox, that is a fixture bug — fix the JSON
shape until check passes.
If ail check reports an internal error or panics, STOP and route
through debug skill RED-first — that is a real compiler bug
surfaced by 22c, not a fixture issue.
Step 1.3 — Verify the fixture compiles and runs
cargo run -q -p ail -- build examples/test_22c_user_class_e2e.ail.json -o /tmp/test_22c
/tmp/test_22c
Expected: stdout 42\n, exit 0.
If ail build errors during monomorphisation / codegen / linking,
STOP and route through debug skill — the fixture proves a real
compiler-side gap (most likely candidates: mono pass on instance
where the type is a user ADT, or match-lowering inside a synthesised
mono fn).
Step 1.4 — Commit the fixture
git add examples/test_22c_user_class_e2e.ail.json
git commit -m "iter 22c.1: fixture — user class + user ADT + instance, prints 42"
Task 2: E2E test
Files:
- Create:
crates/ail/tests/typeclass_22c.rs
Step 2.1 — Write the e2e test
Use the same shape as crates/ail/tests/typeclass_22b3.rs's
synthetic_fixture_runs_and_prints_five test (an existing precedent
for "build + run a fixture and assert stdout"). Concretely:
//! Iter 22c: user-class end-to-end. The single 22c fixture defines a
//! user class `Foo`, a user ADT `IntBox`, an `instance Foo IntBox`,
//! and a `main` that calls `foo (MkIntBox 42)` through the
//! monomorphisation pass and prints the result via `io/print_int`.
//! This is the milestone-22 acceptance fixture for the user-class
//! path (no Prelude involvement).
use std::process::Command;
fn workspace_root() -> std::path::PathBuf {
let manifest = std::path::PathBuf::from(env!("CARGO_MANIFEST_DIR"));
manifest.parent().unwrap().parent().unwrap().to_path_buf()
}
#[test]
fn user_class_fixture_compiles_runs_and_prints_42() {
let root = workspace_root();
let fixture = root.join("examples/test_22c_user_class_e2e.ail.json");
let out = root.join("target/test-22c/test_22c");
std::fs::create_dir_all(out.parent().unwrap()).expect("mkdir");
// Build
let build = Command::new("cargo")
.args([
"run", "-q", "-p", "ail", "--",
"build", fixture.to_str().unwrap(),
"-o", out.to_str().unwrap(),
])
.current_dir(&root)
.output()
.expect("spawn cargo run ail build");
assert!(
build.status.success(),
"ail build failed:\nstdout:\n{}\nstderr:\n{}",
String::from_utf8_lossy(&build.stdout),
String::from_utf8_lossy(&build.stderr)
);
// Run
let run = Command::new(&out)
.output()
.expect("spawn produced binary");
assert!(
run.status.success(),
"binary returned non-zero:\nstdout:\n{}\nstderr:\n{}",
String::from_utf8_lossy(&run.stdout),
String::from_utf8_lossy(&run.stderr)
);
assert_eq!(
String::from_utf8_lossy(&run.stdout).trim_end(),
"42",
"expected stdout `42` (with trailing newline)"
);
}
If crates/ail/tests/typeclass_22b3.rs uses a different
"build + run + capture stdout" idiom, mirror it instead — the
project's existing convention takes priority over this template.
Step 2.2 — Run the test
cargo test -p ail --test typeclass_22c
Expected: 1 passed; 0 failed.
Step 2.3 — Run the full bench gate
python3 bench/check.py
python3 bench/compile_check.py
python3 bench/cross_lang.py
Expected: all three exit 0/0/0.
Step 2.4 — Commit
git add crates/ail/tests/typeclass_22c.rs
git commit -m "iter 22c.2: e2e test — user-class fixture compiles, runs, prints 42"
Task 3: JOURNAL entry + milestone-close handoff
Files:
- Modify:
docs/JOURNAL.md— append iteration entry.
Step 3.1 — Append the iteration entry
## 2026-05-09 — Iteration 22c: User-class e2e fixture
The milestone-22 acceptance fixture for the user-class path. One new
`examples/test_22c_user_class_e2e.ail.json` defines a user class
`Foo a where foo : (a borrow) -> Int`, a user ADT `IntBox = MkIntBox
Int`, an `instance Foo IntBox` whose body matches on the ctor and
returns the wrapped int, and a `main` that calls `foo (MkIntBox 42)`
through the 22b.3 monomorphisation pass and prints the result via
`io/print_int`. Stdout: `42`.
**Per-task subjects (mirror commit messages):**
- 22c.1: fixture — user class + user ADT + instance, prints 42
- 22c.2: e2e test — user-class fixture compiles, runs, prints 42
**No production-code changes.** 22b.3 shipped every compiler path
this fixture exercises (mono of class methods, call-site rewrite,
synthesised symbol naming via `__` separator, match lowering inside
synthesised bodies). The fact that 22c lands as a pure-fixture iter
is the strongest possible evidence that the typeclass machinery is
done — there is nothing left to wire.
**Spec amendments folded into 22b.4a's amendment block:**
22b.4b (Prelude for primitives) dropped from milestone scope; 22c's
fixture re-scoped to vocabulary AILang already has (no `++`, no
named-field record syntax, no top-level polymorphic `print`). See
`docs/specs/0002-22-typeclasses.md` "Amendments" §"22b.4b
dropped" and §"22c scope tightened".
**Milestone-22 acceptance check:**
1. ✓ 22b.1 + 22b.2 + 22b.3 + 22b.4a + 22c JOURNAL entries committed.
2. (pending) Audit suite green — `audit` skill runs next.
3. ✓ 22c user-class fixture compiles, runs, prints `42` (this iter).
4. ✓ Round-trip filter retired in 22b.4a; all `examples/*.ail.json`
pass round-trip.
**Bench gates:** all three exited 0/0/0.
**Next step:** dispatch `audit` skill to close the milestone —
architect drift review against `docs/DESIGN.md` plus the three
regression scripts (already 0/0/0 here but `audit` re-runs as
its own gate). On clean audit: milestone 22 closes; queue moves
to whatever's next. On drift: tidy iteration first.
Step 3.2 — Commit
git add docs/JOURNAL.md
git commit -m "iter 22c: journal entry — milestone-22 acceptance fixture lands"
Acceptance gate (run after Task 3)
cargo test --workspace
python3 bench/check.py
python3 bench/compile_check.py
python3 bench/cross_lang.py
All four green is the iteration close. The orchestrator then
dispatches skills/audit to close milestone 22.
Self-review checklist (orchestrator, pre-commit)
- Spec coverage: every spec acceptance criterion has a task that satisfies it. Criterion 1 (JOURNAL) → Task 3; Criterion 3 (22c fixture compiles + runs) → Tasks 1+2; Criterion 4 (round-trip) → already satisfied by 22b.4a. ✓
- Placeholder scan: grep this file for "TBD" / "TODO" / "fill in" / "implement later" / "similar to" / "add appropriate". Zero hits. ✓
- Type consistency: the fixture's class name, method name, ADT
name, ctor name, instance binding, and main call site all use
Foo/foo/IntBox/MkIntBoxconsistently. ✓ - Step granularity: every step ≤ 5 minutes. The two ambiguous
steps are 1.2 and 1.3 (verify check + verify build) — they may
surface compiler bugs that aren't 5-minute fixes. The plan
instructs STOP and route to
debugin that case, which is the right discipline. ✓