Hypothesis-driven measurement of "did monomorphisation actually
buy us performance?" on a 100M-iter LCG hot loop, AILang mono'd
code vs. four C reference variants (direct-inlinable, direct-
noinline, indirect-monomorphic, indirect-polymorphic). Zen 3,
clang -O2, median-of-15.
Headline: H1 supported, but the mechanism is inlining, not
dispatch shape. AILang mono = hand-C direct (1.000x). Indirect-
monomorphic = direct-noinline (1.000x) — saturating branch
predictor makes the indirect-call cost vanish on this hardware.
Inlining is the actual 3.31x win; polymorphic indirect adds
another 21% predictor-miss penalty.
DESIGN.md Decision 11 gains a rationale paragraph reframing mono
as inlining-enabler rather than indirect-call-eliminator, with
explicit pointer to the bench. JOURNAL entry records the full
methodology, ratios, limitations, and the side-effect mono-pass
env.globals-seeding bug surfaced while building the AILang fixture
(separate RED-first debug iter to follow).
Adds DESIGN.md "Feature-acceptance criterion" as a top-level section:
a feature ships only if (1) an LLM author naturally produces code
that uses it, and (2) it measurably improves correctness or removes
redundancy. Aesthetic appeal and human ergonomics do not count.
Mirrored in CLAUDE.md as "Feature acceptance: LLM utility", paired
with the existing "Design rationale != implementation effort". The
two together narrow valid feature rationales to one thing: what the
LLM author gets out of the feature.
Trigger: the typeclass-design conversation around 22a. Rule was
implicit in many past decisions (Decision 1's JSON-over-text choice,
Decision 10's "what LLMs are good at" reasoning) but never stated
as a feature-gate. Codifying it now means future feature proposals
get evaluated against an articulated criterion instead of being
re-derived each time.
Documentation-only; no Rust, schema, or bench changes. Test state
288/0/3 unchanged.
Three drift items from ailang-architect, plus one false-positive
surfaced during verification:
1. DESIGN.md silent on closure-pair 4.14x finding (21'b).
Decision-10 ratified: "Workload scope of the 1.3x target"
paragraph scopes the retirement gate to linear/tree/poly-ADT
workloads; closure-pair carve-out documented as
representational cost (closure cell + env struct = 2 allocs
per step) until a slab/pool answer ships.
2. bench/compile_check.py corpus drift. Three fixtures added
(bench_compute_intsum, bench_compute_collatz,
bench_list_sum_explicit), re-baselined. Now 12 fixtures x
2 ops = 24 compile-time metrics.
3. baseline.json convention not codified. Note field gains
"max-of-distribution gets wider band than percentile"
convention discovered in 21'd.
4. (verification finding) bench_compute_intsum cross_lang
tolerances at 15%/12% fire on subprocess-spawn jitter for
sub-millisecond fixtures. Widened to 35% across all five
intsum metrics; convention recorded in baseline_cross_lang
note field (sub-ms fixtures need looser bands).
All three bench gates re-run sequentially after edits:
bench/check.py — 63 metrics; 63 stable
bench/compile_check.py — 24 metrics; 24 stable
bench/cross_lang.py — 25 metrics; 25 stable
Total: 112 metrics under regression coverage, all green.
288 tests passing, 3 ignored. No Rust changes.
The dual-allocator policy from 2026-05-08 made Boehm the CLI default
"for general workloads", with rc selected explicitly for real-time-
sensitive code. That asymmetry contradicted Decision 10: rc + uniqueness
inference is the canonical memory model AILang's typechecker enforces,
and the CLI was teaching the opposite mental model to users and prompt-
fed LLMs.
This iter flips the asymmetry without removing Boehm:
- main.rs: default_value gc -> rc for build/run; help text rewritten
so rc is the canonical path and gc is the parity oracle.
- DESIGN.md Decision 9 retitled "RC canonical, Boehm parity oracle"
and rewritten end-to-end. Migration plan step 6 updated to a
two-step retirement (default-flip done, full removal gated).
- DESIGN.md pipeline diagram: rc-first, gc-as-oracle.
- JOURNAL: 2026-05-09 entry recording the call, what shipped,
the bug the flip surfaced, the gating condition for full
Boehm removal.
Bug surfaced by the flip and fixed in the same iter:
codegen/drop.rs build_pair_drop_fn emitted `getelementptr inbounds
{{ ptr, ptr }}, ...` via push_str (not format!), so the doubled
braces leaked verbatim into the IR. LLVM parsed it as a struct-of-
struct, the GEP referenced a non-existent field, and clang failed.
Invisible under the old gc default (no per-type drop fns) and
invisible to the 8 explicit _with_alloc("rc") tests (none of them
escaped a closure-with-captures). Caught immediately by two corpus
tests once rc became the baseline: closure_captures_let_n,
local_rec_as_value_capture_demo. Fix: single braces. Tests stay
as the regression guard.
The episode is the canonical justification for keeping the GC arm
as a parity oracle for now — a months-old codegen bug found by
flipping the baseline column.
288 passed / 0 failed / 3 ignored, unchanged.
Architect drift review after Family 20 surfaced four items; (1) and
(2) are doc drift fixed here, (3) and (4) are recorded as acceptable.
DESIGN.md changes:
- Data model section rewritten from "Data model (MVP)" claiming
only fn/const exist to "Data model" reflecting all three Def
kinds. Added the missing schema fields the architect named:
`tail` flag on app/do, the `seq` / `clone` / `reuse-as` / `letrec`
terms, `paramModes` / `retMode` on Type::Fn, `suppress` on FnDef,
`drop-iterative` on TypeDef, `args` on Type::Con, `vars` on
TypeDef. Each additive field's hash-stable serialisation is
called out once at the top so individual mentions stay terse.
`Suppress`, `ParamMode`, `Literal`, `Pattern` get their own
sub-blocks.
- Pipeline diagram now shows `--alloc=gc` (links libgc;
transitional) and `--alloc=rc` (emits ailang_rc_inc / _dec;
canonical) as two backends sharing the same MIR.
- Ecosystem inventory gained a "Surface forms" bullet for
ailang-surface (Iter 14c, lossless Form-A printer/parser) and
ailang-prose (Family 20, lossy Form-B projection). CLI bullet
now lists parse, render, prose, merge-prose.
JOURNAL entry records the architect findings, the resolutions for
items (1) and (2), and the acceptable-drift status of (3)
FnDef::synthetic factor-out (defer to next schema-additive
FnDef field) and (4) linearity.rs spurious warning (not
corpus-triggered).
Pure documentation iter — no code changes. 288 workspace tests
unchanged and green.
Closes a design hole shipped in 20d: the merge-prose prompt instructed
the LLM to emit JSON-AST and gave a 12-line schema-essentials reminder.
JSON-AST is the canonical hashable artefact, not a writing surface; the
reminder was not a language spec. Foreign LLMs had no realistic shot.
20f makes two coupled changes:
1) The LLM now emits Form-A (the canonical authoring surface fixed by
Decision 6). merge-prose loads the original via ailang_core::load_module
and re-renders via ailang_surface::print before embedding (round-trip
is a gating contract on the surface crate, so this is lossless). The
user runs `ail parse foo.new.ailx` to recover JSON, then `ail check`.
2) crates/ailang-core/specs/form_a.md is the complete LLM-targeted
Form-A specification — grammar, every term/pattern/type/def keyword,
schema invariants, pitfall catalogue, four few-shot modules from
examples/*.ailx. Exported as ailang_core::FORM_A_SPEC via include_str!
and embedded verbatim in every merge-prose prompt.
Drift detection in crates/ailang-core/tests/spec_drift.rs: every variant
of Term, Pattern, Type, Def, Literal is reached via exhaustive `match`.
The arms are not the assertion — adding a new variant without updating
the match is a compile error before the test runs. Once matched, an
anchor string is asserted to appear in FORM_A_SPEC. 8 tests, all green.
The hand-written + mechanical-drift-test combo addresses the user's
"distance to code is too big" concern about a docs-only spec. Generator
overkill rejected: structural drift is mechanically caught, but prose
explanation, schema-invariant catalogue, pitfall list, and few-shot
corpus cannot be emitted from AST shape alone.
Tests:
- ailang-core: +8 spec_drift tests; existing 12 unchanged
- ail unit (3): rewritten in lockstep — assert (own T)/(effects IO)
landmarks, FORM-A SPECIFICATION header, FORM_A_SPEC body verbatim
- ail e2e merge_prose_prints_framed_prompt: rewritten — assert
`(module foo` + `FORM-A SPECIFICATION` instead of `ailang/v0`
- Workspace: all green
Richer integration paths (LLM tool-use, MCP server, LSP) were named
in the design discussion and deferred per "kiss". All three layer
additively on the static-prompt path; static prompt remains the
lowest-common-denominator fallback.
Per CLAUDE.md tidy-iter at family boundaries. ailang-architect
flagged the canonical 'DESIGN.md silent' finding for the entire
19a-arc — same shape as family-20's 20e tidy.
Decision 10 schema-additions block gained 'Iter 19b —
FnDef.suppress' entry. New subsection 'Advisory diagnostics —
Iter 19a-arc' documents the over-strict-mode lint rule (incl.
heap-type filter rationale), Severity::Warning introduction, CLI
exit-on-Error gating, and the suppress mechanism. New subsection
'Why advisory + suppress instead of inference' pins the three
substantive reasons (annotation-states-intent, drift-bremse,
LLM-authoring forcing function) so future iters can't reopen the
decision without engaging the recorded rationale.
Migration plan extended with step 7 covering 19a/19a.1/19b.
Decision 10's mandatory-annotation rule is explicitly reaffirmed
as unchanged.
No code changes. Two architect findings deferred (codegen FnDef
fan-out, snapshot-fixture coupling). Data model (MVP) drift
predates 19b — separate iter.
Family-20 tidy-iter: ailang-architect drift review surfaced
DESIGN.md silence on the new prose surface as the highest-leverage
gap. Decision 6 now has a 'Form (B) — human prose projection'
subsection naming the lossy-vs-lossless contract, the no-parser-
by-design choice, and the LLM-mediated round-trip. CLI block
gained 'ail prose' and 'ail merge-prose'.
No code changes. Form (B) does not weaken any Decision 6
invariant: JSON-AST remains the only hashable artefact, form (A)
the canonical authoring surface, 30-production grammar unchanged.
Two architect findings deferred (prompt-template lockstep test;
snapshot-fixture cross-family coupling). Family 20 closes.
JOURNAL queue: empty.
The 18-arc validated the RC pipeline end-to-end; the orchestrator's
call (JOURNAL 2026-05-08) was to keep Boehm as default and RC as
the ready alternative rather than retire Boehm. Decision 9's
'transitional Boehm' framing was wrong shape for that policy —
this edit re-casts it as 'dual allocator: Boehm GC for general
workloads, RC for real-time-sensitive workloads', preserving the
rest of the section.
Decision 10 holds as the specification of the RC alternative; this
edit only changes the framing of how the two allocators relate.
Resolves the architect's drift report on the 18g sub-arc:
- DESIGN.md: ratify mode-metadata's codegen role. param_modes /
ret_mode were already on Type::Fn since 18a; with 18d.4 / 18g
they became load-bearing for drop-emission decisions in
codegen. The new 'Mode metadata is load-bearing for codegen'
subsection records the four seams (Iter A + Iter B + 18g.1 +
18g.2) and names the let-alias-of-borrow carve-out the gates
do not propagate through.
- bench/run.sh: post-throughput, invoke bench/latency_harness.py
on the three canonical arms (Implicit @ gc, explicit @ rc,
Implicit @ rc control). The Boehm-retirement bench numbers
are now reproducible by anyone running the harness, not just
by hand on a specific host.
- Negative coverage: examples/rc_let_implicit_returning_app
+ alloc_rc_let_binder_for_implicit_returning_app_does_not_drop
pin the asymmetry to the (own)-ret-mode test (live=0 there,
live=1 here). The Borrow-ret-mode case is covered by the
language design itself — typechecker rejects 'borrow-
passthrough' shapes with consume-while-borrowed.
JOURNAL entry records four items as deferred known debt:
emit_inlined_partial_drop dynamic-tag fallback, carve-out
diagnostics surface, bench-number stat-of-N, and the
cross-family ordering observation about CLAUDE.md's tidy-iter
rule.
The 18-arc is formally closed with this tidy. Next iter is
the orchestrator's Boehm-retirement decision (Path A vs Path B
from JOURNAL 2026-05-08 18f entry, joined by the post-18g.2
re-bench numbers).
The schema sketch in Decision 10 already committed to
Term::ReuseAs { source, body } as a wrapper, but did not record
WHY the wrapper form was chosen over a reuse_from: Option<String>
modifier on Term::Ctor. Adds the two substantive reasons:
- Compositional flexibility: wrapper generalises to record
literals, opaque box wrappers, capability cells — anything
future iters might add as an allocating construct. Modifier
would replicate per Term variant.
- Source-locality: (reuse-as SRC NEW-CTOR) reads as a sentence
with the source-binder at the head. Modifier scatters intent.
The trade-off accepted: schema permits Term::ReuseAs around a
non-allocating body. Caught at typecheck via
reuse-as-non-allocating-body diagnostic. Same shape as
Decision 7's Term::If composability accepts.
Recording before 18d.1 ships — per CLAUDE.md "design rationale
ne implementation effort", a schema choice has to name its
substantive reasons before code is written, not retroactively.
User correction during the post-RC-overnight check-in: the 18a
"Type::Fn metadata vs. new Type variant" call was justified
across DESIGN.md, JOURNAL, and the commit message primarily by
"avoids ~250 match-arm sites". That is an observation about the
current state of the code, not a design rationale.
CLAUDE.md gains two binding rules:
- Design rationale ≠ implementation effort. Effort is at most a
tiebreaker; a choice whose only stated reason is effort is
suspect. The rule names the 18a misstep as the canonical
anti-example so future sessions catch it earlier.
- Direction freedom + bounce-back conditions inlined (was
previously a cross-reference to a private auto-memory file
outside the repo, which the user couldn't see).
DESIGN.md Decision 10's Schema-additions block now leads with
the substantive reasons for per-position metadata: semantic
locality (modes belong to fn-parameter positions, not to types
in general — Decision 1 line); compositional clarity (type
identity vs. calling convention factor apart); future-proofing
(per-position metadata generalises; Type-variant approach
combinatoric blows up). The match-arm count remains parenthetical,
explicitly named a tiebreaker.
JOURNAL records the correction itself — the mistake stays as a
data point because how design discipline corrupts is informative.
Adds form-A surface (borrow T) / (own T) wrappers in fn-type
params and ret slots. Internally these are not new Type variants
but per-position metadata on Type::Fn:
Type::Fn { params, param_modes, ret, ret_mode, effects }
enum ParamMode { Implicit, Own, Borrow }
This keeps Type itself unchanged, so unification, occurs, apply,
and ~190 other Type match-arms in the typechecker need no new
branch. Fields use serde skip-if-default predicates so canonical
JSON hashes for every pre-18a fixture stay bit-identical (sum,
list, hof, closure, list_map, etc.: zero diff under git).
Iter 18a is purely additive: typechecker treats modes as
transparent (mode-compat check deferred to 18c), no codegen
change (--alloc=gc / Boehm path unchanged), no linearity
enforcement. The annotation info flows into the JSON side-table
that 18c will consume.
Equality of mode slices is length-tolerant: a vec![] (the form
written by mechanically-updated construction sites that elide
modes) compares equal to vec![Implicit; n]. This kept the patch
from being a 100-site mechanical migration through the
typechecker. 18c will choose: keep the slack, or normalise to
full-length on construction.
New fixture examples/borrow_own_demo.{ailx,ail.json} exercises
both modes. list_length declares (borrow (con List)); sum_list
declares (own (con List)). Stdout: 3 then 6.
Documentation: DESIGN.md schema-additions block clarified that
modes are Type::Fn metadata (not Type variants); migration plan
flag rename --memory=rc → --alloc=rc to match the existing CLI
flag.
Verification:
- cargo build --workspace green
- cargo test --workspace green (154 tests, +1 vs baseline 153)
- git diff examples/{sum,list,hof,closure,list_map,...}.ail.json: empty
- borrow_own_demo runtime stdout: "3\n6\n"
- canonical JSON contains "param_modes":["borrow"] and ["own"]
Avoids the ~190 Type match-arms in the typechecker that a new
Type variant would require. ParamMode enum (Implicit | Own |
Borrow) attaches per-param to Type::Fn, with serde defaults so
existing fixture hashes stay bit-identical.
Replaces the earlier "RC + inference, no annotations" position
with a two-layer architecture: inference (intra-fn) + mandatory
LLM-author mode annotations on fn signatures (inter-fn). Adds
structured rationale for rejecting region inference (regions
assume stack-shaped lifetimes; real programs need
non-stack-shaped lifetimes for caches / memo tables) and linear
types as primary mechanism.
Schema additions enumerated for the 18-series:
Type::Borrow / Type::Own (Iter 18a)
Term::Clone (Iter 18c)
Term::ReuseAs (Iter 18d)
TypeDef.drop_iterative (Iter 18e)
Migration plan extended from 18a-d (4 iters) to 18a-f (6 iters).
JOURNAL records the regions excursion, the scope-shape reduction,
and the LLM-author lever as the structural advantage Decision 10
cashes in on.
Adds a conservative escape analysis that flags Term::Ctor and
Term::Lam allocations as non-escaping when they are bound by a
let, never flow to a fn arg / ctor field / tail-return / lambda
capture, and remain inside the allocating fn's frame. Codegen
lowers flagged sites to LLVM `alloca` instead of `@GC_malloc`;
escaping sites continue to use Boehm.
- escape.rs (new): name-based taint propagation; let-only
candidates; tail-position / app-arg / ctor-field / lam-capture
all taint.
- codegen: three sites switched (lower_ctor, lam env, closure
pair). Save/restore non_escape across lambda thunk emit.
- examples/escape_local_demo.{ailx,ail.json}: focused fixture
exercising both branches (peek and recursive count).
- e2e + escape unit tests: 133 → 141 (+8).
Stop-gate: 17a closes the queue. Memory-management discussion
(GC scope) is the next user-driven step. Per-fn arena observations
appended to JOURNAL — including the headline finding that 0/270
shipped allocations are flagged non-escaping under this rule
(real AILang threads ctors directly between fns; build-locally /
consume-locally is not idiomatic).
All existing fixture stdouts and content hashes bit-identical;
IR snapshots unchanged.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Lifts the Int-only restriction on `==`. Declared type becomes
forall a. Fn(a, a) -> Bool; codegen monomorphises and dispatches:
Int → icmp eq i64
Bool → icmp eq i1
Str → call @strcmp + icmp eq i32 0
Unit → constant true (operands still emitted for side effects)
ADT / Fn / other → CodegenError::Internal
This unblocks 16c's build_eq for non-Int lit patterns. == joins
__unreachable__ as the second polymorphic builtin (same Forall
machinery).
- check/builtins.rs: == registered as Forall(a, Fn(a, a) -> Bool).
- codegen: lower_eq dispatch table; @strcmp declared in IR header
alongside @printf/@GC_malloc/@puts.
- examples/eq_demo.{ailx,ail.json}: covers all four supported
scalars including a Str-lit-pattern match.
- IR snapshots refreshed: only +declare i32 @strcmp(ptr, ptr) in
the header; every define body bit-identical.
- e2e + check + codegen tests: 124 → 133 (+9, of which +1 is the
e2e fixture and the rest exercise the new dispatch / typecheck
surface).
Other comparison ops (<, <=, >, >=, !=) remain Int-only — out of
scope for this iter.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Eliminates the Unit-typed chain default that 16c left in place.
Adds __unreachable__ : forall a. a — codegen lowers to LLVM
unreachable. The 16a/16c chain machinery now uses it as the
deepest fall-through, so exhaustive matches with non-Unit arms
no longer need a (case _ ...) workaround.
- check/builtins.rs: register __unreachable__ as Forall(a, a).
- codegen: Term::Var "__unreachable__" emits LLVM unreachable
and sets block_terminated; If/Match/fn-body gate downstream
work on that flag.
- desugar: chain default switched from Lit{Unit} to Var.
- examples/lit_pat: categorize_first's trailing _ arm removed.
Hash changes from c4faec3abc2ed388 to 644de0c0ec15fc17.
- examples/unreachable_demo: positive fixture using safe_div
with __unreachable__ in the impossible branch.
- e2e + desugar tests: 122 → 124 (+2).
Path-a from 16b.2 planning: real primitive over a desugar-time
exhaustiveness pre-check (which would need cross-module type
registry access from desugar).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds path-2 from the 16b.2 planning entry. Desugar now defers
LetRecs whose captures include Term::Let-bound names; a new
ailang-check::lift_letrecs pass runs after typecheck and uses the
elaborated env to resolve capture types. ailang-check learns a real
Term::LetRec typing rule in synth and verify_tail_positions.
- desugar: Term::LetRec arm gains a defer-arm; helpers promoted to
pub for reuse by the lift pass; find_non_callee_use moved before
classification.
- ailang-check::synth/verify_tail_positions: real LetRec rules
(effect-subset, locals install, recursive name in body+in_term).
- ailang-check::lift.rs (new, 720 LOC): post-typecheck lift with
post-order traversal, env-walk for capture-type resolution,
fast-path skip when no LetRec is present.
- ail::main.rs: build path now does load → check → desugar →
lift_letrecs → codegen.
- examples/local_rec_let_capture.{ailx,ail.json}: new fixture
capturing a let-bound `threshold` in a recursive helper.
- e2e + check unit tests: 106 → 110 (+4).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Desugar `Pattern::Lit` (top-level and nested in Ctor) to
`Term::If { cond = (== sv lit), then = body, else_ = fall_k }`.
After 16c, no Pattern::Lit survives the desugar pass; codegen
and typechecker never see one. `is_flat` reclassifies Lit as
non-flat to force the chain machinery; new `build_eq` helper
constructs the equality test per literal kind.
Tests: 99 → 103 (+1 e2e lit_pat_demo, +3 desugar unit). Demo
fixture exercises top-level lit arms (classify) and nested
lit-in-Ctor (categorize_first) with a local IntList; output
100/200/999/-1/0/7.
Known limitations documented in DESIGN.md and JOURNAL: (a)
chain machinery's Unit terminator means non-Wild-terminated
exhaustive matches with lit arms still need a trailing `_`
catch-all to type-check; (b) `==` is currently Int-only, so
Bool/Str/Unit lit patterns desugar correctly but error at
typecheck.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
After six feature iters since the last docs sweep, DESIGN.md had
visible drift. Patched in place rather than queueing the next
codegen-heavy iter against a stale spec.
DESIGN.md changes (+72 LOC):
- Decision 6: form (A) marked shipped (Iter 14c, sole projection
since 15e); body kept as audit trail.
- Pipeline: desugar pass added between resolve+hash and typecheck;
invariant noted that CheckedModule.symbols hashes from the
original module, not the desugared one (so ail diff/manifest
preserve on-disk identity).
- CLI: added deps, diff, workspace, builtins (shipped earlier but
never doc'd).
- "What is not (yet) supported": re-anchored from "end of Iter 13"
to "as of Iter 16a"; removed lifted gates (cross-module ADTs,
no-GC, flat-pattern-only); added tighter follow-up gates
(literal sub-patterns, local recursive let).
- "What IS supported": promoted nested Ctor patterns (16a),
cross-module ADTs (14h), form-(A) text surface (14b/14c/15e),
Boehm GC (Decision 9 / 14f) into the smoke-test list.
- Smoke tests: added std_list_demo, std_maybe_demo, std_either_demo,
std_pair_demo, nested_pat fixtures.
JOURNAL: 16a-aux entry recording the drift sites and what was
explicitly *not* changed (Goal, Decisions 1-5, 7-9, Mangling,
Convention, Data model, Verification — spot-checked, all current).
Tests: 93/93 unchanged (doc-only). Build clean.
Reconsidered 14d's removal of Term::If. The decision was wrong.
"No redundancies" requires judgment; reducibility (if -> match)
is not redundancy in the strong sense. Term::If is a primitive
control-flow shape; bool branching is the second most common
shape after sequencing, and removing it cost 3x tokens on every
branch site (`(if c a b)` 4 tokens vs the match-on-Bool form
12 tokens).
Meta-pattern fixed: I had been treating user observations as
directives. User said "if is a subset of match"; I jumped to
remove it citing CLAUDE.md, with no independent conviction.
The leak appeared in 14f's JOURNAL prose ("three lines for what
if used to do in one"), which read as regret. Two feedback
memories saved (memory/feedback_user_suggestions_not_directives,
memory/feedback_no_nostalgia_for_removed_features) to head this
off.
Implementation: mechanical reverse-application of 14d's diff at
every site (AST, check including the 14e tail-position arm,
codegen 4 sites, surface parser/printer, pretty, CLI walker,
e2e test mutation). Removed lower_bool_match helper — it existed
only because 14d's migration shape needed codegen for non-ptr
match scrutinees; with Term::If back, match-on-Bool returns to
its pre-14d unsupported state. Three fixtures (sum, sort, max3)
restored to pre-14d shape. gc_stress (added in 14f) also
migrated back to (if ...) since it was authored under the wrong
constraint.
14e (musttail) and 14f (GC_malloc) verified intact in IR.
Hashes restored to pre-14d values:
- sum.sum: db33f57cb329935e
- sort.insert: 697fcb9f30f8633a
- max3.max: 65c45d6a45dd0a72
- max3.max3: 624b14429bf302f5
All other defs across all 18 fixtures keep their post-14f
hashes. Tests 80/80 green; cargo doc 0 warnings. LOC delta
+265/-295 net -30.
DESIGN.md Decision 7 preserved with a "Status: REVERTED" header
for audit trail. Form-(A) `if-term` production restored.
Plan: back to 15a (std_maybe stdlib module).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Decision 9 ships. Through Iter 14e every ADT box, lambda env, and
closure pair was leaked. This iter substitutes GC_malloc for malloc
in all four IR allocation sites and links -lgc. No language change,
no AST change, no schema change.
Diff: 5 files modified, ~30 LOC net.
- codegen/lib.rs: 4 substitutions @malloc -> @GC_malloc.
- ail/main.rs: .arg("-lgc") in the clang invocation.
- 5 IR snapshot files: mechanical s/@malloc/@GC_malloc/, 9
occurrences. IR is bit-identical to pre-14f modulo this
substitution — exactly Decision 9's promise.
- e2e.rs: new test gc_handles_recursive_list_construction.
- examples/gc_stress.{ailx,ail.json}: new fixture, builds a 50-
element list via recursive Cons, sums it (1275).
Hash invariance verified: every existing fixture def hash
unchanged (codegen and link line are downstream of canonical
bytes; AST didn't move).
Tests 79 -> 80, all green. Existing 79 byte-identical stdout.
gc_stress -> 1275. list_map_poly -> 2/3/4 unchanged. sort
sorted-list unchanged. cargo doc 0 warnings.
GC notes (pertinent to future work):
- GC_INIT() not needed on Arch libgc 1.5.6 (auto-init via
__attribute__((constructor))).
- No conservative-scan over-retention observed.
- -lgc alone sufficient for link (pthread/dl transitive).
Pattern-shape note from gc_stress fixture writing: the post-14d
"if-then-else" replacement is `(match (app == n 0) (case
(pat-lit true) ...) (case (pat-wild) ...))`. Three lines for
what `if` used to do in one, but uniform with the language.
Worth flagging for the stdlib brief.
Language is feature-complete enough for stdlib. The three
blockers identified at the 14b boundary (redundancy 14d, tail
calls 14e, GC 14f) are all done. Plan 15a: first stdlib module
std_list.ailx with length/append/reverse/map/filter/fold_left/
fold_right/head/tail/is_empty.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Decision 8 ships. Term::App and Term::Do gain tail: bool with
serde-default false and skip-when-false serialisation. New
typecheck pass verify_tail_positions enforces tail-position rules
(Scheme-style propagation through match arms, seq.rhs, let body,
lam body). Codegen emits musttail call for marked App calls.
Hash invariance verified: only the two migrated print_list defs
(list_map_poly.print_list, sort.print_list) changed hashes; all
other defs across all 18 fixtures kept bit-identical hashes —
confirms the skip-when-false serialisation rule works.
Tests 76 -> 79: tail_call_in_non_tail_position_is_rejected,
tail_call_in_tail_position_is_accepted, plus an IR-grep e2e test
asserting that print_list's recursive call site emits musttail
in the lowered IR. Existing 25 e2e tests unchanged in behaviour
(map -> [2,3,4], sort -> sorted list).
IR evidence at the recursive site:
%v7 = musttail call i8 @ail_list_map_poly_print_list(ptr %v6)
ret i8 %v7
Two deviations called out in the implementer report and JOURNAL:
1. tail-do uses tail call, not musttail. Cross-type return
(runtime helpers return i32, AILang Unit is i8) would have
LLVM reject musttail. Path is implemented but not exercised
by any current fixture; proper fix is runtime-helper signature
change, punted.
2. block_terminated flag in codegen so tail-call emit
(musttail call + ret) doesn't get a duplicate trailing ret
from surrounding code (match-arm phi, fn-body, lambda thunk).
Internal plumbing; required for IR well-formedness.
Form (A) productions now at ~30, exactly the constraint-1
budget. Future surface additions need to retire something or
explicit-budget-rebalance in DESIGN.md.
GC notes from implementer survey land in JOURNAL:
- Allocations cluster in lower_ctor; every term-ctor does
malloc(8+8n).
- Tail recursion does not reduce alloc pressure, only stack.
For map-style ctor-blocked recursions, allocation IS the
bottleneck.
- Per-fn arena is sound only when fn return type contains no
boxed ADT. Most current fixtures violate this.
Plan 14f: Boehm conservative GC (GC_malloc, -lgc) as a first
cut. Single-iter integration, no AST/schema change. Stress
test: build a 100k Cons list, observe RSS doesn't blow up.
After 14f the language is feature-complete enough for stdlib
work (15a).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Term::If was semantically a subset of Term::Match on Bool. Per
CLAUDE.md the language must contain no redundancies; two AST nodes
for the same operation produces an authoring decision with no
semantic content and a duplicate codegen path. Removed.
Migration shape (applied to sum, sort, max3 fixtures):
(if c a b) -> (match c (case (lit-bool true) a) (case _ b))
No schema version bump (per user direction): no third-party consumes
ailang/v0, so version ceremony is pure overhead. Edited AST and
fixtures in place; pinned hashes in hash.rs updated.
Implementer deviation, called out and justified: a tightly-scoped
lower_bool_match helper (~95 LOC) was needed in codegen because
the existing match path rejects i1 scrutinees and Pattern::Lit.
Helper accepts only the canonical two-arm migration shape, errors
on anything else, emits the same br/phi IR Term::If used to. No
generalisation of the ADT-match codegen.
Diff: 13 files, +286/-221 (net +65 LOC). AST got smaller
(one variant gone), form-(A) got smaller (one production gone),
typecheck got smaller (one branch gone). Codegen got slightly
larger by the bool-match helper.
Hash deltas: sum.sum, sort.insert, max3.max, max3.max3 changed.
All other defs (e.g. sum.main, sort.IntList, sort.sort,
sort.print_list, max3.main) kept bit-identical hashes — confirms
canonical-JSON byte format intact.
Verification: 76/76 tests green; sum->55, max3->17, sort->[1,1,2,
3,3,4,5,5,5,6,9] (identical to pre-migration). cargo doc 0 warnings.
Tail-call survey by implementer (informs 14e): print_list
recursions are already in tail position (rhs of seq inside match
arm); map/sort/insert recursions are NOT (constructor-blocked
inside Cons applications). 14e annotation will benefit terminal
recursions; ctor-blocked ones need accumulator-form rewrites in
source, not a compiler-side transform.
Decision 7 added to DESIGN.md. JOURNAL entry has the language-
completion sequence (14d done, 14e tail-calls, 14f GC, 15a stdlib).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Strictly additive new crate per Decision 6 architectural pin.
JSON-AST stays the source of truth; ailang-surface is one
producer/consumer of ailang_core::ast::Module values. ailang-check
and ailang-codegen unchanged.
Crate contents:
- src/lex.rs (~264 LOC): 3-rule lexical core. Whitespace and parens
delimit tokens; semicolon to EOL is comment; first-character
classifier (digit -> int, " -> string, else -> ident).
- src/parse.rs (~1041 LOC): hand-written recursive descent. One Rust
fn per EBNF production. No parser-combinator dep.
- src/print.rs (~371 LOC): deterministic pretty-printer. Round-trip
contract with parse() is the surface's correctness gate.
- tests/round_trip.rs (~128 LOC): integration test runs every
examples/*.ail.json fixture through print -> parse -> canonical
JSON, asserts canonical-byte equality with the original.
Two AST-driven form widenings beyond the 14b sketch (both folded
into DESIGN.md Decision 6):
- lam-term carries (typed name type) params, ret type, and
optional effects (Term::Lam has parallel param_tys/ret_ty/
effects fields).
- import-clause admits (import name (as alias)?) (Import.alias
is Option<String>).
Production count ~28, under 30-rule budget. Constraint 1
(formalisable for foreign LLM) intact.
Verification:
- cargo build --workspace green.
- cargo test --workspace: 76 tests green (was 64; +9 surface unit,
+2 round-trip integration). All 17 fixtures round-trip
byte-identical at canonical level. 3 hand-written .ailx
exhibits parse to canonical JSON identical to their .ail.json
siblings.
- cargo doc --no-deps zero warnings (DESIGN.md item 6 invariant).
Manual smoke test (ail parse → ail run): hello, box,
list_map_poly all produce expected output through the form-(A)
authoring lane end-to-end.
CLI: ail parse <file.ailx> [-o <file.ail.json>]. .ail.json
remains a first-class input to every existing subcommand.
Plan 14d: stdlib (std_list.ailx with length/filter/fold/concat/
reverse/head/tail), authored in form (A) from day one.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User clarified the architectural intent: the data structure is
the source of truth. The textual surface is one projection
among potentially many — visual / graphical front-ends are an
explicitly anticipated future lane.
The previous draft of Decision 6 said "JSON-AST demoted to
storage and exchange only", which was the wrong framing. The
JSON-AST stays canonical; form (A) is the AI authoring
projection but does not displace the AST.
Changes:
- New "Architectural pin: data structure is the source of truth"
subsection.
- Implementation outline reworded: parser is "strictly additive",
no schema changes, .ail.json remains a first-class input to
every existing subcommand.
- New "What this Decision deliberately does not do" subsection
to lock in the architectural commitments.
No code change. Pure documentation correction to the 14b
design pass before 14c starts.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
User redirected at iter boundary: writing a stdlib in JSON was
the wrong move. The language is supposed to be the one I program
*best* in, and JSON-AST authoring is rationalisation, not
strength.
DESIGN.md Decision 6 captures the constraints that fall out of
the "formalisable for a foreign LLM" hard requirement (no
precedence, no semantic indentation, ASCII only, every AST node
a uniquely-tagged form), sketches three candidate notations with
the same `map` encoded in each, and picks form (A) — fully-tagged
S-expressions — as the first attempt with explicit rollback path
to form (C) if (A) hurts authoring.
Form (A) shape:
- 3-rule lexical core: sexpr / atom / token-classified-by-
first-character.
- Every AST node has a unique head keyword. No case-rule (no
"capitalised head means ctor"); ctors are explicit via
`(term-ctor TypeName CtorName args)` and `(pat-ctor CtorName
fields)`.
- Bare atoms get their sort from the parent slot (type-var
inside `(con NAME args)`, term-var inside `(app HEAD args)`,
pat-var inside `(pat-ctor CTOR fields)`, integer literal in
term position, etc.).
Empirical exhibits, hand-encoded:
- examples/hello.ailx 5 LOC (JSON was 36 pretty / 21 canonical)
- examples/box.ailx 25 LOC (JSON was 160 / 88)
- examples/list_map_poly.ailx 50 LOC (JSON was 394 / 230)
4-8x line reduction, ~4x character reduction. Bigger gains on
bigger programs since overhead is proportional to AST depth.
None are parseable yet — header comments say "Iter 14b design
exhibit, parser lands in 14c".
Two small spec issues caught while writing the exhibits and
folded back into DESIGN.md before committing:
- Operator idents (`+`, `==`) need the token-by-first-char
classification rule, not a word-shaped regex.
- Bool literals (`true`/`false`) reserved in term context;
unit is explicit `(lit-unit)`.
Tests unchanged (this iter is paper). 25/25 e2e green.
cargo doc --no-deps zero warnings.
Plan 14c: new crate `ailang-surface` with PEG parser, round-trip
hash-equivalence gate against every existing `examples/*.ail.json`,
CLI subcommand `ail parse`. If round-trip holds, stdlib starts
in `.ailx` form (Iter 14d).
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds ailang-docwriter to /agents/ — a recurring role for keeping
crate-, module-, and pub-item-level rustdoc accurate. First mission:
ailang-core. Crate root, every module root, every pub item documented;
intra-doc links throughout; Iter-13a additions (TypeDef.vars, Type::Con.args)
get an explicit backwards-compat note. Two stale broken-link warnings in
ailang-check fixed in passing. cargo doc --no-deps now warning-free across
the workspace; promoted to verification invariant 6 in DESIGN.md.
DESIGN.md: removes "No parameterised ADTs" from the gap list,
adds an entry under "what is supported" that names the per-use-
site substitution scheme and the `llvm_type(Type::Var)` hard-error
defence. Smoke-test list extended with `box.ail.json` and
`maybe_int.ail.json`. Boundary snapshot moved from "end of Iter 12"
to "end of Iter 13".
JOURNAL.md: single Iter 13 entry covering 13a/b/c. Records the
hash-invariant regression test, the architect-flagged debt I
deliberately did not touch (poly-fn-as-value asymmetry, builtins
triple-source, `synth_arg_type` shortcuts on If/Match), the
KISS observation that 13b's "no mono-queue for types" was the
right call, and the process note that this was the first iter
worked strictly through `/agents/` after the role pin in 3df943d.
Plan iteration 14 queued: list_map-as-`List a` rewrite, GC/arena,
poly-fn-as-value.
DESIGN.md:
- "What is not (yet) supported" updated for end-of-Iter-12 boundary:
parameterised ADTs replace HM-inside-bodies as the headline gap;
polymorphism limitations (direct-call-only, no higher-rank) are
spelled out so future me doesn't trip over them.
- "What is supported" gains the polymorphism + monomorphisation
bullet, with the descriptor scheme written out.
- Smoke-test list extended with poly_id and poly_apply.
JOURNAL.md: Iter 12a/b retrospective. Notes the metavar-encoding
choice (Type::Var{name:"$m<n>"} vs new variant) and why the
codegen-side type tracker was preferable to a typechecker
sidetable for now. Architecture self-check confirms the language
is now usable for poly-flavoured programs over primitives. Plan 13:
parameterised ADTs as the natural next step (without them, generic
map remains hand-monomorphised).
Iter 12c was originally to include a polymorphic map rewrite —
dropped because parameterised ADTs are the prerequisite. The two
new examples (poly_id, poly_apply, both already in 12b) cover the
real e2e proof.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
examples/sort.ail.json: insertion sort over IntList. Defines `insert
:: Int -> IntList -> IntList` and `sort :: IntList -> IntList`
recursively, plus print_list using Iter 10's seq. Sorts an 11-element
input and prints `1 1 2 3 3 4 5 5 5 6 9` one-per-line.
Validates that the language handles deeper ADT recursion + branching
(if + <=) + ctor construction + IO sequencing without surprises.
Wrote, typechecked, ran first try.
The Iter 11 plan called for polymorphism, but on reflection the right
move was one more validation cycle before disturbing the pipeline.
Polymorphism is now queued explicitly as Iter 12 (12a typechecker
substitution, 12b codegen monomorphisation, 12c docs + generic
example).
Tests: 52 green (was 51). New e2e `insertion_sort_orders_list`.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Adds `Term::Seq { lhs, rhs }` (serde tag "seq") as a first-class
AST node for sequencing effectful expressions. Equivalent in
behaviour to `let _ = lhs in rhs`, but the dedicated node gives the
pretty-printer and diagnostics a cleaner shape and surfaces the
intent ("run for effect, then yield rhs") to future tooling.
Typecheck: lhs must be Unit; rhs's type is the result; effects
accumulate.
Codegen: lower lhs (drop SSA), lower rhs (return).
Capture / deps walkers: recurse into both sides.
Refactored examples/list_map.ail.json's print_list to use seq
instead of `let _ = ...`. Output unchanged (2/4/6).
Hash stability: existing examples without Term::Seq serialise
bit-identical; only list_map.ail.json's hashes changed (deliberate
refactor).
Tests: 51 green (was 50). New unit test
`ailang_check::tests::seq_lhs_must_be_unit` covers the type-error
path; existing list_map e2e covers the happy path.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
DESIGN.md:
- CLI block: add `ail run`.
- Smoke-test list: add list_map.ail.json (the dogfood example).
JOURNAL: append Iter 9 entry covering the dogfood result, two
friction points surfaced (single-arg fn-type pretty-print parens;
absence of a sequencing operator), the `ail run` CLI helper, and a
ranked Iter 10 plan with three candidates (polymorphism, `;`, GC).
Tentative pick is (2) `;` next, (1) polymorphism for Iter 11.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
DESIGN.md:
- Term schema: add `ctor`, `match`, `lam` rows. The schema fragment is
now exhaustive for the supported language; prior versions silently
dropped Iter 3/Iter 8 additions.
- "What is not (yet) supported": closures-with-capture moves out of
pending. New positive bullet for anonymous lambdas. Polymorphism
added as an explicit pending bullet (Forall is parseable but not
inferred). Smoke-test list extended with closure.ail.json.
JOURNAL: append Iter 8 entry covering both 8a (closure-pair ABI) and
8b (Term::Lam + capture). Includes the rationale for skipping TIR
(KISS won — `synth` already provides enough type info), the closure-
conversion sketch, hash stability check (existing examples produce
the same fn hashes pre vs post Iter 8), and the Iter 9 plan with
two candidates ranked.
Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
Brummel