fieldtest: raw-buf comprehensive — 4 examples, 8 findings (refs #7)

Comprehensive usability re-test of the RawBuf kernel extension in
natural LLM-author decompositions the prior field test (0058) did not
exercise: a Float RawBuf in a user ADT read through two borrow helpers,
a Bool flag buffer filled with a computed value, an Int fib buffer whose
fill reads its own writes plus two composed borrow summaries, and a
forbidden core-primitive element. All three working fixtures build, run
to expected values, and report live=0 across Float/Bool/Int widths —
the core RawBuf surface (borrow-helper composition, fill loops,
RawBuf-in-ADT, element widths) genuinely holds together.

Orchestrator triage corrected the fieldtester's run, which executed a
stale target/release/ail (built before the #50 fix). Every outcome was
re-verified against a fresh build:

- F2 (bare RawBuf ADT field unresolved) — RETRACTED: a stale-binary
  false positive. On a fresh build the bare name resolves in every
  configuration (ctor name == or != type name; builds/runs live=0). The
  #50 fix is correct and general.
- F7 (NEW, issue #51) — a `check`-clean program crashes `build`: a
  buffer whose element type is never observed by a later get/set
  defaults its element to Unit in monomorphisation and hits an
  unregistered `RawBuf_new__Unit` intercept. Affects a legitimate
  `RawBuf<Int>` used only for its size AND a forbidden locally-
  constructed element. Root: the `(new T ...)` desugar drops the
  explicit element annotation; honouring it reverses the milestone's
  § Term::New desugar decision, so it routes through brainstorm.
- F8 (process) — the field test ran stale; the fieldtester agent now
  builds from the current tree before running (plugin fix).

F1 (spec_gap) resolved here: the design ledger's §Series substrate ctor
wrote the storage field `(own (RawBuf a))`, which does not parse (`own`
is fn-param/ret only). Rewritten to the verified-authorable `(con RawBuf
a)`. The rest of the §Series listing is illustrative and its element-
less `(new RawBuf lookback)` construction is entangled with #51; a full
compile-verification of that listing belongs with the Series-substrate
work.

F5 (param-in rejects a forbidden core primitive, message names the set)
and F4/F3 (borrow composition; substrate composite) carry on. F6
(param-in set rendered as a Rust-debug list) filed as #52.

refs #7
This commit is contained in:
2026-05-30 18:29:00 +02:00
parent f488d314e8
commit 1b2d23ec42
6 changed files with 531 additions and 1 deletions
@@ -0,0 +1,92 @@
; Fieldtest raw-buf comprehensive .1 (Axes 1+2+3+4 combined — the
; "Series substrate" shape with a Float element width).
;
; Task an LLM author is naturally given: "model a fixed-size window of
; Float samples with a running total, fill it from a loop, then report
; the window mean and the window maximum through read-only helpers."
;
; The natural decomposition wraps a RawBuf<Float> together with a
; bookkeeping `count` field in a user ADT (Window), exactly the Series
; substrate the ledger (design/models/0007 §Series) describes — an ADT
; holding `(own (RawBuf a))` plus Int bookkeeping. The buffer is filled
; via a (loop ...)/recur (runtime-N, not literal indices), and the
; read-back is factored into two `borrow (Window)` helpers (`mean`,
; `wmax`) that each pattern-match the ADT and drive a read loop off the
; bookkeeping count, reading the buffer through `RawBuf.get` on a borrow.
;
; This stresses, in one program:
; - Float element width through a fill loop (not just literal sets).
; - A RawBuf stored in a user ADT field (the #50 substrate shape).
; - TWO borrow-receiver helpers composing over the same wrapped buffer
; (the #46 borrow-receiver-read fix, under composition + behind a
; match on the ADT).
; - The drop cascade: the Window ADT owns the buffer; dropping the
; Window frees the slab. Leak-clean expected under AILANG_RC_STATS.
;
; Window of 4 samples: 2.0, 4.0, 6.0, 8.0.
; mean = 20.0 / 4 = 5.0
; wmax = 8.0
; Expected stdout: two lines, "5.0" then "8.0".
(module rbx_1_float_window_stats
; NOTE (field test): the ledger's §Series substrate form writes the
; storage field as `(own (RawBuf a))`. That does NOT parse here — `own`
; in a ctor field position is rejected (surface-parse-error). And the
; bare `(con RawBuf (con Float))` field gives a type-mismatch
; (expected RawBuf<Float>, got raw_buf.RawBuf<Float>). The only form
; that checks is the fully-qualified `raw_buf.RawBuf`. See findings.
(data Window
(ctor W (con raw_buf.RawBuf (con Float)) (con Int)))
(fn build_window
(doc "Allocate a 4-slot Float buffer, fill slot i with (2*(i+1)), wrap in a Window with count=4.")
(type (fn-type (params (con Int)) (ret (own (con Window)))))
(params n)
(body
(let buf (new RawBuf (con Float) 4)
(let filled
(loop (b (con RawBuf (con Float)) buf) (i (con Int) 0)
(if (app ge i n)
b
(recur (app RawBuf.set b i (app int_to_float (app * 2 (app + i 1))))
(app + i 1))))
(term-ctor Window W filled n)))))
(fn mean
(doc "Average of the first count slots, read through a borrow of the Window.")
(type (fn-type (params (borrow (con Window))) (ret (con Float))))
(params w)
(body
(match w
(case (pat-ctor W buf count)
(app /
(loop (acc (con Float) 0.0) (i (con Int) 0)
(if (app ge i count)
acc
(recur (app + acc (app RawBuf.get buf i))
(app + i 1))))
(app int_to_float count))))))
(fn wmax
(doc "Maximum of the first count slots, read through a borrow of the Window.")
(type (fn-type (params (borrow (con Window))) (ret (con Float))))
(params w)
(body
(match w
(case (pat-ctor W buf count)
(loop (mx (con Float) 0.0) (i (con Int) 0)
(if (app ge i count)
mx
(recur (if (app float_gt (app RawBuf.get buf i) mx)
(app RawBuf.get buf i)
mx)
(app + i 1))))))))
(fn main
(type (fn-type (params) (ret (con Unit)) (effects IO)))
(params)
(body
(let w (app build_window 4)
(seq (seq (app print (app mean w)) (do io/print_str "\n"))
(seq (app print (app wmax w)) (do io/print_str "\n")))))))
+57
View File
@@ -0,0 +1,57 @@
; Fieldtest raw-buf comprehensive .2 (Axes 1+2+3 — Bool element width
; through a fill loop, then counted via a borrow helper).
;
; Task an LLM author is naturally given: "mark which of the numbers
; 0..N are even in a flag buffer, then count how many flags are set."
; A flat Bool RawBuf is the natural storage for a flag/bitset array.
;
; The prior field test only stored a single Bool at a literal index
; (rawbuf_3_sensor_pair). This drives the i1/i8 Bool packing edge
; through a (loop ...)/recur FILL of runtime length N (one RawBuf.set
; per slot, the stored value being a *computed* Bool — the result of
; `eq (i % 2) 0`), and reads the flags back through a `borrow` helper
; (`count_set`) that drives off RawBuf.size and accumulates an Int.
;
; This stresses:
; - Bool stored/loaded across a loop, not a literal set — does the
; i1/i8 round-trip hold when the value is computed, not a `true`
; literal?
; - A borrow-receiver helper over a Bool buffer (the #46 fix on the
; Bool width).
; - `RawBuf.get` of a Bool used directly as an `if` condition inside
; the count loop.
;
; N = 6 -> flags for 0,1,2,3,4,5 -> even at 0,2,4 -> 3 set.
; Expected stdout: "3".
(module rbx_2_bool_sieve
(fn mark_even
(doc "Set slot i to (i is even) for i in 0..n. Linear own->own.")
(type (fn-type (params (own (con RawBuf (con Bool))) (con Int)) (ret (own (con RawBuf (con Bool))))))
(params buf n)
(body
(loop (b (con RawBuf (con Bool)) buf) (i (con Int) 0)
(if (app ge i n)
b
(recur (app RawBuf.set b i (app eq (app % i 2) 0))
(app + i 1))))))
(fn count_set
(doc "Count the set flags, reading the Bool buffer through a borrow.")
(type (fn-type (params (borrow (con RawBuf (con Bool)))) (ret (con Int))))
(params buf)
(body
(loop (acc (con Int) 0) (i (con Int) 0)
(if (app ge i (app RawBuf.size buf))
acc
(recur (if (app RawBuf.get buf i) (app + acc 1) acc)
(app + i 1))))))
(fn main
(type (fn-type (params) (ret (con Unit)) (effects IO)))
(params)
(body
(let buf (new RawBuf (con Bool) 6)
(let buf (app mark_even buf 6)
(app print (app count_set buf)))))))
@@ -0,0 +1,29 @@
; Fieldtest raw-buf comprehensive .3 (Axis 5 — the param-in element-type
; restriction on a forbidden *core primitive*).
;
; The prior field test rejected RawBuf<Str> (the shipped fixture) and
; RawBuf<Tick> (a user ADT). This probes the third natural mistake: a
; RawBuf of `Unit` — a core primitive that is NOT in the allowed set
; {Int, Float, Bool}. An LLM author might reach for `RawBuf<Unit>` as a
; "presence buffer" / counter-of-slots before realising a Bool flag
; buffer is the right tool. The restriction must reject it, and per the
; B3 fix (#48) the diagnostic should now NAME the allowed set
; {Int, Float, Bool}, not just the offending type.
;
; Expected: `ail check` exits 1 with [param-not-in-restricted-set]
; naming `Unit` AND the allowed set {Int, Float, Bool}.
(module rbx_3_unit_element_reject
(fn make_presence
(doc "Try to allocate a RawBuf of Unit — forbidden element type.")
(type (fn-type (params (con Int)) (ret (own (con RawBuf (con Unit))))))
(params n)
(body (new RawBuf (con Unit) n)))
(fn main
(type (fn-type (params) (ret (con Unit)) (effects IO)))
(params)
(body
(let buf (app make_presence 3)
(do io/print_str "unreachable\n")))))
@@ -0,0 +1,76 @@
; Fieldtest raw-buf comprehensive .4 (Axes 1+2+3 — Int buffer, an
; own->own fill helper across a function boundary, plus TWO distinct
; borrow-receiver helpers composed over the same buffer).
;
; Task an LLM author is naturally given: "fill a buffer of N slots with
; the Fibonacci sequence, then report both the total and the index of
; the largest element." The natural decomposition is three helpers:
; - `fill` : (own RawBuf, Int) -> own RawBuf (loop/recur, runtime N)
; - `total` : (borrow RawBuf) -> Int (read loop off .size)
; - `argmax`: (borrow RawBuf) -> Int (read loop, tracks idx)
; and a main that threads the owned buffer through `fill`, then borrows
; it TWICE in sequence (`total buf`, then `argmax buf`) — the natural
; "compute several summaries of one read-only buffer" shape.
;
; This stresses the #46 borrow-receiver fix under composition: two
; independent borrows of the same owned binder, each a helper that does
; multiple RawBuf.get / RawBuf.size calls on its borrow parameter. The
; prior rawbuf_1 had a single borrow helper; this has two, called back
; to back on the same buffer, with the buffer still owned (and dropped)
; by main afterwards.
;
; Fib(0..5) = 1,1,2,3,5,8 (seeding fib(0)=fib(1)=1).
; total = 1+1+2+3+5+8 = 20
; argmax = index 5 (value 8 is largest)
; Expected stdout: "20" then "5".
(module rbx_4_int_two_borrow_helpers
(fn fill
(doc "Fill slot i with fib(i) for i in 0..n. Linear own->own.")
(type (fn-type (params (own (con RawBuf (con Int))) (con Int)) (ret (own (con RawBuf (con Int))))))
(params buf n)
(body
; Seed slots 0 and 1 to 1, then each later slot = sum of prior two,
; read back from the buffer itself via RawBuf.get on the owned `b`.
(loop (b (con RawBuf (con Int)) buf) (i (con Int) 0)
(if (app ge i n)
b
(recur (app RawBuf.set b i
(if (app le i 1)
1
(app + (app RawBuf.get b (app - i 1))
(app RawBuf.get b (app - i 2)))))
(app + i 1))))))
(fn total
(doc "Sum every slot, borrow receiver.")
(type (fn-type (params (borrow (con RawBuf (con Int)))) (ret (con Int))))
(params buf)
(body
(loop (acc (con Int) 0) (i (con Int) 0)
(if (app ge i (app RawBuf.size buf))
acc
(recur (app + acc (app RawBuf.get buf i))
(app + i 1))))))
(fn argmax
(doc "Index of the largest slot, borrow receiver. Tracks (best_i, best_v).")
(type (fn-type (params (borrow (con RawBuf (con Int)))) (ret (con Int))))
(params buf)
(body
(loop (best_i (con Int) 0) (best_v (con Int) (app RawBuf.get buf 0)) (i (con Int) 0)
(if (app ge i (app RawBuf.size buf))
best_i
(if (app gt (app RawBuf.get buf i) best_v)
(recur i (app RawBuf.get buf i) (app + i 1))
(recur best_i best_v (app + i 1)))))))
(fn main
(type (fn-type (params) (ret (con Unit)) (effects IO)))
(params)
(body
(let buf (new RawBuf (con Int) 6)
(let buf (app fill buf 6)
(seq (seq (app print (app total buf)) (do io/print_str "\n"))
(seq (app print (app argmax buf)) (do io/print_str "\n"))))))))