Chez Phase 1 (increment 2): live analyzer -> Chez, var cells, RT, mandelbrot

Wire the real pipeline end to end: host/chez/driver.janet boots a compile-mode
jolt ctx, runs the EXISTING Janet-hosted analyzer on actual Clojure source to
real IR, feeds it to the Scheme emitter, and runs the result on Chez. Analysis
stays on Janet (the analyzer ports to Chez in Phase 2); execution is on Chez.

emit.janet now consumes live IR (pv/phm-normalized like the Janet backend) and
covers what the analyzer actually emits, not the hand-built inc-1 shapes:
- core ops arrive as :var clojure.core/+ etc., not :rt — lowered to native
  Scheme via a native-ops table (mirrors backend.janet's), `=` to jolt=.
- var cells (host/chez/rt.ss): :def -> def-var!, :var -> var-deref. Late binding
  so cross-var calls (run -> count-point) and the entry crossing resolve at use.
- named fns (defn / fn self-name) bind via letrec so self-recursion resolves.
- unsupported stdlib/host refs (no core on Chez yet) are rejected at EMIT time
  (clean out-of-subset signal) instead of deref'ing to nil and failing at runtime.

Number model: jolt is all-doubles (no ratios; (/ 1 2) is 0.5), so literals emit
as flonums — matches the Janet host and keeps Chez out of exploding exact
rationals (mandelbrot). jolt-num->string prints integer-valued without ".0".

Two real bugs found via the corpus probe and fixed (regression rows added):
- loop bound in parallel (Scheme named-let) but Clojure loop is sequential — a
  later init must see earlier bindings; wrap a let* around the loop.
- #(...) shorthand gensyms params with a trailing `#`, invalid in Scheme — munge
  it to `_`.

Gate: test/chez/emit-test.janet runs the real analyzer -> Chez for (+ 1 2),
fib(30)=832040, mandelbrot run(40), and the two regressions, parity-checked
against the Janet oracle (6/6). First parity number via the new subset probe
(test/chez/run-corpus-chez.janet, JOLT_CHEZ_CORPUS=1): 182/182 compiled corpus
cases pass, 0 divergences; 2473/2655 out of subset pending core on Chez. Full
jpm/run-tests gate green (125 files). Chez tests skip cleanly without `chez`.

Perf note (unchanged plan): emitted fib(30) ~23ms vs hand-Scheme ~5ms — the
jolt-truthy? wrapper (~3x) plus flonum (not fixnum) arithmetic, both Phase-4
type-specialization levers.

host/chez/driver.janet

@@ -0,0 +1,73 @@
+# Phase 1 (jolt-cf1q.2) — live-analyzer -> Chez driver.
+#
+# Boots a real jolt ctx, runs the EXISTING Janet-hosted analyzer on actual
+# Clojure source to produce host-neutral IR, feeds that IR to the Scheme emitter
+# (emit.janet), and assembles a runnable Chez program. This is the Option-2
+# backend swap end to end: same front end, Scheme back end, run on Chez.
+#
+# Analysis still happens on Janet here (the analyzer is portable Clojure but not
+# yet bootstrapped onto Chez — that's Phase 2); EXECUTION happens on Chez. The
+# point of this increment is to validate that the real IR the analyzer emits
+# compiles to correct, fast Scheme.
+
+(import ../../src/jolt/api :as api)
+(import ../../src/jolt/backend :as backend)
+(import ../../src/jolt/reader :as r)
+(import ../../src/jolt/evaluator :as evlr)
+(import ./emit :as emit)
+
+(defn chez-available?
+  "True when a `chez` binary is on PATH — lets the chez tests skip cleanly on
+  hosts without it (CI without Chez), like the clojure-test-suite skips when its
+  corpus dir is absent."
+  []
+  (def r (protect (let [p (os/spawn ["chez" "--version"] :p {:out :pipe :err :pipe})]
+                    (ev/read (p :out) 1024)
+                    (ev/read (p :err) 1024)
+                    (os/proc-wait p))))
+  (and (r 0) (zero? (r 1))))
+
+(defn make-ctx []
+  "A compile-mode jolt ctx (the analyzer pipeline is only built under :compile?)."
+  (api/init {:compile? true}))
+
+(defn- parse-all [src]
+  (def out @[])
+  (var s src)
+  (while (> (length (string/trim s)) 0)
+    (def parsed (r/parse-next s))
+    (set s (in parsed 1))
+    (def f (in parsed 0))
+    (unless (nil? f) (array/push out f)))
+  out)
+
+(defn compile-program
+  "Compile a Clojure program string to a runnable Chez program. Every top-level
+  form is analyzed to real IR and emitted to Scheme; all but the LAST form are
+  treated as defs (also interned in the ctx so later forms resolve their vars),
+  and the last form is the expression whose value the program prints."
+  [ctx src]
+  (def forms (parse-all src))
+  (assert (> (length forms) 0) "compile-program: empty program")
+  (def n (length forms))
+  (def def-scm @[])
+  (for i 0 (- n 1)
+    (def f (in forms i))
+    # emit the def, then intern it (interpreted) so a later form's reference to
+    # this var resolves to a :var node rather than an unresolved symbol.
+    (array/push def-scm (emit/emit (backend/analyze-form ctx f)))
+    (evlr/eval-form ctx @{} f))
+  (def final-scm (emit/emit (backend/analyze-form ctx (in forms (- n 1)))))
+  (emit/program def-scm final-scm))
+
+(defn run-on-chez
+  "Compile `src` and run it on Chez; returns [exit-code stdout stderr]."
+  [ctx src &opt scheme-out]
+  (def prog (compile-program ctx src))
+  (def path (or scheme-out "/tmp/chez-jolt-prog.ss"))
+  (spit path prog)
+  (def proc (os/spawn ["chez" "--script" path] :p {:out :pipe :err :pipe}))
+  (def out (ev/read (proc :out) 0x100000))
+  (def err (ev/read (proc :err) 0x100000))
+  (def code (os/proc-wait proc))
+  [code (string/trim (if out (string out) "")) (string/trim (if err (string err) ""))])

host/chez/emit.janet

@@ -1,29 +1,64 @@
 # Phase 1 — jolt IR -> Chez Scheme emitter (jolt-cf1q.2).
 #
 # The new back end: consumes the SAME host-neutral IR (jolt.ir, see
-# jolt-core/jolt/ir.clj) the analyzer produces and the Janet backend consumes,
-# but emits Scheme source text instead of Janet. `host/compile` (Chez `eval`)
-# turns that into a procedure. This increment covers the pure-functional subset
-# (const/local/var/rt/if/do/let/fn/invoke/def/loop/recur) — enough to run
-# fib/mandelbrot-shaped code through the REAL IR. Globals are early-bound here;
-# var-cell late binding is the next increment.
+# jolt-core/jolt/ir.clj) the live analyzer produces and the Janet backend
+# consumes, but emits Scheme source text instead of Janet. `host/compile` (Chez
+# `eval`) turns that into a procedure. Covers the pure-functional + numeric
+# subset (const/local/var/host/if/do/let/fn/invoke/def/loop/recur) — enough to
+# run fib/mandelbrot-shaped code through the REAL analyzer.
 #
-# IR nodes are plain :op-tagged structs/tables (keyword keys), matching ir.clj.
+# IR access mirrors the Janet backend: live IR fields are jolt VALUES — vectors
+# are persistent (pv), and a nil-valued node densifies to a phm. `nn`/`vv` below
+# normalize both into Janet structs/arrays, so the same code drives hand-built
+# IR (the unit tests) and live analyzer output (the driver).
 
-(def rt-map
-  # jolt RT primitive name -> Scheme. = is the exactness-aware jolt= from
-  # values.ss; inc/dec/quot get preamble shims. Arithmetic/compare are native.
+(import ../../src/jolt/pv :as pv)
+(import ../../src/jolt/phm :as phm)
+
+# Normalize a node (phm -> struct) and a vector field (pvec -> array view); both
+# pass plain Janet values through untouched, so hand-built IR still works.
+(defn- nn [n] (if (phm/phm? n) (phm/phm-to-struct n) n))
+(defn- vv [x] (if (pv/pvec? x) (pv/pv->array x) x))
+
+# Hot clojure.core primitives lowered to native Scheme, mirroring the Janet
+# backend's native-ops (documented numbers-only relaxation). `=` is the
+# exactness-aware jolt= from values.ss; inc/dec/not are rt shims; mod/rem/quot
+# map to Scheme's (correct: Scheme has all three, unlike Janet which lacked quot).
+(def- native-ops
   {"+" "+" "-" "-" "*" "*" "/" "/"
    "<" "<" ">" ">" "<=" "<=" ">=" ">="
-   "=" "jolt=" "inc" "jolt-inc" "dec" "jolt-dec"
-   "mod" "modulo" "quot" "quotient" "rem" "remainder"})
+   "=" "jolt=" "inc" "jolt-inc" "dec" "jolt-dec" "not" "jolt-not"
+   "min" "min" "max" "max"
+   "mod" "modulo" "rem" "remainder" "quot" "quotient"})
+
+# Unary ops only legal at arity 1; binary at arity 2. Others (arith/compare) are
+# variadic in both Scheme and jolt, so any arity is fine.
+(def- unary-ops {"inc" true "dec" true "not" true})
+(def- binary-ops {"mod" true "rem" true "quot" true})
+
+# If fnode is a clojure.core (or host) ref to a native-op primitive, return the
+# Scheme op string — only at an arity where the Scheme op and the jolt fn agree.
+(defn- native-op [fnode nargs]
+  (def nm (case (get fnode :op)
+            :var (when (= "clojure.core" (get fnode :ns)) (get fnode :name))
+            :host (get fnode :name)
+            nil))
+  (def op (and nm (get native-ops nm)))
+  (cond
+    (nil? op) nil
+    (and (get unary-ops nm) (not= nargs 1)) nil
+    (and (get binary-ops nm) (not= nargs 2)) nil
+    op))
 
 (var- recur-target nil)
 (var- gensym-n 0)
 (defn- fresh-label [prefix] (string prefix (++ gensym-n)))
 
-# MVP: jolt local/var names are valid Scheme identifiers (inc, even?, + all are).
-(defn- munge [name] name)
+# Most jolt names are already valid Scheme identifiers (inc, even?, +, ->str all
+# are — Scheme allows ! $ % & * + - . / : < = > ? @ ^ _ ~). The one that isn't is
+# `#`, which jolt auto-gensyms use as a suffix (e.g. p1__0000X4# from #(...)
+# shorthand) — `#` starts a datum in Scheme, so replace it with `_`.
+(defn- munge [name] (string/replace-all "#" "_" name))
 
 (var emit nil)   # forward declaration (mutual recursion with the helpers below)
 
@@ -31,76 +66,126 @@
   (cond
     (nil? v) "jolt-nil"
     (boolean? v) (if v "#t" "#f")
-    (number? v) (string v)
+    # jolt models every number as a double (no ratios/bignums; see reader.janet).
+    # Emit flonums so arithmetic matches the Janet host and Chez doesn't fall into
+    # exploding exact rationals (mandelbrot). Integer-valued -> append ".0".
+    (number? v) (let [s (string v)]
+                  (if (or (string/find "." s) (string/find "e" s) (string/find "n" s))
+                    s
+                    (string s ".0")))
     (string? v) (string/format "%j" v)   # quoted+escaped string literal
     (errorf "emit-const: unsupported literal %p" v)))
 
 (defn- emit-binding [b]
+  (def b (vv b))
   (string "(" (munge (get b 0)) " " (emit (get b 1)) ")"))
 
 (defn- emit-let [node]
-  (string "(let* (" (string/join (map emit-binding (get node :bindings)) " ") ") "
+  (string "(let* (" (string/join (map emit-binding (vv (get node :bindings))) " ") ") "
           (emit (get node :body)) ")"))
 
 (defn- emit-loop [node]
   (def label (fresh-label "loop"))
-  (def bs (string/join (map emit-binding (get node :bindings)) " "))
+  (def pairs (map vv (vv (get node :bindings))))
+  (def names (map |(munge (get $ 0)) pairs))
+  # inits are evaluated in the OUTER scope (recur-target unchanged) and, like
+  # Clojure loop/let, SEQUENTIALLY — a later init sees earlier bindings. Scheme's
+  # named `let` binds in parallel, so wrap a sequential let* around the loop.
+  (def inits (map |(emit (get $ 1)) pairs))
+  (def seq-bs (string/join (map (fn [n i] (string "(" n " " i ")")) names inits) " "))
+  (def rebinds (string/join (map (fn [n] (string "(" n " " n ")")) names) " "))
   (def prev recur-target)
   (set recur-target label)
   (def body (emit (get node :body)))
   (set recur-target prev)
-  (string "(let " label " (" bs ") " body ")"))
+  (string "(let* (" seq-bs ") (let " label " (" rebinds ") " body "))"))
 
 (defn- emit-recur [node]
   (unless recur-target (error "emit: recur outside a loop/fn target"))
-  (string "(" recur-target " " (string/join (map emit (get node :args)) " ") ")"))
+  (string "(" recur-target " " (string/join (map emit (vv (get node :args))) " ") ")"))
 
 (defn- emit-fn [node]
-  (def arities (get node :arities))
+  (def arities (map nn (vv (get node :arities))))
   (when (not= 1 (length arities)) (error "emit: multi-arity fn not in this increment"))
   (def a (first arities))
   (when (get a :rest) (error "emit: variadic fn not in this increment"))
-  (def params (map munge (get a :params)))
+  (def params (map munge (vv (get a :params))))
   # wrap the body in a named let so fn-level `recur` rebinds the params
   (def label (fresh-label "fnrec"))
   (def prev recur-target)
   (set recur-target label)
   (def body (emit (get a :body)))
   (set recur-target prev)
-  (string "(lambda (" (string/join params " ") ") "
-          "(let " label " (" (string/join (map (fn [p] (string "(" p " " p ")")) params) " ") ") "
-          body "))"))
+  (def lambda
+    (string "(lambda (" (string/join params " ") ") "
+            "(let " label " (" (string/join (map (fn [p] (string "(" p " " p ")")) params) " ") ") "
+            body "))"))
+  # A named fn (defn / (fn self [..])) references itself by name — the analyzer
+  # binds that name as a :local in the body. letrec makes the name visible to the
+  # lambda so self-calls resolve (recur stays a separate self-call to the arity).
+  (if-let [nm (get node :name)]
+    (let [m (munge nm)] (string "(letrec ((" m " " lambda ")) " m ")"))
+    lambda))
+
+# The Clojure stdlib (clojure.core, clojure.math, clojure.string, …) and host
+# interop (Math/sqrt etc.) have no implementation on Chez yet (Phase 2+). A
+# reference to one — except a clojure.core call lowered to a native op — is
+# genuinely uncompilable here. Reject it at emit time (a clean "out of subset"
+# signal) rather than emitting a var-deref that resolves to nil and fails
+# confusingly at runtime.
+(defn- stdlib-var? [n]
+  (and (= :var (get n :op)) (string/has-prefix? "clojure." (or (get n :ns) ""))))
+
+(defn- emit-invoke [node]
+  (def fnode (nn (get node :fn)))
+  (def args (map emit (vv (get node :args))))
+  (def nop (native-op fnode (length args)))
+  (cond
+    # zero-arg + / * : Scheme's identity is the EXACT 0 / 1, but jolt models every
+    # number as a double, so emit the flonum identity to keep (= 0 (+)) true.
+    (and nop (empty? args) (= nop "+")) "0.0"
+    (and nop (empty? args) (= nop "*")) "1.0"
+    nop (string "(" nop " " (string/join args " ") ")")
+    (stdlib-var? fnode)
+    (errorf "emit: unsupported stdlib fn `%s/%s` (no core on Chez yet)" (get fnode :ns) (get fnode :name))
+    (= :host (get fnode :op))
+    (errorf "emit: unsupported host call `%s` (no host interop on Chez yet)" (get fnode :name))
+    (string "(" (emit fnode) " " (string/join args " ") ")")))
 
 (set emit (fn emit [node]
+  (def node (nn node))
   (case (get node :op)
     :const (emit-const (get node :val))
     :local (munge (get node :name))
-    :var   (munge (get node :name))           # early-bound (MVP)
-    :rt    (or (get rt-map (get node :name))
-               (errorf "emit: unmapped rt primitive %s" (get node :name)))
-    :host  (get node :name)
+    # late-bound var: read the cell's current root at use time. A value-position
+    # ref to a stdlib var (e.g. passing `inc` to (map inc xs)) needs a real fn,
+    # which native-op lowering doesn't provide — so it's out of subset regardless.
+    :var   (if (stdlib-var? node)
+             (errorf "emit: unsupported stdlib ref `%s/%s` (no core on Chez yet)" (get node :ns) (get node :name))
+             (string "(var-deref " (string/format "%j" (get node :ns)) " "
+                     (string/format "%j" (get node :name)) ")"))
+    :host  (errorf "emit: unsupported host ref `%s` (no host interop on Chez yet)" (get node :name))
     :if    (string "(if (jolt-truthy? " (emit (get node :test)) ") "
                    (emit (get node :then)) " " (emit (get node :else)) ")")
     :do    (string "(begin "
-                   (string/join (map emit (get node :statements)) " ")
-                   (if (empty? (get node :statements)) "" " ")
+                   (string/join (map emit (vv (get node :statements))) " ")
+                   (if (empty? (vv (get node :statements))) "" " ")
                    (emit (get node :ret)) ")")
-    :invoke (string "(" (emit (get node :fn)) " "
-                    (string/join (map emit (get node :args)) " ") ")")
+    :invoke (emit-invoke node)
     :let   (emit-let node)
     :loop  (emit-loop node)
     :recur (emit-recur node)
     :fn    (emit-fn node)
-    :def   (string "(define " (munge (get node :name)) " " (emit (get node :init)) ")")
+    :def   (string "(def-var! " (string/format "%j" (get node :ns)) " "
+                   (string/format "%j" (get node :name)) " " (emit (get node :init)) ")")
     (errorf "emit: unhandled op %p" (get node :op)))))
 
-# Wrap emitted top-level forms into a runnable Chez program: preamble (value
-# model + rt shims) then the forms, then print `final` (a Scheme expr string).
+# Wrap emitted top-level forms into a runnable Chez program: load the RT, then
+# the def forms, then print `final` (an emitted Scheme expr string) via jolt's
+# number/value printing.
 (defn program [forms-scheme final]
   (string
     "(import (chezscheme))\n"
-    "(load \"host/chez/values.ss\")\n"
-    "(define (jolt-inc x) (+ x 1))\n"
-    "(define (jolt-dec x) (- x 1))\n"
+    "(load \"host/chez/rt.ss\")\n"
     (string/join forms-scheme "\n") "\n"
-    "(printf \"~a\\n\" " final ")\n"))
+    "(printf \"~a\\n\" (jolt-pr-str " final "))\n"))

host/chez/rt.ss

@@ -0,0 +1,52 @@
+;; Phase 1 (jolt-cf1q.2) — the minimal Chez RT the emitted Scheme rests on.
+;;
+;; Sits above the value model (values.ss) and below an emitted program. Adds the
+;; two things the back end's output references that aren't in the value layer:
+;;   1. the var-cell late-binding registry (Clojure vars — a global root that a
+;;      reference reads at call time, so redefinition / mutual recursion work);
+;;   2. the rt primitive shims the emitter names (jolt-inc/dec/not) and jolt's
+;;      number printing (all jolt numbers model Clojure doubles; integer-valued
+;;      print without a trailing ".0", matching the Janet host).
+;;
+;; Emitted programs do `(load "host/chez/rt.ss")`; this loads values.ss in turn.
+
+(load "host/chez/values.ss")
+
+;; --- rt arithmetic / logic shims (named in emit.janet's native-ops) ----------
+(define (jolt-inc x) (+ x 1))
+(define (jolt-dec x) (- x 1))
+;; jolt `not`: only nil and false are falsey.
+(define (jolt-not x) (if (jolt-truthy? x) #f #t))
+
+;; --- var cells: late-bound global roots (Clojure vars) -----------------------
+;; A var is a mutable cell keyed by "ns/name". A `:def` sets the root; a `:var`
+;; reference reads it at use time (late binding), so a forward/mutually-recursive
+;; reference resolves to whatever the cell holds when the call actually runs.
+(define-record-type var-cell (fields ns name (mutable root)) (nongenerative var-cell-v1))
+(define var-table (make-hashtable string-hash string=?))
+(define (jolt-var ns name)
+  (let ((k (string-append ns "/" name)))
+    (or (hashtable-ref var-table k #f)
+        (let ((c (make-var-cell ns name jolt-nil)))
+          (hashtable-set! var-table k c)
+          c))))
+(define (var-deref ns name) (var-cell-root (jolt-var ns name)))
+(define (def-var! ns name v) (var-cell-root-set! (jolt-var ns name) v) v)
+
+;; --- jolt number printing ----------------------------------------------------
+;; jolt models every number as a Clojure double: integer-valued values print
+;; without a ".0" (the Janet host prints (* 1.0 5) as "5", (/ 1 2) as "0.5").
+(define (jolt-num->string x)
+  (if (and (rational? x) (integer? x))
+      (number->string (exact x))
+      (number->string x)))
+
+;; Minimal pr-str for the program's final value (full printer is Phase 2).
+(define (jolt-pr-str x)
+  (cond
+    ((jolt-nil? x) "nil")
+    ((eq? x #t) "true")
+    ((eq? x #f) "false")
+    ((number? x) (jolt-num->string x))
+    ((string? x) x)
+    (else (format "~a" x))))