build+test: -j/ccache build, drop 990 dup compile, add make smoke (#19)

Test-speed "immediate wins" from task #19 (build/test-infra only, no
compiler/cgen change — byte-id-neutral; all 237 pass, 950/990-997 +
combined_ww_fresh unchanged).

#1 Parallel build + ccache. MAKEFLAGS += -j$(NPROC) by default: the
C-compile DAG and the five wwstage builds write disjoint outputs (each
.o distinct; each wwstage tool's side files land at its own
selfhost/cmd/<tool>/main.* stem), so -j is order-independent.
test/run's Phase-2 byte-id gates are a single serial recipe that -j
does not reach. CC is wrapped with ccache when present (content-
addressed, byte-identical to plain cc); falls back to bare $(CC).

#2 Kill 990's duplicate ww1->ww2 compile. probe_ww1_to_ww2 recompiled
main.combined.ww (~70s) to assert the ww2 binary is executable — but
probe_bootstrap_fixed_point already compiles ww1->ww2, assembles,
links, and *runs* ww2 to produce ww3, so executability is proven and
the byte-id assertions (ww2.s==ww3.s, ww2==ww3) are untouched. Drop
the redundant probe. make test ~8:30 -> 7:39.

Add `make smoke [FIXTURE=x.ww]`: inner-loop cross-stage byte-id check
(cstage w6c vs wwstage w6c_ww .s diff) on a small self-contained
fixture, seconds. Catches cs!=ww emission divergence per fold; NOT a
substitute for the full 990-997 gate before landing a cgen/ABI fold.

Makefile

@@ -7,6 +7,21 @@
 PREFIX	?= /usr/local
 CC	?= cc
 AR	?= ar
+
+# Parallelise the build by default. The C-compile DAG and the five
+# wwstage builds write disjoint outputs (each .o is distinct; each
+# wwstage tool's side files land at its own selfhost/cmd/<tool>/main.*
+# stem), so -j is order-independent. test/run's Phase-2 byte-id gates
+# (950/990-997) are a single serial recipe — MAKEFLAGS -j does not
+# reach inside them. Override with `make -j1` if needed.
+NPROC	?= $(shell nproc 2>/dev/null || echo 4)
+MAKEFLAGS += -j$(NPROC)
+
+# Wrap the C compiler with ccache when present — content-addressed, so
+# byte-identical to plain cc; just skips recompiling unchanged TUs.
+# Falls back cleanly to bare $(CC) when ccache is absent. A command-line
+# CC= override wins and is left unwrapped.
+CC	:= $(shell command -v ccache >/dev/null 2>&1 && echo "ccache $(CC)" || echo "$(CC)")
 CFLAGS	?= -O0 -g -Wall -Wextra -Wpedantic -Wno-unused-parameter
 CFLAGS	+= -std=c99 -fno-strict-aliasing -D_POSIX_C_SOURCE=200809L
 INCS	= -Icmd/wcc
@@ -1675,6 +1690,28 @@ test: all sizelint $(TESTS)
 test-unit: all sizelint $(TESTS)
 	@WW=$(BIN)/ww BIN=$(BIN) UNIT=1 sh test/run
 
+# ---- smoke: inner-loop cross-stage byte-id check on a SMALL input ------
+# Compiles one self-contained fixture through cstage w6c AND wwstage
+# w6c_ww and diffs the emitted .s. Catches a cs!=ww emission divergence
+# (rule 10) in seconds — the per-fold inner-loop check Rob's cadence
+# calls for. NOT a substitute for the full 990-997 byte-id gate, which
+# self-compiles main.combined.ww: ALWAYS run `make test` before landing
+# a cgen/ABI/compiler-lib fold. w6c consumes a flat unit (no import
+# resolution), so FIXTURE must be import-free or a pre-built .combined.ww.
+#   make smoke                          # default: rt/malloc.ww
+#   make smoke FIXTURE=rt/ensure.ww     # any self-contained .ww
+FIXTURE	?= rt/malloc.ww
+smoke: $(BIN)/w6c $(BIN)/w6c_ww
+	@$(BIN)/w6c $(FIXTURE) > $(OUT)/smoke.cs.s || { echo "smoke: cstage w6c failed on $(FIXTURE)"; exit 1; }
+	@$(BIN)/w6c_ww $(FIXTURE) > $(OUT)/smoke.ww.s || { echo "smoke: wwstage w6c_ww failed on $(FIXTURE)"; exit 1; }
+	@if cmp -s $(OUT)/smoke.cs.s $(OUT)/smoke.ww.s; then \
+		echo "smoke OK: cs == ww .s for $(FIXTURE) (`wc -l < $(OUT)/smoke.cs.s` lines)"; \
+	else \
+		echo "smoke FAIL: cs != ww .s for $(FIXTURE)"; \
+		diff -u $(OUT)/smoke.cs.s $(OUT)/smoke.ww.s | head -40; \
+		exit 1; \
+	fi
+
 install: all
 	mkdir -p $(PREFIX)/bin $(PREFIX)/lib
 	cp $(BIN)/ww $(PREFIX)/bin/
@@ -1820,4 +1857,4 @@ nocc:
 	@echo
 	@echo "NOCC OK: $(STAGE0)/* reproduces itself from source. cc not invoked."
 
-.PHONY: all cstage wwstage test test-unit sizelint install clean bootstrap nocc bootstrap-snapshot
+.PHONY: all cstage wwstage test test-unit smoke sizelint install clean bootstrap nocc bootstrap-snapshot

test/wcc/990_selfhost.c

@@ -709,60 +709,6 @@ cleanup:
 	return fail ? -1 : 0;
 }
 
-/* Probe 8: ww1 → ww2 ladder. wwdump_ww (= ww1, built by C tools)
- * compiles its own concatenated source. The output assembles + links
- * via w6a/w6l into ww2 (a fresh ww-built compiler binary). Confirms
- * the full ww-cgen pipeline is functionally correct end-to-end on
- * the entire frontend (lex+parse+check+cgen+ast+typ+sym+mem+err+
- * driver). Pre-fix-set this segfaulted at 5919 lines of asm; post-
- * fix-set it produces 43K+ lines that assemble/link cleanly. */
-static int
-probe_ww1_to_ww2(const char *bin)
-{
-	char cwd[1024];
-	if (getcwd(cwd, sizeof cwd) == NULL) return -1;
-	char tmpdir[64];
-	snprintf(tmpdir, sizeof tmpdir, "/tmp/ww12_%d", getpid());
-	mkdir(tmpdir, 0755);
-	char asm_p[512], obj_p[512], exe_p[512], cmd[4096];
-	snprintf(asm_p, sizeof asm_p, "%s/ww2.s", tmpdir);
-	snprintf(obj_p, sizeof obj_p, "%s/ww2.o", tmpdir);
-	snprintf(exe_p, sizeof exe_p, "%s/ww2",   tmpdir);
-	int fail = 0;
-	snprintf(cmd, sizeof cmd,
-	    "timeout 180 %s/wwdump_ww -c %s/selfhost/cmd/wwdump/main.combined.ww > %s 2>/dev/null",
-	    bin, cwd, asm_p);
-	if (runwait(cmd) != 0) {
-		fprintf(stderr, "ww1->ww2 FAIL: wwdump_ww segfaulted on its own source\n");
-		fail++;
-		goto cleanup;
-	}
-	snprintf(cmd, sizeof cmd, "timeout 180 %s/w6a -o %s %s 2>/dev/null", bin, obj_p, asm_p);
-	if (runwait(cmd) != 0) {
-		fprintf(stderr, "ww1->ww2 FAIL: w6a errored on ww1 output\n");
-		fail++;
-		goto cleanup;
-	}
-	char rt[1024];
-	snprintf(rt, sizeof rt, "%s/../lib/libwwrt.a", bin);
-	snprintf(cmd, sizeof cmd, "timeout 180 %s/w6l -o %s %s %s 2>/dev/null",
-	    bin, exe_p, obj_p, rt);
-	if (runwait(cmd) != 0) {
-		fprintf(stderr, "ww1->ww2 FAIL: w6l errored\n");
-		fail++;
-		goto cleanup;
-	}
-	struct stat st;
-	if (stat(exe_p, &st) != 0 || !(st.st_mode & 0111)) {
-		fprintf(stderr, "ww1->ww2 FAIL: ww2 binary not executable\n");
-		fail++;
-	}
-cleanup:
-	snprintf(cmd, sizeof cmd, "rm -rf %s", tmpdir);
-	runwait(cmd);
-	return fail ? -1 : 0;
-}
-
 /* Probe 7: end-to-end via wwdump_ww. Use `ww build` to produce the
  * concatenated source as a side effect, then drive ww-cgen → w6a →
  * w6l → run, comparing the binary's exit code to the fixture's
@@ -930,7 +876,6 @@ main(void)
 	if (probe_ww_compile(bin)  != 0) fail++;
 	if (probe_cgen_match(bin)  != 0) fail++;
 	if (probe_ww_links(bin)    != 0) fail++;
-	if (probe_ww1_to_ww2(bin)  != 0) fail++;
 	if (probe_bootstrap_fixed_point(bin) != 0) fail++;
 
 	if (fail) {