fix: opt predict_version into SQLITE_RESULT_SUBTYPE; harden CI subtype check

The Windows (MinGW) CI job surfaced a real latent bug, not a platform
quirk: predict_version() calls sqlite3_result_subtype() to tag its JSON
result, but SQLite 3.45+ refuses that call unless the function was
registered with SQLITE_RESULT_SUBTYPE. macOS shipped an older SQLite that
never enforced it, so local tests passed; MSYS2's newer sqlite3 CLI
returned 'misuse of sqlite3_result_subtype()'.

- Register predict_version with SQLITE_RESULT_SUBTYPE (guarded #ifdef so
  a header bump either direction stays safe; scoped to the one function
  that sets a subtype).
- Compile the static test builds (soak, asan, wasm, fuzz) with
  -DSQLITE_STRICT_SUBTYPE=1 and exercise predict_version()/predict_debug()
  in the soak, so this bug class now fails under our own ASan/valgrind/
  wasm runs instead of only on Windows CI.
- wasm job: install emscripten directly from emscripten-core/emsdk
  instead of an unresolvable third-party action.

Co-Authored-By: Claude Fable 5 <noreply@anthropic.com>
Signed-off-by: mstrathman <matthew.strathman@gmail.com>

Author: mstrathman

.github/workflows/ci.yml

@@ -61,7 +61,8 @@ jobs:
         run: |
           make vendor/sqlite3ext.h sqlite-predict.h
           clang -std=c99 -g -O1 -fsanitize=fuzzer,address \
-            -DSQLITE_CORE -DSQLITE_PREDICT_STATIC -Ivendor/ -I./ \
+            -DSQLITE_CORE -DSQLITE_PREDICT_STATIC -DSQLITE_STRICT_SUBTYPE=1 \
+            -Ivendor/ -I./ \
             fuzz/fuzz_predict.c sqlite-predict.c predict-forecast.c \
             predict-tabular.c predict-receipts.c vendor/sha256.c \
             vendor/sqlite3.c -o fuzz_predict -lm -lpthread -ldl
@@ -94,8 +95,14 @@ jobs:
     runs-on: ubuntu-latest
     steps:
       - uses: actions/checkout@v4
-      - uses: mymindstorm/setup-emsdk@v14
+      - name: Install emscripten
+        run: |
+          git clone --depth 1 https://github.com/emscripten-core/emsdk.git
+          ./emsdk/emsdk install latest
+          ./emsdk/emsdk activate latest
       - name: Compile soak to WebAssembly
-        run: make soak-wasm CC=emcc
+        run: |
+          source ./emsdk/emsdk_env.sh
+          make soak-wasm
       - name: Run under node
         run: node dist/soak.js

Makefile

@@ -73,15 +73,17 @@ test-asan: vendor/sqlite3ext.h sqlite-predict.h
 	mkdir -p $(prefix)
 	clang -std=c99 -g -O1 -fsanitize=address,undefined \
 	  -fno-omit-frame-pointer -fno-sanitize-recover=undefined \
-	  -DSQLITE_CORE -DSQLITE_PREDICT_STATIC -Ivendor/ -I./ \
+	  -DSQLITE_CORE -DSQLITE_PREDICT_STATIC -DSQLITE_STRICT_SUBTYPE=1 \
+	  -Ivendor/ -I./ \
 	  tests/soak.c $(OBJS) vendor/sqlite3.c -o $(prefix)/soak-asan
 	UBSAN_OPTIONS=print_stacktrace=1 ./$(prefix)/soak-asan
 
 # libFuzzer harness (statically links sqlite3.c; SQLITE_CORE build)
 fuzz-build: vendor/sqlite3ext.h sqlite-predict.h
 	mkdir -p $(prefix)
 	clang -std=c99 -g -O1 -fsanitize=fuzzer,address,undefined \
-	  -DSQLITE_CORE -DSQLITE_PREDICT_STATIC -Ivendor/ -I./ \
+	  -DSQLITE_CORE -DSQLITE_PREDICT_STATIC -DSQLITE_STRICT_SUBTYPE=1 \
+	  -Ivendor/ -I./ \
 	  fuzz/fuzz_predict.c $(OBJS) vendor/sqlite3.c \
 	  -o $(prefix)/fuzz_predict
 
@@ -115,12 +117,14 @@ test-valgrind: vendor/sqlite3ext.h sqlite-predict.h
 # standalone soak binary (used by CI valgrind + Windows, runnable directly)
 soak: $(prefix) vendor/sqlite3ext.h sqlite-predict.h
 	$(CC) -std=c99 -g -O0 -Ivendor/ -I./ -DSQLITE_CORE -DSQLITE_PREDICT_STATIC \
+	  -DSQLITE_STRICT_SUBTYPE=1 \
 	  tests/soak.c $(OBJS) vendor/sqlite3.c -o $(prefix)/soak $(LDFLAGS)
 
 # WebAssembly soak (emscripten): compiles the static soak to wasm. CI runs
 # it under node to prove the code is wasm-portable and correct there.
 soak-wasm: $(prefix) vendor/sqlite3ext.h sqlite-predict.h
 	emcc -std=c99 -O1 -Ivendor/ -I./ -DSQLITE_CORE -DSQLITE_PREDICT_STATIC \
+	  -DSQLITE_STRICT_SUBTYPE=1 \
 	  -sALLOW_MEMORY_GROWTH -sEXIT_RUNTIME=1 -sSTACK_SIZE=1048576 \
 	  tests/soak.c $(OBJS) vendor/sqlite3.c -o $(prefix)/soak.js
 

sqlite-predict.c

@@ -338,9 +338,18 @@ __declspec(dllexport)
 #endif
   int rc = SQLITE_OK;
   const int flags = SQLITE_UTF8 | SQLITE_INNOCUOUS | SQLITE_DETERMINISTIC;
+  /* predict_version tags its result with the JSON subtype. SQLite 3.45+
+   * refuses sqlite3_result_subtype() unless the function opted in with
+   * this flag (older SQLite ignores the unknown flag and never enforced
+   * the check). Scoped to the one function that sets a subtype. */
+#ifdef SQLITE_RESULT_SUBTYPE
+  const int subtype_flags = flags | SQLITE_RESULT_SUBTYPE;
+#else
+  const int subtype_flags = flags;
+#endif
 
-  rc = sqlite3_create_function_v2(db, "predict_version", 0, flags, NULL,
-                                  predict_version_fn, NULL, NULL, NULL);
+  rc = sqlite3_create_function_v2(db, "predict_version", 0, subtype_flags,
+                                  NULL, predict_version_fn, NULL, NULL, NULL);
   if (rc != SQLITE_OK)
     return rc;
   rc = sqlite3_create_function_v2(db, "predict_debug", 0, flags, NULL,

tests/soak.c

@@ -32,6 +32,12 @@ int main(void) {
   if (sqlite3_predict_init(db, NULL, NULL) != SQLITE_OK)
     return fail("init", db);
 
+  /* meta functions — predict_version() sets the JSON result subtype, so
+   * this exercises that path under -DSQLITE_STRICT_SUBTYPE (see Makefile) */
+  if (run_discard(db, "SELECT predict_version()", 1) ||
+      run_discard(db, "SELECT predict_debug()", 1))
+    goto done_fail;
+
   if (run_discard(
           db,
           "CREATE TABLE series(ts TEXT, value REAL, grp TEXT);", 1) ||