Compiler: accept unnumbered programs, fix string concat in PRINT

Three fixes that lift seven test programs from skipped to passing,
bringing the AOT compiler harness from 56/56 to 63/63.

- Unnumbered programs (compiler_main.c): src/compiler_main.c skipped
  any line that didn't start with a digit, so direct-mode .bas files
  like hello.bas, math_ops.bas, string_ops.bas (no line numbers)
  failed with "No program lines found".  load_file now auto-assigns
  line numbers (last_num + 10) to unnumbered lines, with overflow
  protection at line 65520.

- String concatenation in PRINT (codegen.c): emit_str_atom had a
  broken concat loop that emitted "; _cat = gw_str_concat(&...
  _cat.sval ...)" — _cat was never declared, so any program with a
  string-literal concat in PRINT (like PRINT "ABC" + "DEF") failed
  to link.  Concat is properly handled by emit_str_expr's outer
  loop; remove the dead/broken code in the atom.  Fixes
  string_ops.bas.

- Transcendental result type (codegen.c): peek_expr_type returned
  VT_DBL for ATN/LOG/EXP/VAL, so PRINT formatted them with 15-digit
  double precision (e.g. 3.141592653589793) while real GW-BASIC and
  the interpreter format the single-precision result as 3.141593.
  Real GW-BASIC's transcendentals are single-precision; only CDBL
  forces double.  Demote ATN/LOG/EXP/VAL to VT_SNG; CDBL stays
  VT_DBL.  Fixes math_ops.bas.

Also: tests/run_compiler_tests.sh now runs the compiled binary from
the project root rather than the tempdir where it was built, so
test programs that reference tests/programs/ via relative paths
(chain_test, common_test, run_file, misc_stmts) resolve their
targets.  Earlier I'd misdiagnosed those failures as ON ERROR
divergence — they were just CWD-dependent path lookups.

Doc/test counts: 56 → 63 in README, docs/index.md, docs/development.md,
docs/roadmap.md.  Roadmap updated to note the compiler now accepts
unnumbered programs.

README.md

@@ -55,7 +55,7 @@ to produce native executables linked against `libgwrt.a`.
 
 Pipeline: `.bas` → tokenizer → analysis → C codegen → `gcc` → native binary.
 
-56 of 56 eligible test programs compile via `gwbasic-compile` and produce
+63 of 63 eligible test programs compile via `gwbasic-compile` and produce
 output matching the interpreter's golden files. Run `bash
 tests/run_compiler_tests.sh` to verify.
 
@@ -111,7 +111,7 @@ jupyter notebook  # select "GW-BASIC 2026" kernel
 
 ## Tests
 
-72 interpreter tests, 14 kernel tests, 56 compiler tests -- all passing.
+72 interpreter tests, 14 kernel tests, 63 compiler tests -- all passing.
 
 ```bash
 bash tests/run_tests.sh              # interpreter

docs/development.md

@@ -26,8 +26,8 @@
 
 72 automated test programs in `tests/programs/`, plus 4 classic interactive
 programs in `tests/classic/` (Hamurabi, Lunar Lander, Gunner, Diamond from
-David Ahl's *BASIC Computer Games*).  14 Jupyter kernel tests.  56 compiler
-tests (eligible numbered programs compiled to native executables via
+David Ahl's *BASIC Computer Games*).  14 Jupyter kernel tests.  63 compiler
+tests (eligible programs compiled to native executables via
 `gwbasic-compile`; run `bash tests/run_compiler_tests.sh`).
 
 Run the full automated suite:

docs/index.md

@@ -28,7 +28,7 @@ version is structured as modular C suitable for new feature development.
 - **Classic programs** -- Hamurabi, Lunar Lander, Gunner, and Diamond from
   David Ahl's *BASIC Computer Games* (1978) run out of the box
 - **Ahead-of-time compiler** -- `gwbasic-compile prog.bas -c` produces native
-  executables via C codegen + GCC (56/56 eligible tests pass, 100%)
+  executables via C codegen + GCC (63/63 eligible tests pass, 100%)
 - **Jupyter kernel** -- inline Sixel graphics, INPUT support, Pygments syntax
   highlighting; `pip install -e . && gwbasickernel-install --user`
 - **72 test programs** with golden-file regression testing and DOSBox-X

docs/roadmap.md

@@ -9,10 +9,10 @@ invokes GCC to produce native executables linked against `libgwrt.a`.
 
 Pipeline: `.bas` → `gw_crunch()` → analysis pass → C codegen → `gcc` → native binary.
 
-**56 of 56 eligible tests pass (100%)** via `tests/run_compiler_tests.sh`.
-The harness skips chain/common multi-file flows, hardware-dependent tests
-(graphics/sound/timer), and unnumbered direct-mode programs that the
-compiler does not currently accept.
+**63 of 63 eligible tests pass (100%)** via `tests/run_compiler_tests.sh`.
+The harness only skips hardware-dependent tests (graphics/sound/timer)
+and CHAIN/RUN target files that aren't standalone.  The compiler now
+accepts unnumbered direct-mode programs by auto-numbering them.
 
 Language coverage:
 

src/codegen.c

@@ -962,7 +962,9 @@ static void emit_str_atom(void)
     skip_spaces();
     uint8_t tok = cur();
 
-    /* String literal */
+    /* String literal.  Concatenation is handled by emit_str_expr (the atom
+     * just emits the literal value); a previous attempt to handle '+' at
+     * the atom level emitted references to an undeclared `_cat` variable. */
     if (tok == '"') {
         EMIT("gw_str_from_cstr(\"");
         tp++;
@@ -973,14 +975,6 @@ static void emit_str_atom(void)
         }
         if (*tp == '"') tp++;
         EMIT("\")");
-        /* Handle concatenation */
-        skip_spaces();
-        while (cur() == TOK_PLUS) {
-            advance();
-            EMIT("; _cat = gw_str_concat(&(gw_value_t){.type=VT_STR,.sval=_cat.sval}, &(gw_value_t){.type=VT_STR,.sval=");
-            emit_str_expr();
-            EMIT("})");
-        }
         return;
     }
 
@@ -1305,9 +1299,13 @@ static gw_valtype_t peek_expr_type(void)
             tp = save;
             return arg_type;
         }
-        case FUNC_VAL: case FUNC_ATN: case FUNC_LOG: case FUNC_EXP:
+        /* GW-BASIC's transcendentals are single-precision; only CDBL
+         * explicitly forces double.  Promoting them to VT_DBL here makes
+         * PRINT use 15-digit format instead of the 7-digit single form. */
         case FUNC_CDBL:
             return VT_DBL;
+        case FUNC_VAL: case FUNC_ATN: case FUNC_LOG: case FUNC_EXP:
+            return VT_SNG;
         default:
             return VT_SNG;
         }

src/compiler_main.c

@@ -92,7 +92,10 @@ static int parse_line_num(const char *text, uint16_t *num)
     return (int)(p - text);
 }
 
-/* Load a .bas file (ASCII format) */
+/* Load a .bas file (ASCII format).  Lines that don't begin with a number
+ * are treated as direct-mode statements and given auto-assigned numbers
+ * (10, 20, ...) so the compiler can handle scratchpad-style programs the
+ * interpreter accepts via stdin or `gwbasic file.bas`. */
 static int load_file(const char *filename)
 {
     FILE *f = fopen(filename, "r");
@@ -103,6 +106,7 @@ static int load_file(const char *filename)
 
     char buf[256];
     uint8_t kbuf[256];
+    uint16_t last_num = 0;
     while (fgets(buf, sizeof(buf), f)) {
         int len = strlen(buf);
         while (len > 0 && (buf[len-1] == '\n' || buf[len-1] == '\r'))
@@ -111,11 +115,23 @@ static int load_file(const char *filename)
 
         uint16_t num;
         int skip = parse_line_num(buf, &num);
-        if (skip == 0) continue;  /* skip non-numbered lines */
+        const char *content;
+        if (skip == 0) {
+            /* Unnumbered direct-mode line: auto-assign next number. */
+            if (last_num >= 65520) {
+                fprintf(stderr, "Auto line numbering overflow in %s\n", filename);
+                fclose(f);
+                return 1;
+            }
+            num = last_num + 10;
+            content = buf;
+            while (*content == ' ') content++;
+        } else {
+            content = buf + skip;
+            while (*content == ' ') content++;
+        }
+        last_num = num;
 
-        /* Tokenize the line content (after the line number) */
-        const char *content = buf + skip;
-        while (*content == ' ') content++;
         int tok_len = gw_crunch(content, kbuf, sizeof(kbuf));
         store_line(num, kbuf, tok_len);
     }

tests/run_compiler_tests.sh

@@ -20,20 +20,13 @@ if [ ! -f "$PROJECT_DIR/build/libgwrt.a" ]; then
     exit 1
 fi
 
-# Programs that are not meaningful for the AOT path:
-#   - chain/common targets are not standalone
-#   - interactive / timing-dependent / hardware tests
-#   - unnumbered direct-mode programs (compiler requires numbered lines)
-#   - misc_stmts.bas / run_file.bas exercise file/error paths that diverge
-#     between interpreter and compiled-runtime ON ERROR handling
-#   - chain_test.bas / common_test.bas need their target .bas in the same
-#     directory as the compiled binary; this harness compiles in a tmpdir
-#     and doesn't stage the targets
+# Programs not meaningful for the AOT path:
+#   - chain_target.bas / common_target.bas are not standalone (loaded via
+#     CHAIN, never run directly)
+#   - interactive / timing-dependent / hardware-output programs
 SKIP=(
     chain_target.bas
-    chain_test.bas
     common_target.bas
-    common_test.bas
     datetime.bas
     on_timer.bas
     timer_stop.bas
@@ -43,11 +36,6 @@ SKIP=(
     play_scale.bas
     speaker_out.bas
     text_adventure.bas
-    hello.bas
-    math_ops.bas
-    string_ops.bas
-    misc_stmts.bas
-    run_file.bas
 )
 should_skip() {
     local n="$1"
@@ -81,16 +69,17 @@ for bas in "$SCRIPT_DIR"/programs/*.bas; do
         popd > /dev/null
         continue
     fi
+    popd > /dev/null
 
+    # Run from project root so test programs that reference tests/programs/
+    # (chain_test, common_test, run_file, misc_stmts) resolve relative paths.
     actual=$(mktemp)
-    if ! timeout 5 "./$stem" > "$actual" 2>&1; then
+    if ! ( cd "$PROJECT_DIR" && timeout 5 "$WORK_DIR/$stem" > "$actual" 2>&1 ); then
         printf "  RUN-FAIL  %s\n" "$name"
         fail=$((fail + 1))
         rm -f "$actual"
-        popd > /dev/null
         continue
     fi
-    popd > /dev/null
 
     expected="$EXPECTED_DIR/${stem}.expected"
     if [ -f "$expected" ]; then