Allow exporting of symbols that are not valid JS identifiers (#25922)

In this case we now generate a warning instead of an errors.

Such symbols are not directly accessible in the module scope (since we
cannot declare them there). They are only accessible via `wasmExports`
or `Module` dictionary objects.

See: llvm/llvm-project#169043
Fixes: #24825, #23560

Author: sbc100

test/test_other.py

@@ -10237,13 +10237,13 @@ def test_dash_s_list_parsing(self):
       # Simple one-per-line response file format
       ("[email protected]", ''),
       # stray slash
-      ("EXPORTED_FUNCTIONS=['_a', '_b', \\'_c', '_d']", '''invalid export name: "\\\\'_c'"'''),
+      ("EXPORTED_FUNCTIONS=['_a', '_b', \\'_c', '_d']", r'''emcc: error: undefined exported symbol: "\\'_c'"'''),
       # stray slash
-      ("EXPORTED_FUNCTIONS=['_a', '_b',\\ '_c', '_d']", '''invalid export name: "\\\\ '_c'"'''),
+      ("EXPORTED_FUNCTIONS=['_a', '_b',\\ '_c', '_d']", r'''emcc: error: undefined exported symbol: "\\ '_c'"'''),
       # stray slash
-      ('EXPORTED_FUNCTIONS=["_a", "_b", \\"_c", "_d"]', 'invalid export name: "\\\\"_c""'),
+      ('EXPORTED_FUNCTIONS=["_a", "_b", \\"_c", "_d"]', r'emcc: error: undefined exported symbol: "\\"_c""'),
       # stray slash
-      ('EXPORTED_FUNCTIONS=["_a", "_b",\\ "_c", "_d"]', 'invalid export name: "\\\\ "_c"'),
+      ('EXPORTED_FUNCTIONS=["_a", "_b",\\ "_c", "_d"]', r'emcc: error: undefined exported symbol: "\\ "_c""'),
       # missing comma
       ('EXPORTED_FUNCTIONS=["_a", "_b" "_c", "_d"]', 'wasm-ld: error: symbol exported via --export not found: b" "_c'),
     ]:
@@ -14852,21 +14852,45 @@ def test_cxx20_modules_std_headers(self):
     self.do_runf('main.cpp', 'Hello Module!', cflags=['-std=c++20', '-fmodules'])
 
   def test_invalid_export_name(self):
-    create_file('test.c', '__attribute__((export_name("my.func"))) void myfunc() {}')
-    expected = 'emcc: error: invalid export name: "_my.func"'
-    self.assert_fail([EMCC, 'test.c'], expected)
+    create_file('main.c', r'''
+      #include <emscripten.h>
+      #include <stdio.h>
+
+      __attribute__((export_name("my.func"))) int myfunc() { return 42; }
+
+      int main() {
+        int rtn = EM_ASM_INT(return Module['_my.func']());
+        printf("got: %d\n", rtn);
+
+        int rtn2 = EM_ASM_INT(return wasmExports['my.func']());
+        printf("got2: %d\n", rtn2);
+        return 0;
+      }
+    ''')
+    expected = 'emcc: error: export name is not a valid JS symbol: "_my.func".  Use `Module` or `wasmExports` to access this symbol'
+    self.assert_fail([EMCC, '-Werror', 'main.c'], expected)
+
+    # With warning suppressed the above program should work.
+    self.do_runf('main.c', 'got: 42\ngot2: 42\n', cflags=['-Wno-js-compiler'])
 
     # When we are generating only wasm and not JS we don't need exports to
     # be valid JS symbols.
-    self.run_process([EMCC, 'test.c', '--no-entry', '-o', 'out.wasm'])
-
-    # GCC (and clang) and JavaScript also allow $ in symbol names
-    create_file('valid.c', '''
-                #include <emscripten.h>
-                EMSCRIPTEN_KEEPALIVE
-                void my$func() {}
-                ''')
-    self.run_process([EMCC, 'valid.c'])
+    self.run_process([EMCC, '-Werror', 'main.c', '--no-entry', '-o', 'out.wasm'])
+
+  def test_export_with_dollarsign(self):
+    # GCC (and clang) and JavaScript both allow $ in symbol names
+    create_file('main.c', r'''
+      #include <emscripten.h>
+      #include <stdio.h>
+
+      EMSCRIPTEN_KEEPALIVE int my$func() { return 42; }
+
+      int main() {
+        int rtn = EM_ASM_INT(return _my$func());
+        printf("got: %d\n", rtn);
+        return 0;
+      }''')
+    self.do_runf('main.c', 'got: 42\n')
 
   @also_with_modularize
   def test_instantiate_wasm(self):

tools/emscripten.py

@@ -419,7 +419,7 @@ def emscript(in_wasm, out_wasm, outfile_js, js_syms, finalize=True, base_metadat
 
   for e in settings.EXPORTED_FUNCTIONS:
     if not js_manipulation.isidentifier(e):
-      exit_with_error(f'invalid export name: "{e}"')
+      diagnostics.warning('js-compiler', f'export name is not a valid JS symbol: "{e}".  Use `Module` or `wasmExports` to access this symbol')
 
   # memory and global initializers
 
@@ -1019,10 +1019,16 @@ def create_receiving(function_exports, other_exports, library_symbols, aliases):
     # folks try to call/use a reference that was taken before the
     # wasm module is available.
     for sym in mangled:
-      assignment = sym
-      if (settings.MODULARIZE or not settings.MINIMAL_RUNTIME) and should_export(sym) and settings.MODULARIZE != 'instance':
-        assignment += f" = Module['{sym}']"
-      receiving.append(f"var {assignment} = makeInvalidEarlyAccess('{sym}');")
+      module_export = (settings.MODULARIZE or not settings.MINIMAL_RUNTIME) and should_export(sym) and settings.MODULARIZE != 'instance'
+      if not js_manipulation.isidentifier(sym) and not module_export:
+        continue
+      assignment = f'var {sym}'
+      if module_export:
+        if js_manipulation.isidentifier(sym):
+          assignment += f" = Module['{sym}']"
+        else:
+          assignment = f"Module['{sym}']"
+      receiving.append(f"{assignment} = makeInvalidEarlyAccess('{sym}');")
   else:
     # Declare all exports in a single var statement
     sep = ',\n  '
@@ -1057,7 +1063,15 @@ def create_receiving(function_exports, other_exports, library_symbols, aliases):
       sig_str = sym.removeprefix('dynCall_')
       assignment += f" = dynCalls['{sig_str}']"
     if do_module_exports and should_export(mangled):
-      assignment += f" = Module['{mangled}']"
+      if js_manipulation.isidentifier(mangled):
+        assignment += f" = Module['{mangled}']"
+      else:
+        assignment = f"Module['{mangled}']"
+    elif not js_manipulation.isidentifier(mangled):
+      # Symbol is not a valid JS identify and also not exported. In this case we
+      # have nothing to do here.
+      continue
+
     if sym in alias_inverse_map:
       for target in alias_inverse_map[sym]:
         assignment += f" = {target}"