Simplify clang vs clang++ selection process when compiling sources

ChangeLog.md

@@ -20,6 +20,13 @@ See docs/process.md for more on how version tagging works.
 
 3.1.50 (in development)
 -----------------------
+- When compiling, Emscripten will now invoke `clang` or `clang++` depending only
+  on whether `emcc` or `em++` was run.  Previously it would determine which to
+  run based on individual file extensions.  One side effect of this is that you
+  may now see a clang warning when building `.c` source files using `em++`:
+  `warning: treating 'c' input as 'c++' when in C++ mode`.  This also means that
+  the `DEFAULT_TO_CXX` setting now only applies when linking and not when
+  compiling. (#20712)
 
 3.1.49 - 11/14/23
 -----------------

emcc.py

@@ -1494,9 +1494,6 @@ def phase_setup(options, state, newargs):
     diagnostics.warning('deprecated', 'RUNTIME_LINKED_LIBS is deprecated; you can simply list the libraries directly on the commandline now')
     newargs += settings.RUNTIME_LINKED_LIBS
 
-  if settings.STRICT:
-    default_setting('DEFAULT_TO_CXX', 0)
-
   # Find input files
 
   # These three arrays are used to store arguments of different types for
@@ -2067,6 +2064,7 @@ def phase_linker_setup(options, state, newargs):
   if settings.STRICT:
     if not settings.MODULARIZE and not settings.EXPORT_ES6:
       default_setting('STRICT_JS', 1)
+    default_setting('DEFAULT_TO_CXX', 0)
     default_setting('AUTO_JS_LIBRARIES', 0)
     default_setting('AUTO_NATIVE_LIBRARIES', 0)
     default_setting('AUTO_ARCHIVE_INDEXES', 0)
@@ -3039,12 +3037,14 @@ def is_link_flag(flag):
       return True
     return flag.startswith(('-l', '-L', '-Wl,'))
 
-  CXX = [shared.CLANG_CXX]
-  CC = [shared.CLANG_CC]
+  if run_via_emxx:
+    compiler = [shared.CLANG_CXX]
+  else:
+    compiler = [shared.CLANG_CC]
+
   if config.COMPILER_WRAPPER:
     logger.debug('using compiler wrapper: %s', config.COMPILER_WRAPPER)
-    CXX.insert(0, config.COMPILER_WRAPPER)
-    CC.insert(0, config.COMPILER_WRAPPER)
+    compiler.insert(0, config.COMPILER_WRAPPER)
 
   compile_args = [a for a in newargs if a and not is_link_flag(a)]
   system_libs.ensure_sysroot()
@@ -3062,39 +3062,16 @@ def get_language_mode(args):
     return ''
 
   language_mode = get_language_mode(newargs)
-
-  def use_cxx(src):
-    if 'c++' in language_mode or run_via_emxx:
-      return True
-    suffix = shared.suffix(src)
-    # Next consider the filename
-    if suffix in C_ENDINGS + OBJC_ENDINGS:
-      return False
-    if suffix in CXX_ENDINGS:
-      return True
-    # Finally fall back to the default
-    if settings.DEFAULT_TO_CXX:
-      # Default to using C++ even when run as `emcc`.
-      # This means that emcc will act as a C++ linker when no source files are
-      # specified.
-      # This differs to clang and gcc where the default is always C unless run as
-      # clang++/g++.
-      return True
-    return False
-
-  def get_compiler(src_file):
-    if use_cxx(src_file):
-      return CXX
-    return CC
+  use_cxx = 'c++' in language_mode or run_via_emxx
 
   def get_clang_command(src_file):
-    return get_compiler(src_file) + get_cflags(state.orig_args, use_cxx(src_file)) + compile_args + [src_file]
+    return compiler + get_cflags(state.orig_args, use_cxx) + compile_args + [src_file]
 
   def get_clang_command_preprocessed(src_file):
-    return get_compiler(src_file) + get_clang_flags(state.orig_args) + compile_args + [src_file]
+    return compiler + get_clang_flags(state.orig_args) + compile_args + [src_file]
 
   def get_clang_command_asm(src_file):
-    return get_compiler(src_file) + get_target_flags() + compile_args + [src_file]
+    return compiler + get_target_flags() + compile_args + [src_file]
 
   # preprocessor-only (-E) support
   if state.mode == Mode.PREPROCESS_ONLY:

src/settings.js

@@ -1944,9 +1944,9 @@ var USE_OFFSET_CONVERTER = false;
 var LOAD_SOURCE_MAP = false;
 
 // Default to c++ mode even when run as `emcc` rather then `emc++`.
-// When this is disabled `em++` is required when compiling and linking C++
-// programs. This which matches the behaviour of gcc/g++ and clang/clang++.
-// [compile+link]
+// When this is disabled `em++` is required linking C++ programs. Disabling
+// this will match the behaviour of gcc/g++ and clang/clang++.
+// [link]
 var DEFAULT_TO_CXX = true;
 
 // While LLVM's wasm32 has long double = float128, we don't support printing

test/test_other.py

@@ -11720,29 +11720,16 @@ def test_default_to_cxx(self):
     create_file('foo.h', '#include <string.h>')
     create_file('cxxfoo.h', '#include <string>')
 
-    # The default bahviour is to default to C++, which means the C++ header can be compiled even
-    # with emcc.
-    self.run_process([EMCC, '-c', 'cxxfoo.h'])
+    # Compiling a C++ header using `em++` works.
+    self.run_process([EMXX, '-c', 'cxxfoo.h'])
 
-    # But this means that C flags can't be passed (since we are assuming C++)
-    err = self.expect_fail([EMCC, '-std=gnu11', '-c', 'foo.h'])
-    self.assertContained("'-std=gnu11' not allowed with 'C++'", err)
-
-    # If we disable DEFAULT_TO_CXX the emcc can be used with C-only flags (e.g. -std=gnu11),
-    self.run_process([EMCC, '-std=gnu11', '-c', 'foo.h', '-sDEFAULT_TO_CXX=0'])
-
-    # But can't be used to build C++ headers
-    err = self.expect_fail([EMCC, '-c', 'cxxfoo.h', '-sDEFAULT_TO_CXX=0'])
-    self.assertContained("'string' file not found", err)
-
-    # Check that STRICT also disables DEFAULT_TO_CXX
+    # Compiling the same header using `emcc` fails, just like `clang`
     err = self.expect_fail([EMCC, '-c', 'cxxfoo.h', '-sSTRICT'])
     self.assertContained("'string' file not found", err)
 
-    # Using em++ should alwasy work for C++ headers
-    self.run_process([EMXX, '-c', 'cxxfoo.h', '-sDEFAULT_TO_CXX=0'])
-    # Or using emcc with `-x c++`
-    self.run_process([EMCC, '-c', 'cxxfoo.h', '-sDEFAULT_TO_CXX=0', '-x', 'c++-header'])
+    # But it works if we pass and explicit language mode.
+    self.run_process([EMCC, '-c', 'cxxfoo.h', '-x', 'c++-header'])
+    self.run_process([EMCC, '-c', 'cxxfoo.h', '-x', 'c++'])
 
   @parameterized({
     '': ([],),

tools/settings.py

@@ -70,7 +70,6 @@
     'USE_PTHREADS', # legacy name of PTHREADS setting
     'SHARED_MEMORY',
     'SUPPORT_LONGJMP',
-    'DEFAULT_TO_CXX',
     'WASM_OBJECT_FILES',
     'WASM_WORKERS',
     'BULK_MEMORY',