Support arbitrary -gX debug options, and limit DWARF to line-tables-only (emscripten-core#9872)

Fixes emscripten-core#9611

Author: kripken

emcc.py

@@ -1821,6 +1821,7 @@ def check_human_readable_list(items):
       shared.Settings.BUNDLED_CD_DEBUG_FILE = target + ".cd"
       shared.Settings.SYSTEM_JS_LIBRARIES.append(shared.path_from_root('src', 'library_cyberdwarf.js'))
       shared.Settings.SYSTEM_JS_LIBRARIES.append(shared.path_from_root('src', 'library_debugger_toolkit.js'))
+      newargs.append('-g')
 
     if options.tracing:
       if shared.Settings.ALLOW_MEMORY_GROWTH:
@@ -1891,11 +1892,8 @@ def check_human_readable_list(items):
         logger.debug("running (for precompiled headers): " + clang_compiler + ' ' + ' '.join(args))
         return run_process([clang_compiler] + args, check=False).returncode
 
-      if need_llvm_debug_info(options):
-        newargs.append('-g')
-
       # For asm.js, the generated JavaScript could preserve LLVM value names, which can be useful for debugging.
-      if options.debug_level >= 3 and not shared.Settings.WASM:
+      if options.debug_level >= 3 and not shared.Settings.WASM and not shared.Settings.WASM_BACKEND:
         newargs.append('-fno-discard-value-names')
 
       def is_link_flag(flag):
@@ -2640,10 +2638,26 @@ def check_bad_eq(arg):
       newargs[i] = ''
       newargs[i + 1] = ''
     elif newargs[i].startswith('-g'):
-      requested_level = newargs[i][2:] or '3'
-      options.debug_level = validate_arg_level(requested_level, 4, 'Invalid debug level: ' + newargs[i])
       options.requested_debug = newargs[i]
-      newargs[i] = ''
+      requested_level = newargs[i][2:] or '3'
+      if is_int(requested_level):
+        # the -gX value is the debug level (-g1, -g2, etc.)
+        options.debug_level = validate_arg_level(requested_level, 4, 'Invalid debug level: ' + newargs[i])
+        # if we don't need to preserve LLVM debug info, do not keep this flag
+        # for clang
+        if options.debug_level < 3:
+          newargs[i] = ''
+        else:
+          # until we support full DWARF info, limit the clang frontend to just
+          # emit line tables, which can be represented in source maps
+          newargs[i] = '-gline-tables-only'
+      else:
+        # a non-integer level can be something like -gline-tables-only. keep
+        # the flag for the clang frontend to emit the appropriate DWARF info.
+        # set the emscripten debug level to 3 so that we do not remove that
+        # debug info during link (during compile, this does not make a
+        # difference).
+        options.debug_level = 3
     elif newargs[i] == '-profiling' or newargs[i] == '--profiling':
       options.debug_level = max(options.debug_level, 2)
       options.profiling = True
@@ -3676,6 +3690,14 @@ def validate_arg_level(level_string, max_level, err_msg, clamp=False):
   return level
 
 
+def is_int(s):
+  try:
+    int(s)
+    return True
+  except ValueError:
+    return False
+
+
 if __name__ == '__main__':
   try:
     sys.exit(run(sys.argv))

site/source/docs/tools_reference/emcc.rst

@@ -112,7 +112,7 @@ Options that are modified or new in *emcc* are listed below:
   Preserve debug information.
 
   - When compiling to object files, this is the same as in *Clang* and *gcc*, it adds debug information to the object files.
-  - When linking, this is equivalent to :ref:`-g3 <emcc-g3>` (preserve JS whitespace and compiled function names).
+  - When linking, this is equivalent to :ref:`-g3 <emcc-g3>`.
 
 .. _emcc-gN:
 
@@ -137,7 +137,7 @@ Options that are modified or new in *emcc* are listed below:
     -
       .. _emcc-g3:
 
-      ``-g3``: When compiling to object files, keep debug info (this is the same as :ref:`-g <emcc-g>`).
+      ``-g3``: When compiling to object files, keep debug info, including JS whitespace, function names, and LLVM debug info if any (this is the same as :ref:`-g <emcc-g>`).
 
     .. _emcc-g4:
 

tests/test_other.py

@@ -2329,6 +2329,33 @@ def test_debuginfo(self):
         else:
           self.assertIn('strip-debug', err)
 
+  @no_fastcomp()
+  def test_debuginfo_line_tables_only(self):
+    def test(do_compile):
+      do_compile([])
+      no_size = os.path.getsize('a.out.wasm')
+      do_compile(['-gline-tables-only'])
+      line_size = os.path.getsize('a.out.wasm')
+      do_compile(['-g'])
+      full_size = os.path.getsize('a.out.wasm')
+      self.assertLess(no_size, line_size)
+      # currently we don't support full debug info anyhow, so line tables
+      # is all we have
+      self.assertEqual(line_size, full_size)
+
+    def compile_directly(compile_args):
+      run_process([PYTHON, EMCC, path_from_root('tests', 'hello_world.cpp')] + compile_args)
+
+    test(compile_directly)
+
+    def compile_to_object_first(compile_args):
+      # compile with the specified args
+      run_process([PYTHON, EMCC, path_from_root('tests', 'hello_world.cpp')] + compile_args + ['-c', '-o', 'a.o'])
+      # link with debug info
+      run_process([PYTHON, EMCC, 'a.o', '-g'])
+
+    test(compile_to_object_first)
+
   @unittest.skipIf(not scons_path, 'scons not found in PATH')
   @with_env_modify({'EMSCRIPTEN_ROOT': path_from_root()})
   def test_scons(self):