SIDE_DEBUG option: emit DWARF in a wasm on the side

emcc.py

@@ -2821,6 +2821,11 @@ def check_bad_eq(arg):
           newargs[i] = '-g'
           shared.Settings.FULL_DWARF = 1
           shared.warning('gforce_dwarf is a temporary option that will eventually disappear')
+        elif requested_level.startswith('side'):
+          # Emit full DWARF but also emit it in a file on the side
+          newargs[i] = '-g'
+          shared.Settings.FULL_DWARF = 1
+          shared.Settings.SIDE_DEBUG = 1
         # 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
@@ -3339,6 +3344,9 @@ def run_closure_compiler(final):
     with open(final, 'w') as f:
       f.write(js)
 
+  if shared.Settings.FULL_DWARF and shared.Settings.SIDE_DEBUG:
+    shared.Building.emit_debug_on_side(wasm_binary_target)
+
 
 def modularize():
   global final

src/settings.js

@@ -1492,6 +1492,14 @@ var ELIMINATE_DUPLICATE_FUNCTIONS_PASSES = 5;
 // the ctors.
 var EVAL_CTORS = 0;
 
+// Whether to emit DWARF in a wasm file on the side (this is not called
+// "split"/"separate" because there is already a DWARF concept by that name).
+// When DWARF is on the side, the main file has no DWARF info, while the side
+// file, ending in .debug.wasm, has the same wasm binary + all the debug
+// sections.
+// This has no effect if DWARF is not being emitted.
+var SIDE_DEBUG = 0;
+
 // see http://kripken.github.io/emscripten-site/docs/debugging/CyberDWARF.html
 // [fastcomp-only]
 var CYBERDWARF = 0;

tests/test_other.py

@@ -9033,6 +9033,21 @@ def test(infile, source_map_added_dir=''):
     ensure_dir('inner')
     test('inner/a.cpp', 'inner')
 
+  @no_fastcomp('dwarf')
+  def test_side_debug(self):
+    run_process([PYTHON, EMCC, path_from_root('tests', 'hello_world.c'), '-gforce_dwarf'])
+    self.assertExists('a.out.wasm')
+    self.assertNotExists('a.out.wasm.debug.wasm')
+    run_process([PYTHON, EMCC, path_from_root('tests', 'hello_world.c'), '-gside'])
+    self.assertExists('a.out.wasm')
+    self.assertExists('a.out.wasm.debug.wasm')
+    self.assertLess(os.path.getsize('a.out.wasm'), os.path.getsize('a.out.wasm.debug.wasm'))
+    # the special section should also exist, that refers to the side debug file
+    with open('a.out.wasm', 'rb') as f:
+      wasm = f.read()
+      self.assertIn(b'external_debug_info', wasm)
+      self.assertIn(b'a.out.wasm.debug.wasm', wasm)
+
   def test_wasm_producers_section(self):
     # no producers section by default
     run_process([PYTHON, EMCC, path_from_root('tests', 'hello_world.c')])

tools/shared.py

@@ -704,6 +704,7 @@ def replace_or_append_suffix(filename, new_suffix):
 LLVM_INTERPRETER = os.path.expanduser(build_llvm_tool_path(exe_suffix('lli')))
 LLVM_COMPILER = os.path.expanduser(build_llvm_tool_path(exe_suffix('llc')))
 LLVM_DWARFDUMP = os.path.expanduser(build_llvm_tool_path(exe_suffix('llvm-dwarfdump')))
+LLVM_OBJCOPY = os.path.expanduser(build_llvm_tool_path(exe_suffix('llvm-objcopy')))
 WASM_LD = os.path.expanduser(build_llvm_tool_path(exe_suffix('wasm-ld')))
 
 EMSCRIPTEN = path_from_root('emscripten.py')
@@ -2756,6 +2757,27 @@ def wasm2js(js_file, wasm_file, opt_level, minify_whitespace, use_closure_compil
       f.write(all_js)
     return js_file
 
+  @staticmethod
+  def emit_debug_on_side(wasm_file):
+    # extract the DWARF info from the main file, and leave the wasm with
+    # debug into as a file on the side
+    # TODO: emit only debug sections in the side file, and not the entire
+    #       wasm as well
+    wasm_file_with_dwarf = wasm_file + '.debug.wasm'
+    shutil.move(wasm_file, wasm_file_with_dwarf)
+    run_process([LLVM_OBJCOPY, '--remove-section=.debug*', wasm_file_with_dwarf, wasm_file])
+
+    # embed a section in the main wasm to point to the file with external DWARF,
+    # see https://yurydelendik.github.io/webassembly-dwarf/#external-DWARF
+    section_name = b'\x13external_debug_info' # section name, including prefixed size
+    contents = asbytes(wasm_file_with_dwarf)
+    section_size = len(section_name) + len(contents)
+    with open(wasm_file, 'ab') as f:
+      f.write(b'\0') # user section is code 0
+      f.write(WebAssembly.lebify(section_size))
+      f.write(section_name)
+      f.write(contents)
+
   @staticmethod
   def apply_wasm_memory_growth(js_file):
     logger.debug('supporting wasm memory growth with pthreads')