Add custom "emscripten_metadata" section to standalone WASM

Currently it's not possible execute Emscripten-generated WASM files
without parsing certain data from the accompanying JS. This change adds
that data to the wasm file itself so that standalone WASM files
are executable by third-party WASM runtimes.

emcc.py

@@ -2605,6 +2605,8 @@ def do_binaryen(target, asm_target, options, memfile, wasm_binary_target,
     if not shared.Settings.WASM_BACKEND and not DEBUG:
       os.unlink(asm_target) # we don't need the asm.js, it can just confuse
     sys.exit(0) # and we are done.
+  else:
+    shared.WebAssembly.add_emscripten_metadata(final, wasm_binary_target)
   if options.opt_level >= 2:
     # minify the JS
     optimizer.do_minify() # calculate how to minify

tests/test_other.py

@@ -7873,6 +7873,7 @@ def test_binaryen_metadce(self):
     def test(filename, expectations, size_slack):
       # in -Os, -Oz, we remove imports wasm doesn't need
       for args, expected_len, expected_exists, expected_not_exists, expected_wasm_size, expected_wasm_imports, expected_wasm_exports, expected_wasm_funcs in expectations:
+        expected_wasm_size += 29 # account for 'emscripten_metadata' section
         print(args, expected_len, expected_exists, expected_not_exists, expected_wasm_size, expected_wasm_imports, expected_wasm_exports, expected_wasm_funcs)
         run_process([PYTHON, EMCC, filename, '-g2'] + args)
         # find the imports we send from JS

tools/shared.py

@@ -507,6 +507,16 @@ def get_emscripten_version(path):
 EMSCRIPTEN_VERSION = get_emscripten_version(path_from_root('emscripten-version.txt'))
 parts = [int(x) for x in EMSCRIPTEN_VERSION.split('.')]
 EMSCRIPTEN_VERSION_MAJOR, EMSCRIPTEN_VERSION_MINOR, EMSCRIPTEN_VERSION_TINY = parts
+# For the Emscripten-specific WASM metadata section, follows semver, changes
+# whenever metadata section changes structure
+# NB: major version 0 implies no compatibility
+(EMSCRIPTEN_METADATA_MAJOR, EMSCRIPTEN_METADATA_MINOR) = (0, 0)
+# For the JS/WASM ABI, follows semver, changes whenever musl C types
+# change size/signedness or syscalls change signature. On an ABI
+# change, increment EMSCRIPTEN_ABI_MINOR if EMSCRIPTEN_ABI_MAJOR == 0
+# or the ABI change is backwards compatible, otherwise increment
+# EMSCRIPTEN_ABI_MAJOR and set EMSCRIPTEN_ABI_MINOR = 0
+(EMSCRIPTEN_ABI_MAJOR, EMSCRIPTEN_ABI_MINOR) = (0, 0)
 
 
 def generate_sanity():
@@ -2974,16 +2984,61 @@ def lebify(x):
     return bytearray(ret)
 
   @staticmethod
-  def make_shared_library(js_file, wasm_file, needed_dynlibs):
-    # a wasm shared library has a special "dylink" section, see tools-conventions repo
+  def get_js_data(js_file, shared=False):
     js = open(js_file).read()
     m = re.search("var STATIC_BUMP = (\d+);", js)
     mem_size = int(m.group(1))
     m = re.search("Module\['wasmTableSize'\] = (\d+);", js)
     table_size = int(m.group(1))
-    m = re.search('gb = alignMemory\(getMemory\(\d+ \+ (\d+)\), (\d+) \|\| 1\);', js)
-    assert m.group(1) == m.group(2), 'js must contain a clear alignment for the wasm shared library'
-    mem_align = int(m.group(1))
+    if shared:
+      m = re.search('gb = alignMemory\(getMemory\(\d+ \+ (\d+)\), (\d+) \|\| 1\);', js)
+      assert m.group(1) == m.group(2), 'js must contain a clear alignment for the wasm shared library'
+      mem_align = int(m.group(1))
+    else:
+      mem_align = None
+    return (mem_size, table_size, mem_align)
+
+  @staticmethod
+  def add_emscripten_metadata(js_file, wasm_file):
+    (mem_size, table_size, _) = WebAssembly.get_js_data(js_file)
+    wasm = open(wasm_file, 'rb').read()
+
+    raw_name = b'emscripten_metadata'
+    name = bytes(bytearray([len(raw_name)])) + raw_name
+    contents = b''.join(map(bytes, map(WebAssembly.lebify, [
+      # metadata section version
+      EMSCRIPTEN_METADATA_MAJOR,
+      EMSCRIPTEN_METADATA_MINOR,
+      # NB: The structure of the following should only be changed
+      #     if EMSCRIPTEN_METADATA_MAJOR is incremented
+      # ABI version
+      EMSCRIPTEN_ABI_MAJOR,
+      EMSCRIPTEN_ABI_MINOR,
+      # static bump
+      mem_size,
+      # table size
+      table_size,
+      # NB: more data can be appended here as long as you increase
+      #     the EMSCRIPTEN_METADATA_MINOR
+    ])))
+
+    new_section = b''.join([
+      b'\0', # custom section byte ID (== 0)
+      bytes(WebAssembly.lebify(len(name) + len(contents))),
+      name,
+      contents,
+    ])
+
+    with open(wasm_file, 'wb') as f:
+      f.write(wasm[0:8]) # copy magic number and version
+      # Emscripten metadata should be before any standard wasm section
+      f.write(new_section)
+      f.write(wasm[8:])
+
+  @staticmethod
+  def make_shared_library(js_file, wasm_file, needed_dynlibs):
+    # a wasm shared library has a special "dylink" section, see tools-conventions repo
+    (mem_size, table_size, mem_align) = WebAssembly.get_js_data(js_file, True)
     mem_align = int(math.log(mem_align, 2))
     logger.debug('creating wasm dynamic library with mem size %d, table size %d, align %d' % (mem_size, table_size, mem_align))
     wso = js_file + '.wso'