[EH] Use LLVM implementation for new Wasm EH (#23469)

This adds a new `WASM` mode in `ExceptionLibrary`, which uses the new
standard Wasm (exnref) EH, adds a library variant for it, and make the
tests use the new LLVM implementation instead of the Binaryen
translator.

The CI won't pass until llvm/llvm-project#123915
lands.

This requires adding `-wasmexcept` and `-wasmsjlj` variants to
`MINIMAL_TASKS` in `embuilder.py`
because they are necessary to run coreX tests in CircleCI, because
dylink coreX tests in CircleCI rellies on `./embuilder build MINIMAL_PIC
--pic`, which is the sum of `MINIMAL_TASKS` and `MINIMAL_PIC_TASKS`.

Because a new variant is added to `ExceptionLibrary`, this increases
`SYSTEM` library size from 366M to 400M, roughly by 11%. We discussed
about the possibility of not shipping the exnref libraries and let them
be built on demand, but given that `SYSTEM` include all variants, I'm
not sure how we are going to do that:
https://github.com/emscripten-core/emscripten/blob/73ebb91029948e62b3a4cea9ccc8db2dd87162b5/embuilder.py#L174-L177

Author: aheejin

Diff for: ChangeLog.md

@@ -20,6 +20,11 @@ See docs/process.md for more on how version tagging works.
 
 4.0.2 (in development)
 ----------------------
+- The standard Wasm EH, enabled by `-sWASM_LEGACY_EXCEPTIONS=0`, now uses the
+  LLVM backend implementation rather than the previously used Binaryen
+  translator
+  (https://github.com/WebAssembly/binaryen/blob/main/src/passes/TranslateEH.cpp).
+  (#23469) No specific action from the user is required.
 - Added support for compiling AVX2 intrinsics, 256-bit wide intrinsic is emulated
   on top of 128-bit Wasm SIMD instruction set. (#23035). Pass `-msimd128 -mavx2`
   to enable targeting AVX2.

Diff for: embuilder.py

@@ -32,6 +32,7 @@
 MINIMAL_TASKS = [
     'libcompiler_rt',
     'libcompiler_rt-legacysjlj',
+    'libcompiler_rt-wasmsjlj',
     'libcompiler_rt-ww',
     'libc',
     'libc-debug',
@@ -40,13 +41,16 @@
     'libc_optz-debug',
     'libc++abi',
     'libc++abi-legacyexcept',
+    'libc++abi-wasmexcept',
     'libc++abi-noexcept',
     'libc++abi-debug',
     'libc++abi-debug-legacyexcept',
+    'libc++abi-debug-wasmexcept',
     'libc++abi-debug-noexcept',
     'libc++abi-debug-ww-noexcept',
     'libc++',
     'libc++-legacyexcept',
+    'libc++-wasmexcept',
     'libc++-noexcept',
     'libc++-ww-noexcept',
     'libal',
@@ -80,6 +84,7 @@
     'crt1_proxy_main',
     'crtbegin',
     'libunwind-legacyexcept',
+    'libunwind-wasmexcept',
     'libnoexit',
     'sqlite3',
     'sqlite3-mt',

Diff for: site/source/docs/tools_reference/settings_reference.rst

@@ -1165,6 +1165,8 @@ https://github.com/WebAssembly/exception-handling/blob/main/proposals/exception-
 If false, emit instructions for the standardized exception handling proposal:
 https://github.com/WebAssembly/exception-handling/blob/main/proposals/exception-handling/Exceptions.md
 
+.. note:: Applicable during both linking and compilation
+
 Default value: true
 
 .. _nodejs_catch_exit:

Diff for: src/settings.js

@@ -781,7 +781,7 @@ var EXCEPTION_STACK_TRACES = false;
 // https://github.com/WebAssembly/exception-handling/blob/main/proposals/exception-handling/legacy/Exceptions.md
 // If false, emit instructions for the standardized exception handling proposal:
 // https://github.com/WebAssembly/exception-handling/blob/main/proposals/exception-handling/Exceptions.md
-// [link]
+// [compile+link]
 var WASM_LEGACY_EXCEPTIONS = true;
 
 // Emscripten throws an ExitStatus exception to unwind when exit() is called.

Diff for: test/other/codesize/test_codesize_cxx_except_wasm.size

@@ -1 +1 @@
-144652
+144531

Diff for: test/test_other.py

@@ -3535,7 +3535,10 @@ def test_embind_tsgen_jspi(self):
     'legacy': [1]
   })
   def test_embind_tsgen_exceptions(self, legacy):
+    if not legacy and shared.get_node_version(config.NODE_JS)[0] < 22:
+      self.skipTest('Node version needs to be 22 or greater to run tsgen with exnref')
     self.set_setting('WASM_LEGACY_EXCEPTIONS', legacy)
+
     # Check that when Wasm exceptions and assertions are enabled bindings still generate.
     self.run_process([EMXX, test_file('other/embind_tsgen.cpp'),
                       '-lembind', '-fwasm-exceptions', '-sASSERTIONS',

Diff for: tools/building.py

@@ -93,6 +93,12 @@ def llvm_backend_args():
   elif settings.SUPPORT_LONGJMP == 'wasm':
     args += ['-wasm-enable-sjlj']
 
+  if settings.WASM_EXCEPTIONS:
+    if settings.WASM_LEGACY_EXCEPTIONS:
+      args += ['-wasm-use-legacy-eh']
+    else:
+      args += ['-wasm-use-legacy-eh=0']
+
   # better (smaller, sometimes faster) codegen, see binaryen#1054
   # and https://bugs.llvm.org/show_bug.cgi?id=39488
   args += ['-disable-lsr']
@@ -277,6 +283,10 @@ def link_lld(args, target, external_symbols=None):
 
   if settings.WASM_EXCEPTIONS:
     cmd += ['-mllvm', '-wasm-enable-eh']
+    if settings.WASM_LEGACY_EXCEPTIONS:
+      cmd += ['-mllvm', '-wasm-use-legacy-eh']
+    else:
+      cmd += ['-mllvm', '-wasm-use-legacy-eh=0']
   if settings.WASM_EXCEPTIONS or settings.SUPPORT_LONGJMP == 'wasm':
     cmd += ['-mllvm', '-exception-model=wasm']
 

Diff for: tools/link.py

@@ -431,10 +431,6 @@ def check_human_readable_list(items):
     extras = settings.BINARYEN_EXTRA_PASSES.split(',')
     passes += [('--' + p) if p[0] != '-' else p for p in extras if p]
 
-  # Run the translator to the standardized EH instructions.
-  if not settings.WASM_LEGACY_EXCEPTIONS:
-    passes += ['--emit-exnref']
-
   # If we are going to run metadce then that means we will be running binaryen
   # tools after the main invocation, whose flags are determined here
   # (specifically we will run metadce and possibly also wasm-opt for import/

Diff for: tools/settings.py

@@ -84,6 +84,7 @@
     'INLINING_LIMIT',
     'DISABLE_EXCEPTION_CATCHING',
     'DISABLE_EXCEPTION_THROWING',
+    'WASM_LEGACY_EXCEPTIONS',
     'MAIN_MODULE',
     'SIDE_MODULE',
     'RELOCATABLE',

Diff for: tools/shared.py

@@ -379,11 +379,13 @@ def node_reference_types_flags(nodejs):
 
 def node_exception_flags(nodejs):
   node_version = get_node_version(nodejs)
-  # Exception handling was enabled by default in node v17.
+  # Legacy exception handling was enabled by default in node v17.
   if node_version and node_version < (17, 0, 0):
     return ['--experimental-wasm-eh']
-  else:
-    return []
+  # Standard exception handling was supported behind flag in node v22.
+  if node_version and node_version >= (22, 0, 0) and not settings.WASM_LEGACY_EXCEPTIONS:
+    return ['--experimental-wasm-exnref']
+  return []
 
 
 def node_pthread_flags(nodejs):

Diff for: tools/system_libs.py

@@ -775,10 +775,12 @@ class Exceptions(IntEnum):
   - EMSCRIPTEN: Emscripten provides exception handling capability using JS
     emulation. This causes code size increase and performance degradation.
   - WASM_LEGACY: Wasm native exception handling support (legacy)
+  - WASM: Wasm native exception handling support
   """
   NONE = auto()
   EMSCRIPTEN = auto()
   WASM_LEGACY = auto()
+  WASM = auto()
 
 
 class ExceptionLibrary(Library):
@@ -793,7 +795,9 @@ def get_cflags(self):
     elif self.eh_mode == Exceptions.EMSCRIPTEN:
       cflags += ['-sDISABLE_EXCEPTION_CATCHING=0']
     elif self.eh_mode == Exceptions.WASM_LEGACY:
-      cflags += ['-fwasm-exceptions']
+      cflags += ['-fwasm-exceptions', '-sWASM_LEGACY_EXCEPTIONS']
+    elif self.eh_mode == Exceptions.WASM:
+      cflags += ['-fwasm-exceptions', '-sWASM_LEGACY_EXCEPTIONS=0']
 
     return cflags
 
@@ -805,19 +809,25 @@ def get_base_name(self):
       name += '-noexcept'
     elif self.eh_mode == Exceptions.WASM_LEGACY:
       name += '-legacyexcept'
+    elif self.eh_mode == Exceptions.WASM:
+      name += '-wasmexcept'
     return name
 
   @classmethod
   def variations(cls):
     combos = super().variations()
     return ([dict(eh_mode=Exceptions.NONE, **combo) for combo in combos] +
             [dict(eh_mode=Exceptions.EMSCRIPTEN, **combo) for combo in combos] +
-            [dict(eh_mode=Exceptions.WASM_LEGACY, **combo) for combo in combos])
+            [dict(eh_mode=Exceptions.WASM_LEGACY, **combo) for combo in combos] +
+            [dict(eh_mode=Exceptions.WASM, **combo) for combo in combos])
 
   @classmethod
   def get_default_variation(cls, **kwargs):
     if settings.WASM_EXCEPTIONS:
-      eh_mode = Exceptions.WASM_LEGACY
+      if settings.WASM_LEGACY_EXCEPTIONS:
+        eh_mode = Exceptions.WASM_LEGACY
+      else:
+        eh_mode = Exceptions.WASM
     elif settings.DISABLE_EXCEPTION_CATCHING == 1:
       eh_mode = Exceptions.NONE
     else:
@@ -838,6 +848,12 @@ def get_cflags(self):
                  '-sDISABLE_EXCEPTION_THROWING=0']
     elif self.eh_mode == Exceptions.WASM_LEGACY:
       cflags += ['-sSUPPORT_LONGJMP=wasm',
+                 '-sWASM_LEGACY_EXCEPTIONS',
+                 '-sDISABLE_EXCEPTION_THROWING',
+                 '-D__WASM_SJLJ__']
+    elif self.eh_mode == Exceptions.WASM:
+      cflags += ['-sSUPPORT_LONGJMP=wasm',
+                 '-sWASM_LEGACY_EXCEPTIONS=0',
                  '-sDISABLE_EXCEPTION_THROWING',
                  '-D__WASM_SJLJ__']
     return cflags
@@ -848,18 +864,24 @@ def get_base_name(self):
     # suffixes. Change the default to wasm exception later.
     if self.eh_mode == Exceptions.WASM_LEGACY:
       name += '-legacysjlj'
+    elif self.eh_mode == Exceptions.WASM:
+      name += '-wasmsjlj'
     return name
 
   @classmethod
   def variations(cls):
     combos = super().variations()
     return ([dict(eh_mode=Exceptions.EMSCRIPTEN, **combo) for combo in combos] +
-            [dict(eh_mode=Exceptions.WASM_LEGACY, **combo) for combo in combos])
+            [dict(eh_mode=Exceptions.WASM_LEGACY, **combo) for combo in combos] +
+            [dict(eh_mode=Exceptions.WASM, **combo) for combo in combos])
 
   @classmethod
   def get_default_variation(cls, **kwargs):
     if settings.SUPPORT_LONGJMP == 'wasm':
-      eh_mode = Exceptions.WASM_LEGACY
+      if settings.WASM_LEGACY_EXCEPTIONS:
+        eh_mode = Exceptions.WASM_LEGACY
+      else:
+        eh_mode = Exceptions.WASM
     else:
       eh_mode = Exceptions.EMSCRIPTEN
     return super().get_default_variation(eh_mode=eh_mode, **kwargs)
@@ -1600,7 +1622,7 @@ def get_files(self):
       filenames += ['cxa_noexception.cpp']
     elif self.eh_mode == Exceptions.EMSCRIPTEN:
       filenames += ['cxa_exception_emscripten.cpp']
-    elif self.eh_mode == Exceptions.WASM_LEGACY:
+    elif self.eh_mode in (Exceptions.WASM_LEGACY, Exceptions.WASM):
       filenames += [
         'cxa_exception_storage.cpp',
         'cxa_exception.cpp',
@@ -1654,7 +1676,7 @@ class libcxx(ExceptionLibrary, MTLibrary):
 
   def get_cflags(self):
     cflags = super().get_cflags()
-    if self.eh_mode == Exceptions.WASM_LEGACY:
+    if self.eh_mode in (Exceptions.WASM_LEGACY, Exceptions.WASM):
       cflags.append('-D__WASM_EXCEPTIONS__')
     return cflags
 
@@ -1677,7 +1699,7 @@ def __init__(self, **kwargs):
     super().__init__(**kwargs)
 
   def can_use(self):
-    return super().can_use() and self.eh_mode == Exceptions.WASM_LEGACY
+    return super().can_use() and self.eh_mode in (Exceptions.WASM_LEGACY, Exceptions.WASM)
 
   def get_cflags(self):
     cflags = super().get_cflags()
@@ -1688,7 +1710,7 @@ def get_cflags(self):
       cflags.append('-D_LIBUNWIND_HAS_NO_EXCEPTIONS')
     elif self.eh_mode == Exceptions.EMSCRIPTEN:
       cflags.append('-D__EMSCRIPTEN_EXCEPTIONS__')
-    elif self.eh_mode == Exceptions.WASM_LEGACY:
+    elif self.eh_mode in (Exceptions.WASM_LEGACY, Exceptions.WASM):
       cflags.append('-D__WASM_EXCEPTIONS__')
     return cflags
 

Diff for: tools/webassembly.py

@@ -75,6 +75,7 @@ class Type(IntEnum):
   V128 = 0x7b # -0x5
   FUNCREF = 0x70 # -0x10
   EXTERNREF = 0x6f # -0x11
+  EXNREF = 0x69 # -0x17
   VOID = 0x40 # -0x40