Enable reporting of undefined symbols in wasm-ld

ChangeLog.md

@@ -17,6 +17,10 @@ See docs/process.md for how version tagging works.
 
 Current Trunk
 -------------
+- Add LLD_REPORT_UNDEFINED option that should allow for more detailed
+  diagnostics when symbols are undefined at link time.  This currently has
+  some limitations and is not enabled by default. For example, EM_JS symbols
+  are reported as undefined at link time, as are `__invoke_*` functions.
 
 v1.39.7: 02/03/2020
 -------------------

emcc.py

@@ -512,6 +512,41 @@ def ensure_archive_index(archive_file):
     run_process([shared.LLVM_RANLIB, archive_file])
 
 
+def get_all_js_library_funcs(temp_files):
+  # Runs the js compiler to generate a list of all functions available in the JS
+  # libraries.  This must be done separately for each linker invokation since the
+  # list of library functions depends on what settings are used.
+  # TODO(sbc): Find a way to optimize this.  Potentially we could add a super-set
+  # mode of the js compiler that would generate a list of all possible symbols
+  # that could be checked in.
+  old_full = shared.Settings.INCLUDE_FULL_LIBRARY
+  old_linkable = shared.Settings.LINKABLE
+  try:
+    # Temporarily define INCLUDE_FULL_LIBRARY since we want a full list
+    # of all available JS library functions.
+    shared.Settings.INCLUDE_FULL_LIBRARY = True
+    # Temporarily set LINKABLE so that the jscompiler doesn't report
+    # undefined symbolls itself.
+    shared.Settings.LINKABLE = True
+    emscripten.generate_struct_info()
+    glue, forwarded_data = emscripten.compile_settings(temp_files)
+    forwarded_json = json.loads(forwarded_data)
+    library_fns = forwarded_json['Functions']['libraryFunctions']
+    library_fns_list = []
+    for name in library_fns:
+      if shared.is_c_symbol(name):
+        name = shared.demangle_c_symbol_name(name)
+        library_fns_list.append(name)
+        # TODO(sbc): wasm-ld shouldn't be reporting errors for symbols
+        # such as __wasi_fd_write which are defined with import_name attibutes
+        # but it currently does.  Remove this once we fix wasm-ld.
+        library_fns_list.append('__wasi_' + name)
+  finally:
+    shared.Settings.INCLUDE_FULL_LIBRARY = old_full
+    shared.Settings.LINKABLE = old_linkable
+  return library_fns_list
+
+
 #
 # Main run() function
 #
@@ -2210,7 +2245,14 @@ def get_final():
             lto_level = options.opt_level
           else:
             lto_level = 0
-          final = shared.Building.link_lld(linker_inputs, DEFAULT_FINAL, lto_level=lto_level)
+          all_externals = None
+          if shared.Settings.LLD_REPORT_UNDEFINED:
+            all_externals = get_all_js_library_funcs(misc_temp_files)
+            log_time('JS symbol generation')
+            # TODO(sbc): This is an incomplete list of __invoke functions.  Perhaps add
+            # support for wildcard to wasm-ld.
+            all_externals += ['emscripten_longjmp_jmpbuf', '__invoke_void', '__invoke_i32_i8*_...']
+          final = shared.Building.link_lld(linker_inputs, DEFAULT_FINAL, lto_level=lto_level, all_external_symbols=all_externals)
         else:
           final = shared.Building.link(linker_inputs, DEFAULT_FINAL, force_archive_contents=force_archive_contents, just_calculate=just_calculate)
       else:

emscripten.py

@@ -2769,11 +2769,8 @@ def normalize_line_endings(text):
   return text
 
 
-def run(infile, outfile, memfile):
-  temp_files = get_configuration().get_temp_files()
-  infile, outfile = substitute_response_files([infile, outfile])
-
-  if not shared.Settings.BOOTSTRAPPING_STRUCT_INFO:
+def generate_struct_info():
+  if not shared.Settings.STRUCT_INFO and not shared.Settings.BOOTSTRAPPING_STRUCT_INFO:
     generated_struct_info_name = 'generated_struct_info.json'
 
     def generate_struct_info():
@@ -2785,6 +2782,12 @@ def generate_struct_info():
     shared.Settings.STRUCT_INFO = shared.Cache.get(generated_struct_info_name, generate_struct_info)
   # do we need an else, to define it for the bootstrap case?
 
+
+def run(infile, outfile, memfile):
+  temp_files = get_configuration().get_temp_files()
+  infile, outfile = substitute_response_files([infile, outfile])
+  generate_struct_info()
+
   outfile_obj = open(outfile, 'w')
 
   emscripter = emscript_wasm_backend if shared.Settings.WASM_BACKEND else emscript_fastcomp

src/jsifier.js

@@ -162,11 +162,15 @@ function JSify(data, functionsOnly) {
         return;
       } else if ((!LibraryManager.library.hasOwnProperty(ident) && !LibraryManager.library.hasOwnProperty(ident + '__inline')) || SIDE_MODULE) {
         if (!(finalName in IMPLEMENTED_FUNCTIONS) && !LINKABLE) {
+          var msg = 'undefined symbol: ' + ident;
           if (ERROR_ON_UNDEFINED_SYMBOLS) {
-            error('undefined symbol: ' + ident);
+            error(msg);
+            if (WASM_BACKEND) {
+              warnOnce('Link with `-s LLD_REPORT_UNDEFINED` to get more information on undefined symbols');
+            }
             warnOnce('To disable errors for undefined symbols use `-s ERROR_ON_UNDEFINED_SYMBOLS=0`')
           } else if (VERBOSE || WARN_ON_UNDEFINED_SYMBOLS) {
-            warn('undefined symbol: ' + ident);
+            warn(msg);
           }
         }
         if (!RELOCATABLE) {

src/library.js

@@ -4791,40 +4791,41 @@ LibraryManager.library = {
     {{{ makeDynCall('vii') }}}(func, Math.min(base, end), Math.max(base, end));
   },
 
-  // misc definitions to avoid unnecessary unresolved symbols from fastcomp
+  // misc definitions to avoid unnecessary unresolved symbols being reported
+  // by fastcomp or wasm-ld
 #if SUPPORT_LONGJMP
-  emscripten_prep_setjmp: true,
-  emscripten_cleanup_setjmp: true,
-  emscripten_check_longjmp: true,
-  emscripten_get_longjmp_result: true,
-  emscripten_setjmp: true,
+  emscripten_prep_setjmp: function() {},
+  emscripten_cleanup_setjmp: function() {},
+  emscripten_check_longjmp: function() {},
+  emscripten_get_longjmp_result: function() {},
+  emscripten_setjmp: function() {},
 #endif
-  emscripten_preinvoke: true,
-  emscripten_postinvoke: true,
-  emscripten_resume: true,
-  emscripten_landingpad: true,
-  getHigh32: true,
-  setHigh32: true,
-  FtoILow: true,
-  FtoIHigh: true,
-  DtoILow: true,
-  DtoIHigh: true,
-  BDtoILow: true,
-  BDtoIHigh: true,
-  SItoF: true,
-  UItoF: true,
-  SItoD: true,
-  UItoD: true,
-  BItoD: true,
-  llvm_dbg_value: true,
-  llvm_debugtrap: true,
-  llvm_ctlz_i32: true,
-  emscripten_asm_const: true,
-  emscripten_asm_const_int: true,
-  emscripten_asm_const_double: true,
-  emscripten_asm_const_int_sync_on_main_thread: true,
-  emscripten_asm_const_double_sync_on_main_thread: true,
-  emscripten_asm_const_async_on_main_thread: true,
+  emscripten_preinvoke: function() {},
+  emscripten_postinvoke: function() {},
+  emscripten_resume: function() {},
+  emscripten_landingpad: function() {},
+  getHigh32: function() {},
+  setHigh32: function() {},
+  FtoILow: function() {},
+  FtoIHigh: function() {},
+  DtoILow: function() {},
+  DtoIHigh: function() {},
+  BDtoILow: function() {},
+  BDtoIHigh: function() {},
+  SItoF: function() {},
+  UItoF: function() {},
+  SItoD: function() {},
+  UItoD: function() {},
+  BItoD: function() {},
+  llvm_dbg_value: function() {},
+  llvm_debugtrap: function() {},
+  llvm_ctlz_i32: function() {},
+  emscripten_asm_const: function() {},
+  emscripten_asm_const_int: function() {},
+  emscripten_asm_const_double: function() {},
+  emscripten_asm_const_int_sync_on_main_thread: function() {},
+  emscripten_asm_const_double_sync_on_main_thread: function() {},
+  emscripten_asm_const_async_on_main_thread: function() {},
 
   // ======== compiled code from system/lib/compiler-rt , see readme therein
   __muldsi3__asm: true,

src/settings.js

@@ -1733,6 +1733,13 @@ var DISABLE_EXCEPTION_THROWING = 0;
 // across function boundaries.
 var USE_OFFSET_CONVERTER = 0;
 
+// If set to 1, the JS compiler is run before wasm-ld so that the linker can
+// report undefined symbols within the binary.  Without this option that linker
+// doesn't know which symmbols might be defined JS and so reporting of undefined
+// symbols is deleyed until the JS compiler is run.
+// [link]
+var LLD_REPORT_UNDEFINED = 0;
+
 //===========================================
 // Internal, used for testing only, from here
 //===========================================

tests/test_other.py

@@ -10301,3 +10301,10 @@ def test_signature_mismatch(self):
     self.assertContained('function signature mismatch: foo', stderr)
     self.expect_fail([PYTHON, EMCC, '-Wl,--fatal-warnings', 'a.c', 'b.c'])
     self.expect_fail([PYTHON, EMCC, '-s', 'STRICT', 'a.c', 'b.c'])
+
+  @no_fastcomp('lld only')
+  def test_lld_report_undefined(self):
+    create_test_file('main.c', 'void foo(); int main() { foo(); return 0; }')
+    stderr = self.expect_fail([PYTHON, EMCC, '-s', 'LLD_REPORT_UNDEFINED', 'main.c'])
+    self.assertContained('wasm-ld: error:', stderr)
+    self.assertContained('main_0.o: undefined symbol: foo', stderr)

tools/shared.py

@@ -1447,6 +1447,10 @@ def demangle_c_symbol_name(name):
   return name[1:] if name.startswith('_') else '$' + name
 
 
+def is_c_symbol(name):
+  return name.startswith('_')
+
+
 def treat_as_user_function(name):
   if name.startswith('dynCall_'):
     return False
@@ -1881,7 +1885,7 @@ def link_llvm(linker_inputs, target):
     return target
 
   @staticmethod
-  def link_lld(args, target, opts=[], lto_level=0):
+  def link_lld(args, target, opts=[], lto_level=0, all_external_symbols=None):
     if not os.path.exists(WASM_LD):
       exit_with_error('linker binary not found in LLVM directory: %s', WASM_LD)
     # runs lld to link things.
@@ -1903,10 +1907,17 @@ def link_lld(args, target, opts=[], lto_level=0):
         WASM_LD,
         '-o',
         target,
-        '--allow-undefined',
         '--lto-O%d' % lto_level,
     ] + args
 
+    if all_external_symbols:
+      undefs = configuration.get_temp_files().get('.undefined').name
+      with open(undefs, 'w') as f:
+        f.write('\n'.join(all_external_symbols))
+      cmd.append('--allow-undefined-file=%s' % undefs)
+    else:
+      cmd.append('--allow-undefined')
+
     # wasi does not import the memory (but for JS it is efficient to do so,
     # as it allows us to set up memory, preload files, etc. even before the
     # wasm module arrives)
@@ -1941,8 +1952,14 @@ def link_lld(args, target, opts=[], lto_level=0):
 
       cmd += ['--export', '__data_end']
 
-      for export in Settings.EXPORTED_FUNCTIONS:
-        cmd += ['--export', export[1:]] # Strip the leading underscore
+      c_exports = [e for e in Settings.EXPORTED_FUNCTIONS if is_c_symbol(e)]
+      # Strip the leading underscores
+      c_exports = [demangle_c_symbol_name(e) for e in c_exports]
+      if all_external_symbols:
+        # Filter out symbols external/JS symbols
+        c_exports = [e for e in c_exports if e not in all_external_symbols]
+      for export in c_exports:
+        cmd += ['--export', export]
 
     if Settings.RELOCATABLE:
       if Settings.SIDE_MODULE:
@@ -2314,7 +2331,7 @@ def get_safe_internalize():
     if len(internalize_list) > 8192:
       logger.debug('using response file for EXPORTED_FUNCTIONS in internalize')
       finalized_exports = '\n'.join([exp[1:] for exp in exps])
-      internalize_list_file = configuration.get_temp_files().get(suffix='.response').name
+      internalize_list_file = configuration.get_temp_files().get('.response').name
       with open(internalize_list_file, 'w') as f:
         f.write(finalized_exports)
       internalize_public_api += 'file=' + internalize_list_file