Avoid always including getWasmTableEntry/setWasmTableEntry library functions (#17373)

Author: sbc100

ChangeLog.md

@@ -36,6 +36,9 @@ See docs/process.md for more on how version tagging works.
   added to `DEFAULT_LIBRARY_FUNCS_TO_INCLUDE`. (#17370)
 - The `run` runtime function is no longer exported by default.  It can be added
   to `EXPORTED_RUNTIME_METHODS` if needed.
+- The getWasmTableEntry/setWasmTableEntry library function are no longer
+  included by default.  Add them to `DEFAULT_LIBRARY_FUNCS_TO_INCLUDE` or
+  `EXPORTED_RUNTIME_METHODS` if you want to use them outside of JS library code.
 
 3.1.15 - 07/01/2022
 -------------------

emcc.py

@@ -1851,8 +1851,6 @@ def phase_linker_setup(options, state, newargs, user_settings):
     settings.DEFAULT_LIBRARY_FUNCS_TO_INCLUDE += ['$dynCall']
 
   if not settings.BOOTSTRAPPING_STRUCT_INFO:
-    # Include the internal library function since they are used by runtime functions.
-    settings.DEFAULT_LIBRARY_FUNCS_TO_INCLUDE += ['$getWasmTableEntry', '$setWasmTableEntry']
     if settings.SAFE_HEAP:
       settings.DEFAULT_LIBRARY_FUNCS_TO_INCLUDE += ['$getValue_safe', '$setValue_safe']
     if not settings.MINIMAL_RUNTIME:

emscripten.py

@@ -347,6 +347,10 @@ def emscript(in_wasm, out_wasm, outfile_js, memfile):
     if settings.INITIAL_TABLE == -1:
       settings.INITIAL_TABLE = dylink_sec.table_size + 1
 
+  invoke_funcs = metadata['invokeFuncs']
+  if invoke_funcs:
+    settings.DEFAULT_LIBRARY_FUNCS_TO_INCLUDE += ['$getWasmTableEntry']
+
   glue, forwarded_data = compile_settings()
   if DEBUG:
     logger.debug('  emscript: glue took %s seconds' % (time.time() - t))
@@ -397,7 +401,6 @@ def emscript(in_wasm, out_wasm, outfile_js, memfile):
     out.write(normalize_line_endings(pre))
     pre = None
 
-    invoke_funcs = metadata['invokeFuncs']
     sending = create_sending(invoke_funcs, metadata)
     receiving = create_receiving(exports)
 

src/jsifier.js

@@ -143,10 +143,12 @@ ${argConvertions}
     });
   }
 
-  function processLibraryFunction(snippet, ident, finalName) {
-    // It is possible that when printing the function as a string on Windows, the js interpreter we are in returns the string with Windows
-    // line endings \r\n. This is undesirable, since line endings are managed in the form \n in the output for binary file writes, so
-    // make sure the endings are uniform.
+  function processLibraryFunction(snippet, ident, finalName, deps) {
+    // It is possible that when printing the function as a string on Windows,
+    // the js interpreter we are in returns the string with Windows line endings
+    // \r\n. This is undesirable, since line endings are managed in the form \n
+    // in the output for binary file writes, so make sure the endings are
+    // uniform.
     snippet = snippet.toString().replace(/\r\n/gm, '\n');
 
     // name the function; overwrite if it's already named
@@ -163,20 +165,37 @@ function ${name}(${args}) {
   return ret;
 }`);
     }
+
     if (MEMORY64) {
       const sig = LibraryManager.library[ident + '__sig'];
       if (sig && sig.includes('p')) {
         snippet = convertPointerParams(snippet, sig);
       }
     }
+
     return snippet;
   }
 
+  function addImplicitDeps(snippet, deps) {
+    // There are some very common dependencies that we inject automatically
+    // by conservatively scanning the input functions for their usage.
+    // Specifically, these are dependencies that are automatically generated by
+    // the {{{ makeDynCall }}} macro which is very common.
+    const autoDeps = ['getDynCaller', 'getWasmTableEntry'];
+    for (const dep of autoDeps) {
+      if (snippet.includes(dep + '(')) {
+        deps.push('$' + dep);
+      }
+    }
+  }
+
   // functionStub
   function functionStubHandler(item) {
-    // In LLVM, exceptions generate a set of functions of form __cxa_find_matching_catch_1(), __cxa_find_matching_catch_2(), etc.
-    // where the number specifies the number of arguments. In Emscripten, route all these to a single function '__cxa_find_matching_catch'
-    // that variadically processes all of these functions using JS 'arguments' object.
+    // In LLVM, exceptions generate a set of functions of form
+    // __cxa_find_matching_catch_1(), __cxa_find_matching_catch_2(), etc.  where
+    // the number specifies the number of arguments. In Emscripten, route all
+    // these to a single function '__cxa_find_matching_catch' that variadically
+    // processes all of these functions using JS 'arguments' object.
     if (item.identMangled.startsWith('___cxa_find_matching_catch_')) {
       if (DISABLE_EXCEPTION_THROWING) {
         error('DISABLE_EXCEPTION_THROWING was set (likely due to -fno-exceptions), which means no C++ exception throwing support code is linked in, but exception catching code appears. Either do not set DISABLE_EXCEPTION_THROWING (if you do want exception throwing) or compile all source files with -fno-except (so that no exceptions support code is required); also make sure DISABLE_EXCEPTION_CATCHING is set to the right value - if you want exceptions, it should be off, and vice versa.');
@@ -301,9 +320,11 @@ function ${name}(${args}) {
         }
       } else if (typeof snippet == 'object') {
         snippet = stringifyWithFunctions(snippet);
+        addImplicitDeps(snippet, deps);
       } else if (typeof snippet == 'function') {
         isFunction = true;
-        snippet = processLibraryFunction(snippet, ident, finalName);
+        snippet = processLibraryFunction(snippet, ident, finalName, deps);
+        addImplicitDeps(snippet, deps);
         libraryFunctions.push(finalName);
       }
 

tests/other/metadce/test_metadce_cxx_except.jssize

@@ -1 +1 @@
-31596
+31595

tests/other/metadce/test_metadce_cxx_mangle.jssize

@@ -1 +1 @@
-31696
+31695

tests/other/test_unoptimized_code_size.js.size

@@ -1 +1 @@
-93189
+92392

tests/other/test_unoptimized_code_size_no_asserts.js.size

@@ -1 +1 @@
-54734
+54045