Allow character substitution of wasm symbol names (#13477)

* Allow character substitution of wasm symbol names to avoid working around complex string escaping issues in operating system shells and build systems.

* Fix doc.

* Fix doc.

* Add ChangeLog entry

* Fix test_asyncify_tricky_function_sig

Author: juj

ChangeLog.md

@@ -47,6 +47,9 @@ Current Trunk
   of ccache via emsdk, or from https://github.com/juj/ccache/tree/emscripten,
   and run explicitly with "ccache emcc ..." after installing, or automatically
   just with "emcc ..." after activating ccache via emsdk (#13498).
+- Added support to use a custom set of substitution characters . # and ? to
+  ease passing arrays of C symbols on the command line to ASYNCIFY_* settings.
+  (#13477)
 - Using EM_ASM and EM_JS in a side module will now result in an error (since
   this is not implemented yet).  This could effect users were previously
   inadvertently including (but not actually using) EM_ASM or EM_JS functions in

emcc.py

@@ -627,6 +627,15 @@ def is_dash_s_for_emcc(args, i):
   return arg.isidentifier() and arg.isupper()
 
 
+def unmangle_symbols_from_cmdline(symbols):
+  def unmangle(x):
+    return x.replace('.', ' ').replace('#', '&').replace('?', ',')
+
+  if type(symbols) is list:
+    return [unmangle(x) for x in symbols]
+  return unmangle(symbols)
+
+
 def parse_s_args(args):
   settings_changes = []
   for i in range(len(args)):
@@ -1459,6 +1468,10 @@ def default_setting(name, new_default):
                                              '_emscripten_stack_get_end',
                                              '_emscripten_stack_set_limits']
 
+    shared.Settings.ASYNCIFY_ADD = unmangle_symbols_from_cmdline(shared.Settings.ASYNCIFY_ADD)
+    shared.Settings.ASYNCIFY_REMOVE = unmangle_symbols_from_cmdline(shared.Settings.ASYNCIFY_REMOVE)
+    shared.Settings.ASYNCIFY_ONLY = unmangle_symbols_from_cmdline(shared.Settings.ASYNCIFY_ONLY)
+
     # SSEx is implemented on top of SIMD128 instruction set, but do not pass SSE flags to LLVM
     # so it won't think about generating native x86 SSE code.
     newargs = [x for x in newargs if x not in SIMD_INTEL_FEATURE_TOWER and x not in SIMD_NEON_FLAGS]

src/settings.js

@@ -719,6 +719,7 @@ var ASYNCIFY_STACK_SIZE = 4096;
 // to know you got this right), so this is not recommended unless you
 // really know what are doing, and need to optimize every bit of speed
 // and size.
+//
 // The names in this list are names from the WebAssembly Names section. The
 // wasm backend will emit those names in *human-readable* form instead of
 // typical C++ mangling. For example, you should write Struct::func()
@@ -730,7 +731,23 @@ var ASYNCIFY_STACK_SIZE = 4096;
 // changes which would mean a single list couldn't work for both -O0 and -O1
 // builds, etc.). You can inspect the wasm binary to look for the actual names,
 // either directly or using wasm-objdump or wasm-dis, etc.
+//
 // Simple '*' wildcard matching is supported.
+//
+// To avoid dealing with limitations in operating system shells or build system
+// escaping, the following substitutions can be made:
+// - ' ' -> '.',
+// - '&' -> '#',
+// - ',' -> '?'.
+//
+// That is, the function
+//    "foo(char const*, int&)" can be inputted as
+//    "foo(char.const*?.int#)" on the command line instead.
+//
+// Note: Whitespace is part of the function signature! I.e.
+//    "foo(char const *, int &)" will not match "foo(char const*, int&)", and
+// neither would "foo(const char*, int &)".
+//
 // [link]
 var ASYNCIFY_REMOVE = [];
 

tests/browser/test_asyncify_tricky_function_sig.cpp

@@ -0,0 +1,38 @@
+#include <stdio.h>
+#include <emscripten.h>
+
+volatile int x;
+
+int bar() { return 1; }
+
+__attribute__((noinline))
+int foo(const char *str, int &i)
+{
+  if (x == 1337) return bar(); // don't inline me
+  printf("foo %s\n", str);
+  emscripten_sleep(1);
+  printf("foo %d\n", i);
+  return 42;
+}
+
+__attribute__((noinline))
+int foo2()
+{
+  if (x == 1337) return bar(); // don't inline me
+  printf("foo2 1\n");
+  emscripten_sleep(1);
+  printf("foo2 2\n");
+  return 43;
+}
+
+int main()
+{
+  printf("main 1\n");
+  int j = 42;
+  const char *str = "hello";
+  int ret = foo(str, j);
+  printf("main %d\n", ret);
+  int ret2 = foo2();
+  printf("ret2 %d\n", ret2);
+  REPORT_RESULT(ret + ret2);
+}

tests/test_browser.py

@@ -3210,6 +3210,9 @@ def test_async(self):
       print(opts)
       self.btest('browser/async.cpp', '1', args=['-O' + str(opts), '-g2', '-s', 'ASYNCIFY'])
 
+  def test_asyncify_tricky_function_sig(self):
+    self.btest('browser/test_asyncify_tricky_function_sig.cpp', '85', args=['-s', 'ASYNCIFY_ONLY=[foo(char.const*?.int#),foo2(),main,__original_main]', '-s', 'ASYNCIFY=1'])
+
   @requires_threads
   def test_async_in_pthread(self):
     self.btest('browser/async.cpp', '1', args=['-s', 'ASYNCIFY', '-s', 'USE_PTHREADS', '-s', 'PROXY_TO_PTHREAD', '-g'])