Mojo + NaCl prototype.

mojo/public/cpp/system/handle.h

 // Copyright 2014 The Chromium Authors. All rights reserved.
 // Use of this source code is governed by a BSD-style license that can be
 // found in the LICENSE file.
 
 #ifndef MOJO_PUBLIC_CPP_SYSTEM_HANDLE_H_
 #define MOJO_PUBLIC_CPP_SYSTEM_HANDLE_H_
 
 #include <assert.h>
 #include <limits>
 
@@ skipping 220 matching lines @@
 template <class HandleType>
 inline void Close(ScopedHandleBase<HandleType> /*handle*/) {}
 
 // Most users should typically use |Close()| (above) instead.
 inline MojoResult CloseRaw(Handle handle) {
   return MojoClose(handle.value());
 }
 
 // Strict weak ordering, so that |Handle|s can be used as keys in |std::map|s,
 // etc.
+#ifdef __native_client__
+// Passing the operands by reference confuses the STL implementation in NaCl's

[viettrungluu, 2014/08/20 19:52:38]

Probably we can just make it take |Handle|s by value permanently.

We already mostly pass |Handle|s by value[*], and where we don't it's mostly out
of habit (or misguided sense of consistency, leading to even more inconsistency
of course).

(If we do that, we should probably make |Wait()|, above, just take a |Handle|
instead of a |const Handle&|.)

[*] This should never be less efficient, and sometimes potentially more
efficient.

[Nick Bray (chromium), 2014/09/03 23:45:03]

Done.  The only quirk would be if Handle got a destructor.

+// GCC toolchain and results in specialized templates that break strict aliasing

[Mark Seaborn, 2014/08/26 16:25:54]

I don't think we build with -fstrict-aliasing though, do we?  What's the problem
you see?  A warning or something else?

Is there a GCC/libstdc++ issue filed about this?

[Nick Bray (chromium), 2014/09/03 23:45:03]

No bug filed because NaCl GCC is old and awful, and the only practical fix is to
make it new and less awful.

For posterity:
../mojo/public/cpp/system/handle.h: In member function ‘virtual void
mojo::<unnamed>::CoreCppTest_Basic_Test::TestBody()’:
../mojo/public/cpp/system/handle.h:241: error: dereferencing pointer
‘<anonymous>’ does break strict-aliasing rules [-Wstrict-aliasing]
/Users/ncbray/repos/gitchrome/src/out/Debug/gen/sdk/mac_x86/nacl_x86_newlib/bin/../lib/gcc/x86_64-nacl/4.4.3/../../../../x86_64-nacl/include/c++/4.4.3/bits/stl_tree.h:175:
note: initialized from here

I get this with a vanilla build_nexe.py build.  I believe -Wall -Werror is the
reason it happens.

+// rules.
+// TODO(ncbray): unhack this.
+inline bool operator<(const Handle a, const Handle b) {
+#else
 inline bool operator<(const Handle& a, const Handle& b) {
+#endif
   return a.value() < b.value();
 }
 
 }  // namespace mojo
 
 #endif  // MOJO_PUBLIC_CPP_SYSTEM_HANDLE_H_