Linking TensorFlow to C++. Protobuf (/usr/local/include/google/protobuf) - c++

Question

System information
Have I written custom code: Custom project
OS Platform and Distribution: macOS Sierra (10.12.6)
TensorFlow installed from: Source
TensorFlow version: Git tagged at 1.3 and master at d27ed9c
Python version: Python 2.7.13 :: Anaconda 4.4.0
Bazel version: 0.5.3_1-homebrew
Protobuf version: 3.3.2-homebrew
CUDA/cuDNN version: N/A
GPU model and memory: N/A
Exact command to reproduce:
Problem description
I compiled TensorFlow source code for Python & C++ API bindings following the steps shown in http://www.blitzblit.com/2017/06/11/creating-tensorflow-c-headers-and-libraries/ (We have to take into account that this tutorial is not updated). After compiling the TF source code I include the TF library in the C++ project:
#include <tensorflow/core/public/session.h>
Finally, I compile the C++ project (Until here everything seems fine). Then, when I try to launch the C++ code I get the following message:
/Applications/CLion.app/Contents/bin/cmake/bin/cmake --build /Users/arcadillanzacarmona/Desktop/FaceSDK/ cmake-build-debug --target FaceSDK -- -j 2
[ 96%] Built target dlib
[ 96%] Building CXX object CMakeFiles/FaceSDK.dir/main.cpp.o
clang: warning: -lcurl: 'linker' input unused [-Wunused-command-line-argument]
In file included from /Users/arcadillanzacarmona/Desktop/FaceSDK/main.cpp:33:
In file included from /usr/local/include/tensorflow/core/public/session.h:22:
In file included from /usr/local/include/tensorflow/core/framework/device_attributes.pb.h:29:
/usr/local/include/google/protobuf/repeated_field.h:328:33: error: expected a qualified name after 'typename'
template <typename It, typename VoidPtr> class RepeatedPtrOverPtrsIterator;
^
/opt/local/include/gif_lib.h:286:17: note: expanded from macro 'VoidPtr'
#define VoidPtr void *
^
In file included from /Users/arcadillanzacarmona/Desktop/FaceSDK/main.cpp:33:
In file included from /usr/local/include/tensorflow/core/public/session.h:22:
In file included from /usr/local/include/tensorflow/core/framework/device_attributes.pb.h:29:
/usr/local/include/google/protobuf/repeated_field.h:328:33: error: expected ',' or '>' in template-parameter-list
/opt/local/include/gif_lib.h:286:17: note: expanded from macro 'VoidPtr'
#define VoidPtr void *
^
In file included from /Users/arcadillanzacarmona/Desktop/FaceSDK/main.cpp:33:
In file included from /usr/local/include/tensorflow/core/public/session.h:22:
In file included from /usr/local/include/tensorflow/core/framework/device_attributes.pb.h:29:
/usr/local/include/google/protobuf/repeated_field.h:864:59: error: template argument for non-type template parameter must be an expression
typedef internal::RepeatedPtrOverPtrsIterator<Element*, void*>
^~~~~
/usr/local/include/google/protobuf/repeated_field.h:328:33: note: template parameter is declared here
template <typename It, typename VoidPtr> class RepeatedPtrOverPtrsIterator;
^
/opt/local/include/gif_lib.h:286:17: note: expanded from macro 'VoidPtr'
#define VoidPtr void *
^
In file included from /Users/arcadillanzacarmona/Desktop/FaceSDK/main.cpp:33:
In file included from /usr/local/include/tensorflow/core/public/session.h:22:
In file included from /usr/local/include/tensorflow/core/framework/device_attributes.pb.h:29:
/usr/local/include/google/protobuf/repeated_field.h:867:55: error: template argument for non-type template parameter must be an expression
const void* const>
^~~~~
/usr/local/include/google/protobuf/repeated_field.h:328:33: note: template parameter is declared here
template <typename It, typename VoidPtr> class RepeatedPtrOverPtrsIterator;
^
/opt/local/include/gif_lib.h:286:17: note: expanded from macro 'VoidPtr'
#define VoidPtr void *
^
In file included from /Users/arcadillanzacarmona/Desktop/FaceSDK/main.cpp:33:
In file included from /usr/local/include/tensorflow/core/public/session.h:22:
In file included from /usr/local/include/tensorflow/core/framework/device_attributes.pb.h:29:
/usr/local/include/google/protobuf/repeated_field.h:2258:38: error: expected a qualified name after 'typename'
template <typename Element, typename VoidPtr>
^
/opt/local/include/gif_lib.h:286:17: note: expanded from macro 'VoidPtr'
#define VoidPtr void *
^
In file included from /Users/arcadillanzacarmona/Desktop/FaceSDK/main.cpp:33:
In file included from /usr/local/include/tensorflow/core/public/session.h:22:
In file included from /usr/local/include/tensorflow/core/framework/device_attributes.pb.h:29:
/usr/local/include/google/protobuf/repeated_field.h:2258:38: error: expected ',' or '>' in template-parameter-list
/opt/local/include/gif_lib.h:286:17: note: expanded from macro 'VoidPtr'
#define VoidPtr void *
^
In file included from /Users/arcadillanzacarmona/Desktop/FaceSDK/main.cpp:33:
In file included from /usr/local/include/tensorflow/core/public/session.h:22:
In file included from /usr/local/include/tensorflow/core/framework/device_attributes.pb.h:29:
/usr/local/include/google/protobuf/repeated_field.h:2262:48: error: template argument for non-type template parameter must be an expression
typedef RepeatedPtrOverPtrsIterator<Element, VoidPtr> iterator;
^~~~~~~
/opt/local/include/gif_lib.h:286:17: note: expanded from macro 'VoidPtr'
#define VoidPtr void *
^~~~~~
/usr/local/include/google/protobuf/repeated_field.h:2258:38: note: template parameter is declared here
template <typename Element, typename VoidPtr>
^
/opt/local/include/gif_lib.h:286:17: note: expanded from macro 'VoidPtr'
#define VoidPtr void *
^
In file included from /Users/arcadillanzacarmona/Desktop/FaceSDK/main.cpp:33:
In file included from /usr/local/include/tensorflow/core/public/session.h:22:
In file included from /usr/local/include/tensorflow/core/framework/device_attributes.pb.h:29:
/usr/local/include/google/protobuf/repeated_field.h:2289:61: error: member reference base type 'const iterator' ( aka 'const int') is not a structure or union
bool operator==(const iterator& x) const { return it_ == x.it_; }
~^~~~
/usr/local/include/google/protobuf/repeated_field.h:2290:61: error: member reference base type 'const iterator' ( aka 'const int') is not a structure or union
bool operator!=(const iterator& x) const { return it_ != x.it_; }
~^~~~
/usr/local/include/google/protobuf/repeated_field.h:2293:59: error: member reference base type 'const iterator' ( aka 'const int') is not a structure or union
bool operator<(const iterator& x) const { return it_ < x.it_; }
~^~~~
/usr/local/include/google/protobuf/repeated_field.h:2294:61: error: member reference base type 'const iterator' ( aka 'const int') is not a structure or union
bool operator<=(const iterator& x) const { return it_ <= x.it_; }
~^~~~
/usr/local/include/google/protobuf/repeated_field.h:2295:59: error: member reference base type 'const iterator' ( aka 'const int') is not a structure or union
bool operator>(const iterator& x) const { return it_ > x.it_; }
~^~~~
/usr/local/include/google/protobuf/repeated_field.h:2296:61: error: member reference base type 'const iterator' ( aka 'const int') is not a structure or union
bool operator>=(const iterator& x) const { return it_ >= x.it_; }
~^~~~
/usr/local/include/google/protobuf/repeated_field.h:2324:70: error: member reference base type 'const iterator' ( aka 'const int') is not a structure or union
difference_type operator-(const iterator& x) const { return it_ - x.it_; }
~^~~~
In file included from /Users/arcadillanzacarmona/Desktop/FaceSDK/main.cpp:33:
In file included from /usr/local/include/tensorflow/core/public/session.h:24:
In file included from /usr/local/include/tensorflow/core/framework/tensor.h:21:
In file included from /usr/local/include/tensorflow/core/framework/allocator.h:26:
In file included from /usr/local/include/tensorflow/core/framework/variant.h:30:
In file included from /usr/local/include/tensorflow/core/platform/mutex.h:31:
/usr/local/include/tensorflow/core/platform/default/mutex.h:25:10: fatal error: 'nsync_cv.h' file not found
#include "nsync_cv.h"
^
15 errors generated.
gmake[3]: *** [CMakeFiles/FaceSDK.dir/build.make:63: CMakeFiles/FaceSDK.dir/main.cpp.o] Error 1
gmake[2]: *** [CMakeFiles/Makefile2:68: CMakeFiles/FaceSDK.dir/all] Error 2
gmake[1]: *** [CMakeFiles/Makefile2:80: CMakeFiles/FaceSDK.dir/rule] Error 2
gmake: *** [Makefile:118: FaceSDK] Error 2
Can anyone help me? Seems that the main problem is in the google/protobuf.

Answer 1

The problem is that gif_lib.h and that protobuf header file are incompatible. More specifically, the problem is that gif_lib.h uses the preprocessor to define a macro called VoidPtr, and then the protobuf header uses VoidPtr as an identifier. The fault firmly lays with gif_lib, not protobuf! Preprocessor macros, by convention, are usually uppercase, should include the library name, and should be avoided entirely if possible (in this case, wouldn't a typedef have done?).
Here are some possible workarounds:
Include the protobuf header (or something that includes it, like a TensorFlow header) before the gif_lib.h one (or, again, something that includes it).
After including gif_lib.h, write #undef VoidPtr on a separate line (but this might cause other mysterious problems, depending on how gif_lib.h is implemented.
Do not #include those both of those two headers from any one of your files; hide them behind your own interfaces.
File a bug with whoever wrote gif_lib and get them to stop using the proprocessor in such a dangerous way.
Use a better designed library in place of gif_lib.

Answer 2

I had the similar issue which is solved by changing compiler and using a different version of C++(i.e. from clang to g++-5, g++-7, etc.). I suggest that you can try to solve it in this way. Sorry for not able to give a direct solution.
For the error about nsync_cv, alc1218's solution on this post may help, which is changing the file mutex.h as follows:
/* Copyright 2015 The TensorFlow Authors. All Rights Reserved.
Licensed under the Apache License, Version 2.0 (the "License");
you may not use this file except in compliance with the License.
You may obtain a copy of the License at
http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software
distributed under the License is distributed on an "AS IS" BASIS,
WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
See the License for the specific language governing permissions and
limitations under the License.
==============================================================================*/
#ifndef TENSORFLOW_PLATFORM_DEFAULT_MUTEX_H_
#define TENSORFLOW_PLATFORM_DEFAULT_MUTEX_H_
// IWYU pragma: private, include "third_party/tensorflow/core/platform/mutex.h"
// IWYU pragma: friend third_party/tensorflow/core/platform/mutex.h
#include <chrono>
#include <condition_variable>
#include <mutex>
#include "tensorflow/core/platform/thread_annotations.h"
namespace tensorflow {
#undef mutex_lock
enum LinkerInitialized { LINKER_INITIALIZED };
// A class that wraps around the std::mutex implementation, only adding an
// additional LinkerInitialized constructor interface.
class LOCKABLE mutex : public std::mutex {
public:
mutex() {}
// The default implementation of std::mutex is safe to use after the linker
// initializations
explicit mutex(LinkerInitialized x) {}
void lock() ACQUIRE() { std::mutex::lock(); }
bool try_lock() EXCLUSIVE_TRYLOCK_FUNCTION(true) {
return std::mutex::try_lock();
};
void unlock() RELEASE() { std::mutex::unlock(); }
};
class SCOPED_LOCKABLE mutex_lock : public std::unique_lock<std::mutex> {
public:
mutex_lock(class mutex& m) ACQUIRE(m) : std::unique_lock<std::mutex>(m) {}
mutex_lock(class mutex& m, std::try_to_lock_t t) ACQUIRE(m)
: std::unique_lock<std::mutex>(m, t) {}
mutex_lock(mutex_lock&& ml) noexcept
: std::unique_lock<std::mutex>(std::move(ml)) {}
~mutex_lock() RELEASE() {}
};
// Catch bug where variable name is omitted, e.g. mutex_lock (mu);
#define mutex_lock(x) static_assert(0, "mutex_lock_decl_missing_var_name");
using std::condition_variable;
inline ConditionResult WaitForMilliseconds(mutex_lock* mu,
condition_variable* cv, int64 ms) {
std::cv_status s = cv->wait_for(*mu, std::chrono::milliseconds(ms));
return (s == std::cv_status::timeout) ? kCond_Timeout : kCond_MaybeNotified;
}
} // namespace tensorflow
#endif // TENSORFLOW_PLATFORM_DEFAULT_MUTEX_H_
or simply find where your "nsync_cv" is and include the whole path.