roscpp: Window macro conflict with ERROR constants defined in some msg's. (ros ticket #3817)

[tfoote]

When including windows.h (mingw cross or msvc builds), we get a conflict due to the constant ERROR defined in some msg files - rosgraph_msg/Log.msg and diagnostics_msg/DiagnosticStatus.msg. They both define constants which later conflict with Windows' ERROR macro.

Possible solutions:

1. Change the constant name in DiagnosticStatus. Personally, I hate the idea of changing something because of a stupidly named windows macro!

2. User has to add some inconvenient ifdef/undef code to any file in which they include Log.h or DiagnosticStatus.h. e.g. this is done in roscpp/rosout_appender.cpp:

{{{

ifdef WIN32

#ifdef ERROR
// ach, windows.h polluting everything again,
// clashes with autogenerated rosgraph_msgs/Log.h
#undef ERROR
#endif

endif

}}}

Really annoying for users of some fairly fundamental messages though.

3. Globally killing the ERROR macro in cpp_common/include/ros/platform.h:

{{{

ifdef WIN32

#ifdef _MSC_VER
#define WIN32_LEAN_AND_MEAN // slimmer compile times
#define WINSOCKAPI // stops windows.h from including winsock.h (and lets us include winsock2.h)
#define NOMINMAX // windows c++ pollutes the environment like any factory
#endif
#include <windows.h>
// *********** Add this check ***********
#ifdef ERROR
// ach, windows.h polluting everything again,
// clashes with autogenerated rosgraph_msgs/Log.h
#undef ERROR
#endif

endif

}}}

Of course, could cause problems for windows users that actually use this macro. I don't know anyone or any program who does these days though. Is it worth finding out if anyone will complain noisily about it?

trac data:

• Owner: straszheim
• Reporter: snorri
• Reported at: Wed Jan 18 23:27:20 2012
• URL: https://code.ros.org/trac/ros/ticket/3817

[dirk-thomas]

The mentioned workarounds are sufficient to deal with the issue. Both changing DiagnosticStatus as well as platform.h would have side effects which might break user code.