Allow using dyn. reconfig. with -Werror=float-equal

[CodeFinder2]

Hi all,

I am using dynamic_reconfigure with much stricter compiler settings, namely -Werror=float-equal (among others). Unfortunately, this results in errors like:

.../MyPackageConfig.h: In instantiation of ‘void my_namespace::MyPackageConfig::ParamDescription<T>::calcLevel(uint32_t&, const my_namespace::MyPackageConfig&, const my_namespace::MyPackageConfig&) const [with T = double; uint32_t = unsigned int]’:
.../my_package.cpp:371:1:   required from here
.../MyPackageConfig.h:77:28: error: comparing floating point with == or != is unsafe [-Werror=float-equal]
         if (config1.*field != config2.*field)
                            ^
cc1plus: all warnings being treated as errors

One can get rid of these problems in 2 ways:

1. move all *.cfg files to separate packages and
2. "patch" the include directory to generated headers in CMake.

Both solutions are somewhat "hacks". (a) requires to create a new ROS package just for the .cfg file (overhead, file bloat). (b) is hackish in a sense that additional CMake code is required which may also hide other issues.

Therefore I would suggest patching the template to allow such flags. (My first idea would be a specialized version of the template for floats and doubles but there's probably also a smarter / shorter way.)

Any comments on this?

[eric-wieser]

It seems like cmake should be using SYSTEM in its includes, which cause warnings to be ignored in the specified files

[CodeFinder2]

Since I am already using SYSTEM for adding library/system include directories in my project: are you (@eric-wieser ) referring to the CMake code of the dynamic_reconfigure ROS package?

[eric-wieser]

I think I might be, yes.

[ernestmc]

Having the same issue and my workaround is to ignore the warning for the generated header in the following way:

#pragma GCC diagnostic push
#pragma GCC diagnostic ignored "-Wfloat-equal"
#include "PackageConfig.h"
#pragma GCC diagnostic pop

See here for more info on these pragmas.