Fix assumption on global locale in URDF parser

[traversaro]

Before this fix, the URDF parser was silently failing to parse the decimal part of any real number parsed from XML if the C++ global locale was set to have a decimal separator different from the point.

Related issues in other projects :

• urdfdom usage of std::stod is error-prone to changes in system locale ros/urdfdom#98
• URDF Parsers' code is implicitly assuming that "C" locale is set robotology/idyntree#288
• https://bitbucket.org/osrf/sdformat/issues/60
• Fix assumption on global locale in URDF parser RobotLocomotion/drake#10326

[gergondet]

Thanks for bringing up the issue. This can be confirmed by explicitly setting the locale to one that has comma decimal separator in the test.

However, the attrToList function is only used to convert double lists (such as in the xyz and rpy attributes). For single attribute double value (e.g. value in a mass tag) we use tinyxml2::XMLElement::DoubleAttribute which is unfortunately also locale dependent. I've pushed a patch to your branch.

Note: AppVeyor failure is not related to the PR, I'll work on a patch to fix this then I will merge this

[traversaro]

However, the attrToList function is only used to convert double lists (such as in the xyz and rpy attributes). For single attribute double value (e.g. value in a mass tag) we use tinyxml2::XMLElement::DoubleAttribute which is unfortunately also locale dependent. I've pushed a patch to your branch.

Sorry, I missed those. Just for reference there is a tinyxml2 issue on this: leethomason/tinyxml2#160 .

[gergondet]

Build should be fixed, I will merge this after the CI succeeds. Thanks.