Runtime Interface Reflection: rosidl

[methylDragon]

Depends (and is based on): #727
This PR is part of the runtime interface reflection subscription feature of REP-2011: ros2/ros2#1374

⚠️ NOTE: This PR has been rebased on the remote mirror of the emersonknapp/type-descripton-runtime branch it builds ontop of to help with reviewing. But when #727 gets merged, this should then retarget rolling

Description

This PR adds utilities for manipulating the rosidl_runtime_c__type_description structs defined in #727

It was migrated (and further modified from): ros2/rcl_interfaces#155

Functionalities:

• Handy print functions
• Hash map construction
• Sorting and pruning of referenced types (to prep them for hashing)
• Checking of type validity
• Appending fields and referenced type descriptions
• Creating TypeDescriptions from referenced type descriptions (description subsetting)
• I also pre-generate the type_description_interfaces sources and headers to create a new package (type_description_interfaces_static) to remove a dependency cycle downstream. (That's why there are so many lines.)

Asan runs green too!

clear && \
colcon build --packages-select rosidl_runtime_c \
  --build-base=build-asan --install-base=install-asan \
  --cmake-args -DOSRF_TESTING_TOOLS_CPP_DISABLE_MEMORY_TOOLS=ON \
               -DINSTALL_EXAMPLES=OFF -DSECURITY=ON --no-warn-unused-cli \
               -DCMAKE_BUILD_TYPE=Debug \
  --mixin asan-gcc && \
colcon test --build-base=build-asan --install-base=install-asan \
  --event-handlers sanitizer_report+ --packages-select rosidl_runtime_c

There are some less important tests waiting to be written (printing), and some tests that are already implicitly covered by other tests (but are not explicit yet.)

TODO

• More minor tests
• 100% docstring coverage

Open Questions

• This only operates on the C version of the type description structs... Do I need to write a C++ version? (Is there a way to convert between the two versions of the message, or are they otherwise interchangeable?)
  • Perhaps that C++ utils library would just have a conversion function, since we know the layout of the struct already???

[wjwwood]

Initial review with a few requested changes, but mostly non-blocking nitpicks.

We're still trying to nail down the double-free issues in some of the tests.

[methylDragon]

I'm not too sure why DCO is kicking up a fuss