Enable forcing build from source, and pass through c++ flags

[emersonknapp]

Utilize the variable names proposed in the libcxx colcon mixin colcon/colcon-mixin-repository#16 to be able to force

1. building the package from source
2. using libcxx instead of libstdc++

Related to ros2/ros2#664

[emersonknapp]

@thomas-moulard https://gist.githubusercontent.com/emersonknapp/4aad1eaadcec008d8e4037362b8620b9/raw/584f088d7bbc8390744ae7e378c958d19eed127b/ros2.repos

This should cover everything for a CI build. We should probably specify --cmake-args -DFORCE_BUILD_POCO to properly test the change

[thomas-moulard]

• Linux
• Linux-aarch64
• macOS
• Windows

[emersonknapp]

Pushed new commit on this and other vendor pkgs, plus the colcon mixin

[nuclearsandwich]

This is a full build and partial test that I originally intended to test that things are good when building forcing the vendor builds but because I messed up the option name now it's testing that this doesn't regress the default case of only building when the package isn't found. Covers all the #ros2/ros2#664 vendor packages.

Edit: re-triggered with an updated gist for rclcpp_action

• Linux retest due to infra failure
• Linux-aarch64
• macOS
• Windows

[nuclearsandwich]

This can be merged once green CI comes back for both the standard configuration and with FORCE_BUILD_VENDOR_PKG=ON 🥗

[nuclearsandwich]

CI with clang-libcxx enabled and FORCE_BUILD_VENDOR_PKG=ON

• Linux
• Linux-aarch64

Mac and Windows builds are not expected to work

• macOS
• macOS

[nuclearsandwich]

@emersonknapp that last CI right might actually be overzealousness on my part. Is the libcxx stuff even expected to be functional on Mac and Win as is?

[emersonknapp]

@emersonknapp that last CI right might actually be overzealousness on my part. Is the libcxx stuff even expected to be functional on Mac and Win as is?

My understanding is that we were not targeting Windows or OSX for the clang-libcxx build yet.

Looking into the OSX failure for rviz

[nuclearsandwich]

It looks like the libcxx build is creating failures with class_loader but I haven't identified a cause yet.

[emersonknapp]

Can we disable tests for this build for now? Or, start it out as being unstable because of failed tests? I think we've made a significant milestone in compiling end to end with libcxx

[nuclearsandwich]

Can we disable tests for this build for now? Or, start it out as being unstable because of failed tests? I think we've made a significant milestone in compiling end to end with libcxx

Based on the failing tests, it doesn't seem like class_loader, and therefore pluginlib is functional with the libcxx changes. As long as those aren't needed by what you plan to test then we can avoid running class loader tests.

[emersonknapp]

The first goal is to test compilation - in order to get thread safety compile-time warnings. I think we can work towards making all the tests pass - it would be great to get it going as a nightly that we can start working on fixes for.

[nuclearsandwich]

Re-triggering CI. Full build, default configuration (vendor packages will use system package when available), testing packages above the vendor packages but omitting class_loader and pluginlib for now.

• Linux
• Linux-aarch64
• macOS
• Windows