XML/YAML Front-end

[hidmic]

Feature request

This is a feature request to implement a declarative markup (XML, YAML, TOML, etc.) front-end for launch descriptions. Acceptance criteria is limited to TBD.

Feature description

• Optionally narrow down feature set for a first iteration.
• Draft a format standardization document in https://github.com/ros2/design.
• Gather feedback from potential users (i.e. community and customers).
• Implement parser, from markup based description into LaunchDescription instances.
• Update ros2launch CLI to also understand declarative launchfiles using the parser.

Implementation considerations

ros2launch

Consider refactoring to ease extension through integration of external parsers (using Python modules?).

launch_[markup_fmt]

Weigh different markup formats. Some tentative XML can be found in ros2/design#163. Some draft YAML can be found here.

With the current launch API being already quite declarative, a 1-on-1 mapping may be possible (though this is still a conjecture).

[dirk-thomas]

Also take the existing ROS 1 roslaunch XML into account: http://wiki.ros.org/roslaunch/XML

[hidmic]

As discussed with @mjcarroll and @sloretz, on a first attempt we'll be pursuing an XML format that encompasses functionality equivalent (though adapted to ROS 2 semantics) to that provided by the following ROS 1 roslaunch tags: <node/>, <remap/>, <param/>, <include/>, <group/>.

[hidmic]

A first XML launch format specification can be found in ros2/design#207. Several approaches on how to implement launch front ends (in general, not specific to XML) are described in ros2/design#208.

@wjwwood and I think that ADP offers the best tradeoff between complexity, scalability and extensibility, and thus we'd rather pursue that one. As a summary of what a first Python implementation of such a design would entail, consider the following todo list:

1. Define an abstract launch_frontend.Entity class. (~1 day for 1 FTE) (Add frontend module in launch, launch_xml and launch_yaml packages #226)
2. Implement (and likely refine base API of) a launch_xml.Entity class to wrap XML tags (as represented by any standard xml parsing module). Thoroughly document the mapping from XML schema to object abstraction. (~4 days for 1 FTE) (Add frontend module in launch, launch_xml and launch_yaml packages #226)
3. Implement a launch_frontend.expose_action(*, aliases) decorator for functions or launch.Action subclasses that have a parse(*, entity: launch_frontend.Entity, parser: launch_frontend.Parser) -> launch.Action class method that keeps the mapping from known actions aliases to parsing functions. (~1 day for 1 FTE)
4. Implement a launch_frontend.expose_substitution(*, aliases) decorator for functions or launch.Substitution subclasses that have a parse(*, args: Text, parser: launch_frontend.Parser) -> launch.Substitution class method that keeps the mapping from known substitutions aliases to parsing functions. (~1 day for 1 FTE)
5. Implement a launch_xml.InterpolationSubstitution class to handle substitutions embedded in text. (~2 day for 1 FTE)
6. Define an abstract launch_frontend.Parser class, with methods: (~1 day for 1 FTE)
   1. parse_action(entity: launch_frontend.Entity) -> launch.Action
   2. parse_substitution(value: Text) -> launch.Substitution
   3. parse_description(stream: io.TextIOBase) -> launch.LaunchDescription
7. Implement (and likely refine base API of) a launch_xml.Parser class that makes use of available registries and associated launch_xml.Entity to actually load some XML. (~2 day for 1 FTE)
8. Extend launch_frontend.expose_action and launch_frontend.expose_substitution decorators to inspect the wrapped class constructor annotations, if any, to provide a generic parsing function as a fallback if one's not available. Thoroughly document inference rules (~3 days for 1 FTE)
9. Implement parsing functions for all actions and substitution for which said functions cannot be reasonably inferred. (~3 days for 1 FTE)
10. Integrate launch_xml.Parser to launch source loading actions/functions. (~3 days for 1 FTE)

That is, roughly (and I want to emphasize that qualifier) three (3) weeks worth of an FTE work are required (including reviews if running at a good pace). We may be able to take some shortcuts, though I'm not sure we can radically reduce the sheer amount of man power needed. @mjcarroll @sloretz @ivanpauno thoughts?