RMW_DURATION_INFINITE constant - use in QoS API contracts

[emersonknapp]

Resolves #210
Depends on #298 (which resolves #215)

Create explicit RMW_DURATION_INFINITE constant for use in requesting/offering QoS policies with durations, to be used as the default value instead of 0. Enforce using these values in the API contract.

Linked PRs:

• Use RMW_DURATION_INFINITE constant rmw_fastrtps#512
• Use RMW_DURATION_INFINITE constant rmw_cyclonedds#287
• Use RMW_DURATION_INFINITE constant rmw_connext#487

Note: non-linked followups:

• ros2cli (print "Infinite" instead of constant literal value)
• rosbag2 (interpret old implementation-specific "infinite" values and rewrite them as RMW_DURATION_INFINITE, for playback)

Testing

Gist:

• src: https://gist.github.com/emersonknapp/5f998a7e585c8a7bdc1f4845a67bd27a
• raw: https://gist.githubusercontent.com/emersonknapp/5f998a7e585c8a7bdc1f4845a67bd27a/raw/d743e24087109283bb9f6f3ae21f1feae4436b80/ros2.repos

Build:

• Linux
• Linux-aarch64
• macOS
• Windows

[emersonknapp]

Closed by unintentional language in another PR comment - still working on this

[emersonknapp]

I am actively working on this right now (for any interested parties). I've gathered more information to illustrate how the RMW API is currently incomplete.

The following values are gathered from get_publishers_info_by_topic when a single publisher is publishing with unspecified Deadline duration. It doesn't matter which RMW implementation is being used to publish, the value only depends on the implementation of the subscripton - I have tested it with only Cyclone and FastDDS, with all 4 pub/sub combinations. These same values are reported for Deadline, Lifespan, and Liveliness Lease Duration

Value Type	Fast sec	Fast nsec	Fast total nsec	Cyclone sec	Cyclone nsec	Cyclone total nsec
decimal	2147483647	4294967295	2147483651294967295	9223372036	854775807	9223372036854775807
hex	0x7FFFFFFF	0xFFFFFFFF	0x1DCD6500C46535FF	0x225C17D04	0x32F2D7FF	0x7FFFFFFFFFFFFFFF
DDS IDL PSM Constant	DURATION_INFINITE_SEC	TIME_INVALID_SEC	nothing	nothing	nothing	nothing
int limits	INT32_MAX	UINT32_MAX (or -1)	nothing	nothing	nothing	INT64_MAX

"DDS IDL PSM Constant" above refers to the values laid out in Section 2.3.3 of the DDS 1.4 spec. From my understanding this PSM is meant to provide common interfaces for programming languages. There are separate docs available as well:

• C language (Cyclone) - https://www.omg.org/spec/DDS/20140501/dds_dcps.idl
• C++ language (FastDDS) - https://www.omg.org/spec/DDS-PSM-Cxx/1.0/PDF
  The C++ one doesn't lay out new constants, from what I understand is it is meant to replace the IDL-generated API with more idiomatic language

Based on the above information, the two implementations are providing values that are neither consistent with each other or guidelines from the specification - seems like we just have exposed implementation details.

My next step is an updated API proposal, which will come with coupled prototype PRs for the RMW implementations.

Interesting note: the IDL PSM specifies Durations as:

struct Duration_t {
  long sec;
  unsigned long nanosec;
};

But we use exclusively unsigned values within the RMW API.

[emersonknapp]

Closing this in favor of #301 for now