Treat bare YAML nan and inf as strings

[Old-Ding]

Description

rcl_yaml_param_parser explicitly handles the YAML 1.1 dot-prefixed non-finite forms such as .nan and .inf. Its generic strtod() fallback also accepted untagged bare nan, inf, +inf, and -inf, even though YAML treats those forms as strings.

Reject non-finite results only during implicit type inference, leaving both the existing dot-prefixed forms and explicitly tagged values such as !!float nan unchanged. Extend the special-float integration fixture to verify the four implicit bare forms remain strings and the four explicitly tagged forms remain doubles.

Fixes #1320.

Is this user-facing behavior change?

Yes. Untagged bare nan and inf parameter values are now strings, matching YAML and the composable-node parameter loading path. Valid dot-prefixed and explicit !!float non-finite values continue to be parsed as doubles.

Did you use Generative AI?

Yes. OpenAI Codex assisted with root-cause analysis and preparing the focused code and regression test. I reviewed the diff and verification results.

Additional Information

Local verification:

• a native C probe verified implicit bare non-finite values are rejected as floats, explicit-tag mode accepts them, dot-prefixed forms remain floats, and finite values are unchanged
• PyYAML verification confirmed the fixture has 11 string entries and four explicitly tagged non-finite doubles
• source invariants confirm only the explicit YAML_FLOAT_TAG path enables generic non-finite parsing
• all changed lines are at most 100 characters
• UTF-8/CRLF and git diff --check

The full rcl_yaml_param_parser gtest requires ROS 2 build dependencies that are unavailable on this Windows host; repository CI covers the integration test.

[mergify]

Tick the box to add this pull request to the merge queue (same as @mergifyio queue).

• Queue this pull request

[fujitatomoya]

@Old-Ding the fix itself looks good to me.

seems like you are creating bunch of PRs in a few days.
please have a close look at https://discourse.openrobotics.org/t/guidelines-for-ai-assisted-contributions-in-the-ros-project/55903.

and can you provide the test result for this? either log or screen capture?

[fujitatomoya]

@Old-Ding the fix itself looks good to me.

seems like you are creating bunch of PRs in a few days. please have a close look at https://discourse.openrobotics.org/t/guidelines-for-ai-assisted-contributions-in-the-ros-project/55903.

and can you provide the test result for this? either log or screen capture?

@Old-Ding friendly ping.

[Old-Ding]

Verification result for the current head: https://github.com/Old-Ding/rcl/actions/runs/29411533231

I ran the target package in an Ubuntu ROS Rolling container:

• colcon build --packages-select rcl_yaml_param_parser
• colcon test --packages-select rcl_yaml_param_parser
• test_parse_yaml passed, including the updated special_float_point coverage.
• colcon test-result --verbose: 183 tests, 0 errors, 0 failures, 32 skipped.

The full build and test log is linked above.

[Old-Ding]

@fujitatomoya Thank you for the reminder. I have read the AI-assisted contribution guidelines. This PR description discloses the use of Codex, and I reviewed the parser paths and the regression coverage before submitting it.

Verification for the current PR source is here: https://github.com/Old-Ding/rcl/actions/runs/29411533231

• Built and tested rcl_yaml_param_parser in an Ubuntu ROS Rolling container.
• test_parse_yaml passed, including the updated special_float_point case.
• colcon test-result --verbose: 183 tests, 0 errors, 0 failures, 32 skipped.

The workflow branch contains only its verification workflow in addition to PR head 0bcf6ec; the source tree under test is otherwise identical. I will continue to keep submissions narrow and include reproducible focused test evidence with them.