fix(ros): make the interop gate count interop tests

[YuanYuYuan]

Summary

The interop gate added in #271 does not count interop tests. It asserts a non-zero package total, then reports that total as the interop count. A healthy jazzy run reads 125 tests run: 125 passed, and the gate prints 125 ROS interop tests ran against rmw_zenoh_cpp. Only 41 of those are interop tests. hiroz-tests also holds lifecycle, cache, hu_meter, parameter_tests and other non-interop suites. Deleting every interop test still leaves the gate printing a confident number. #271 removed that defect, and its own fix reintroduced it.

Interop test functions per suite, counted at 47ddff36d2d99cc1bb36e7791d70de0cc51b6424:

suite	tests
pubsub_interop	14
demo_nodes	9
service_interop	7
type_description_interop	4 (absent on Humble)
action_interop	4
parameter_interop	3
total	41 (37 on Humble)

What this PR does

One file changes: scripts/test-ros.nu.

change	effect
Count the interop binaries by name; require each to be present	A run of only non-interop tests fails and names the missing suites, instead of reporting them as interop
Fail when the output contains ros2 CLI not available	A run whose tests skipped or died for a missing ros2 CLI is no longer a pass
Gate the attempt that decided the outcome	The retry path re-ran nextest, but the gate parsed the first attempt's output
Drop the 0 tests run branch	Unreachable: nextest's --no-tests default is to fail
Print a per-suite breakdown	Replaces the single aggregate number

The required suite list is distro-dependent. type_description_interop.rs is #![cfg(not(ros_humble))], because Humble has no type description service to interoperate with. The gate therefore requires that suite on every distro except Humble, and the other five everywhere. It omits python_interop deliberately: the python-interop feature gates that suite, and this command does not enable it.

The self-skip check is not uniform across suites. The shared message is what makes it detectable.

suite	guard	behaviour with no usable ros2
pubsub_interop	check_ros2_available, at two call sites	prints Skipping …: ros2 CLI not available and returns — passes, and counts
demo_nodes, service_interop, parameter_interop, type_description_interop	same helper	panic! with the same substring — already fails the run
action_interop	none	no guard at all

The gate greps for the shared substring, so it catches both spellings.

Note

Reading, not execution, established the 0 tests run branch as unreachable. The hiroz-union invocation in test.yml passes --no-tests=warn precisely because the default fails the job.

Evidence

check	commit	result
The new suite-presence check fires on a genuinely empty suite	8e68984af7abd15e50aea6981085e5cacdc7671c	ROS Tests (clippy-rmw, interop) (humble) failed with ROS interop suites produced no tests: type_description_interop. The old gate had reported that same Humble run as 116 ROS interop tests ran
Full CI on the head commit	47ddff36d2d99cc1bb36e7791d70de0cc51b6424	28 checks green, 0 failed, 0 pending; license/cla success
Gate logic against recorded nextest output	47ddff36d2d99cc1bb36e7791d70de0cc51b6424	see the table below — a manual check, not an automated test

Humble legitimately has no type_description_interop, so the head commit exempts that suite there. That Humble failure is the proof that the detector detects, rather than merely never firing.

Gate logic fed recorded nextest output:

input	result
healthy jazzy output, all six suites	pass — prints the 41-test breakdown
healthy humble output, five suites	pass
84 passing tests, no interop suites	fail — names the missing suites
healthy output plus one ros2 CLI not available line	fail
empty output	fail — no summary line

Breaking changes

tag	what changes	who is affected	before → after	action
BC1	scripts/test-ros.nu fails when interop suites are absent or self-skipped	anyone running the script locally without a working ros2 CLI	silent green → explicit failure	install and source ROS 2 before running, or run cargo nextest directly

No library, API or message-format change. The change is CI-only, and the gate becomes strictly stricter.

Coverage this does not have

These gaps do not block the fix. They bound what green CI proves.

tag	gap	detail
G1	The four Interop tests with ROS 2 <distro> jobs are ungated	.github/workflows/test.yml runs cargo nextest run directly and never calls scripts/test-ros.nu. Only the ROS Tests (clippy-rmw, interop) jobs in ci.yml invoke the gated script
G2	--no-tests=warn keeps one path green by design	The hiroz-union invocation in test.yml warns instead of failing on a zero-test run
G3	The suite list is hand-maintained	A new interop suite is not required until someone adds its name to interop_suites in test-ros.nu
G4	The gate itself is untested	Its behaviour was checked by hand against recorded output, not by a test that runs in CI
G5	Per-suite counts are not asserted	Only presence is required, so a suite that shrinks from 14 tests to 1 still passes
G6	A broken rmw_zenoh_cpp still passes the availability guard	The helper is Command::new("ros2").arg("--version").output().is_ok(), which is true if the process merely spawned. #271 named this failure mode and did not close it. This PR does not close it either