dangernoodle-io / breadboard / 30182039666

97%

push

github

test: convert SSE fidelity tests to golden-file comparison (B1-1220 PR1) (#1090)

- capture six per-topic goldens (log, ota.progress, health.stack, diag.boot,
  health.display, update.available) as k_golden_* literals from real
  bb_openapi_emit_stream() output, following the assertion idiom of the
  existing meta_golden convention (checked-in string literal, asserted via
  TEST_ASSERT_EQUAL_STRING against production output); these six stay
  co-located in this shared test file rather than one-file-per-golden,
  since all six already share one capture helper
- assert_describe_matches_legacy_and_golden() now asserts three equalities
  (legacy==describe, legacy==golden, describe==golden) instead of one, so
  the byte-identical-emission proof survives PR 2's planned deletion of
  bb_openapi_register_topic_schema() and its registry half
- rename the six tests from *_describe_matches_legacy_register_topic_schema_output
  to *_describe_matches_legacy_and_golden to reflect the stronger assertion
- document the golden regeneration procedure in the test file's header
  comment, including the limit of the three-way check: legacy and describe
  are two registration paths but both route through the same shared
  renderer, so agreement proves the paths feed it identically, not that the
  renderer's output is externally correct; regeneration always needs a
  stated reason (a producer's schema literal changed) visible in the same
  diff, and once PR 2 removes the legacy leg, a golden-only diff with no
  such reason is a red flag reviewers should block on


Claude-Session: https://claude.ai/code/session_01GfDDq2y51mr8JYq7VW5qAh

Co-authored-by: Claude Opus 5 <noreply@anthropic.com>

7774 of 8379 branches covered (92.78%)

Branch coverage included in aggregate %.

13225 of 13382 relevant lines covered (98.83%)

2618.23 hits per line