dangernoodle-io / breadboard / 30061862079

97%

push

github

fix: bbtool docs scaffold detects host-built components as host:yes (B1-1197) (#1041)

The generated "Platform support" table's host column only checked for a
platform/host/<name>/ backend-dispatch split directory -- a signal that
represents "ships a per-platform-TU host implementation", not "is genuinely
host-supported". A component whose host presence is instead a Kconfig/
define-gated single-source build (e.g. bb_serialize_meta, shipped
unconditionally in the native env via BB_SERIALIZE_META_HOST) read host:no
even at 100% host coverage.

_has_host_tests adds the generalizing signal: presence of a matching
test/test_host/test_<name>[.c|_*.c] file, OR'd into the existing directory
check. Longest-match disambiguation against the discovery SSOT's component
name set (_owning_component) is required because dozens of real sibling
pairs share a name prefix (bb_serialize vs bb_serialize_meta/_json/_console/
_logfmt, bb_wifi vs bb_wifi_http/_ap/_prov, bb_led vs bb_led_anim/_gpio/...,
etc.) -- a naive boundary check alone would misattribute a longer sibling's
test file to the shorter name.

Regenerated 12 READMEs whose host column was wrong: bb_attrs, bb_data,
bb_diag_http, bb_filter, bb_fsm, bb_lifecycle, bb_serialize,
bb_serialize_console, bb_serialize_json, bb_serialize_logfmt,
bb_serialize_meta, bb_wifi_prov.


Claude-Session: https://claude.ai/code/session_01UwuRBPS8n5TkVkog1RQyVF

Co-authored-by: Claude Opus 4.8 <noreply@anthropic.com>

7535 of 8133 branches covered (92.65%)

Branch coverage included in aggregate %.

12813 of 12975 relevant lines covered (98.75%)

2507.28 hits per line