AIComfort isn't a distinct thermodynamic operation like Cool/Dry/Heat --
it's an AI-driven overlay on top of the device's own 'Auto' behavior,
confirmed by A-CAWW-TP2-20-COMMON reporting both 'Auto' and 'AIComfort'
as separate, mutually-exclusive entries in /mode/vs/0's supportedModes.
Modeled the idiomatic HA way instead of a flat _DEVICE_TO_HVAC entry:
hvac_mode reports AUTO and a new 'ai_comfort' preset carries the
distinction. Entered/left only via the preset (writes the primary mode
resource, not the convenient one) -- there's no dedicated HVACMode
value for it, so it's not offered in the hvac_mode dropdown directly.
Also adds a once-per-(href, code) warning log when a device-reported
mode has no entry in the relevant map, so a future gap like this one
surfaces in the log instead of silently vanishing -- the exact failure
mode issue #93 called out ("this class of gap is invisible without
diffing against supportedModes").
A-CAWW-TP2-20-COMMON (and likely other CAWW/TP2X-class boards) reports
'AIComfort' as a distinct entry in /mode/vs/0's supportedModes,
alongside 'Auto' (already mapped to HEAT_COOL). _read_modes() silently
drops any code missing from _DEVICE_TO_HVAC, so AIComfort was
unreachable -- one of HA's five other AC hvac_modes, unused by this
device family until now.
The other two gaps in issue #93 are already addressed elsewhere and
not duplicated here:
- Fan-only via the 'Fan' device code, and the WindFree/LongWind/
NanoSleep preset codes, are covered by the still-open PR #91, which
replaces the static preset table with a fully dynamic resolver over
the device's own supportedModes.
- The 'Left_And_Right' -> horizontal swing mapping is already on the
still-open claude/device-support-issue-75-windfree-ac branch.
Since entity.py started routing named descriptors through the catalog
under desc.key, SamsungEntityDescription.name has been read for its
value nowhere -- only twice as a flag, to decide whether an entity was
translated at all. That left 148 English names duplicated between Python
and translations/en.json with nothing keeping them honest: six had
already drifted, invisibly, because editing the Python side changes
nothing a user sees.
So the field is gone, and translation_key defaults to desc.key. A
descriptor now sets translation_key only to share one catalog entry
across descriptors or to point at a differently-named one, and the
catalog is the only place an entity name exists.
Every descriptor resolves to exactly the translation key, icon, entity
category, enabled-default and gating it did before -- with one
deliberate exception: the hood fan, previously the sole descriptor with
no key at all, now resolves to 'fan'. That is inert, because fan.py sets
_attr_name = None so the entity presents as the device itself.
The three helpers that forwarded a name into a descriptor
(laundry.bool_option_switch, washer._bool_option_switch, air_purifier's
sensor table) lose that parameter. test_translations.py now requires a
catalog entry for every descriptor rather than only translated ones.
Claude-Session: https://claude.ai/code/session_01GiibJZZLWVvyxq7mc7EDNp
PR #68 restated its own translation data in Python: a 60-line
TRANSLATED_SELECT_STATES table of frozensets duplicating every
entity.select.*.state key, a second _TRANSLATED_COURSE_TABLES table
naming which course tables have translations, and a strings.json that
was a 835-line byte-for-byte copy of translations/en.json save 43
[%key:...%] references. Each needed hand-syncing, and one was already
drifting.
Home Assistant loads exactly one file per language for a custom
integration -- translations/<lang>.json. It never reads strings.json and
never resolves [%key:...%]; both belong to Core's build tooling, which
custom integrations don't run through (hassfest skips a missing
strings.json and validates translations/en.json instead). So en.json is
the source, and the new catalog.py reads the keys and states back out of
it for the two decisions Python genuinely has to make:
- select._display() normalizes a raw Samsung option to a lowercase
state key only when the catalog knows it, else leaves the vendor's
casing alone. Derived sets are identical to the removed literals.
- laundry.cycle_select() keys off a device-reported course table only
when that table has an entry, else falls back to the name-only
'cycle' key. Translating Table_00 is now a translations-only change.
Also fixes six names that had already drifted between the Python
descriptors and the catalog, restoring HA's sentence case for two
generic ones (Auto release dry, Bubble soak) and taking the catalog's
wording for the rest, and adds a test so the vestigial descriptor names
can't silently disagree with the UI again.
Claude-Session: https://claude.ai/code/session_01GiibJZZLWVvyxq7mc7EDNp
Six fridge SwitchDesc write_fns built their payload with
`'On' if p else 'Off'`. The switch platform passes the literal
string 'Off' on turn-off, which is truthy, so the guard always
produced 'On' -- turning these switches off silently re-sent On
and they could never be turned off:
- ICEMAKER_NIGHTTIME (ice.night.status)
- STATUS_LOCK helper (devicecontrol + device.sound)
- DEFROST_DELAY (delayDefrost)
- WELCOME_LIGHTING (status)
- CABINET_LIGHT dim (light.dimming.status)
- ICEMAKER_STATUS_FALLBACK (iceMaker)
Compare `p == 'On'` instead, matching the pattern the other
capability files already use. Adds a regression test asserting
every affected write_fn sends 'Off' on 'Off' and 'On' on 'On'.
Power users can now pick a resource href from a live dropdown, view its
current value, and POST a minimal patch straight to the device -- to pin
down device-specific write behavior without waiting on a new release.
Bypasses the remote-control block and all write_fn/validate_fn logic by
design; the existing remote-control settings toggle moves behind the same
options-flow menu.
Simplification (feedback: this was overcomplicated): drop the
validated_table gate entirely. cycle_select's table_href now just builds
the translation key directly from whatever course table the device
reports (washer_cycle + Table_02 -> washer_cycle_table_02) instead of
comparing against a hardcoded known-good value and falling back to no key
on any mismatch. A table we haven't shipped translations for yet (e.g.
FlexWash's Table_00) still gets a key built for it -- Home Assistant's own
missing-translation handling takes it from there, the same graceful
fallback already relied on for any individual untranslated code within an
existing table. Adding a newly-confirmed table later is just new
strings.json entries, no code change.
Independent (Opus) review of the prior version caught two real issues,
fixed here regardless of the simplification above:
- translation_key was resolved once at entity construction from whatever
coordinator.last_resources held at that moment. Discovery can run while
a sibling resource is still an empty stub (documented precedent: see
_is_included), so a callable translation_key could permanently bake in
a stale value for the entity's lifetime. Moved resolution into a
translation_key property override (Entity.translation_key is a property
upstream, not a plain attribute), re-evaluated against live coordinator
data on every access, matching how options/current_option already work.
- The supportedOptions fallback's "smallest passing K wins" docstring
claimed every larger passing K is an exact multiple of the true one.
False: the shipped dishwasher fixture has passing K=7 (true) alongside
10, 14, and 35, none of which are multiples of 7 -- position 0 always
lands on the same real course code regardless of K, which alone
satisfies the current-course guard for several unrelated splits.
Corrected the reasoning to what's actually true (an empirically-matched
heuristic across six real dumps, not a proof) and added a regression
test locking in the real dishwasher case so this isn't silently lost.
Course codes on the shared /course/vs/0 contract aren't guaranteed
consistent across board generations: washer/combo devices report course
table Table_02, dryer devices Table_03 (x.com.samsung.da.st.courseTable,
previously fully ignored), and every code in washer_cycle/dryer_cycle was
confirmed exclusively against those. FlexWash's older DA_WM_A51 board
reports Table_00 instead -- applying the same translations there risked
showing a wrong name for any code that happens to numerically collide
between tables, not just an untranslated one.
SelectDesc.translation_key can now be a callable (resources -> key or
None), mirroring the existing pattern for `options`. laundry.cycle_select
gains optional table_href/validated_table params: when given, the renamed
washer_cycle_table_02/dryer_cycle_table_03 keys only apply when the
device's own course table matches exactly -- a different table, or no
table id at all, gets no translation_key (raw code display) rather than
a guess. dishwasher's call site is unchanged (static key, unconditional):
no equivalent table-id resource exists in any dump seen, and no evidence
its course codes vary by table the way washer/dryer's do.
entity.py and select.py resolve a callable translation_key once (via
coordinator.last_resources) and reuse that resolved value everywhere
_display() needs it, rather than re-checking the raw descriptor field.
Some DA_WM_TP1/TP2-class boards populate /wm/editcourse/vs/0 without ever
filling in editCourseList itself (issue #1), so the Cycle select never gets
created even though the device clearly has one (confirmed via SmartThings
app screenshots and a currently-selected course).
/course/vs/0's own x.com.samsung.da.supportedOptions turns out to already
carry the course list, just undocumented: a 1-hex-nibble header followed by
one fixed-width record per course, self-indexed by a course-code first byte
rather than positional like editCourseList. Confirmed against six
independent real-world dumps pulled from open and closed GitHub issues.
cycle_options() now falls back to deriving this when editCourseList is
empty, gated on two checks: the derived codes must all be distinct, and
must include whatever course is currently selected. Larger multiples of
the true record width trivially re-pass both checks too (they're just a
sparser sampling of the same table), so the smallest passing width wins
rather than requiring one unambiguous match.
Also fires on the washer_flexwash fixture, newly creating a Cycle select
there -- unconfirmed against any ground truth for that device (a different,
older board generation with no editCourseList and no screenshots to check
against), flagged for follow-up discussion rather than silently accepted.
Marc Billow