Repository navigation
Six follow-ups from #42, four of them a statement the code contradicts - #78
Merged
Merged
Conversation
#42 was approved and its hardware suite passed on the 4.0.1 unit. These are the non-blocking findings from that review, applied rather than left to the author. None of them changes what the library does. Two of the six refreshed reasons in UNMARKED_OPERATIONS were not true. The wait_for_listing entry said "scratch_preset teardown uses it to confirm deletion"; wait_for_listing appears nowhere under tests/hardware/, and _release_scratch calls delete_preset and nothing else. The reason it replaced was correct, so it is back. The delete_preset entry claimed the teardown "confirms absence"; the fixture's own comment explains that a listing guard was tried there and removed on purpose, because list_presets is unreliable on a single READ, and delete_preset returns None rather than raising when the unit does not answer. Nothing catches either one: the marker test checks only that a reason is longer than 20 characters. docs/protocol.md carried two undo/redo rows in one table. The maintainer's row was already there and said the operations "land in PR #42"; #42 added a second row beside it instead of correcting it. After the merge a reader gets two answers, and one of them says the work has not landed. writing.md asks a correction to replace the wrong sentence rather than sit next to it, so the two rows are folded into the original's position. Every fact from both survives: the 2026-09-03 4.0.1 measurement, the contributor's 4.1.0 run, the bypass edit reversed and reapplied on disposable preset copies, and empty-history behaviour being unknown. The verification column takes the stronger of the two, "read-back + on-unit". The same stale forward reference in the prose at section 11 now names undo() and redo() as the drivers instead of promising them. Three test doubles kept a request(match=) argument that the real transport no longer has, and two of them implemented its semantics fully. A double wider than the thing it stands for is a trap: a later request(msg, match=...) in client.py would pass the whole offline suite and raise TypeError against a unit. In test_client.py the same edit also set self.last_match on every request() call, which would overwrite the predicate await_broadcast stores there and turn the assertions that read it into a TypeError on None. The read-twice comment in the inhibited_modules hardware test said the second read covers a dropped first READ. It does not: _read_state goes to await_broadcast, which raises TimeoutError with no retry. Reading twice is still worth doing, so the loop stays and the comment now says what it is for. profiles.py starts a sentence lowercase, from the semicolon #42 split. Not fixed here: undo, redo and inhibited_modules are absent from QuadCortex41.VERIFIED while their docstrings claim 4.1.0 confirmation, which is the defect #42 fixed for set_device_name alone. Promoting them needs 4.1 hardware evidence this repository does not have, and that evidence is the contributor's. It is raised on #42 for him to settle. No changelog entry: nothing here is visible to someone who installs the package.
Two findings from /triage-pr on the first commit, both the same defect this pull request exists to remove: a sentence the code contradicts. The merged undo/redo coverage row claimed "read-back + on-unit". The table's own legend defines on-unit as confirmed on the unit's own screen, and no record in this repository says anyone watched the screen. Section 11 says "Measured 2026-09-03 on CorOS 4.0.1 by preset read-back". The undo and redo docstrings say a bypass edit was reversed and reapplied, with no screen reading. test_identity_and_history.py asserts only through read_current_preset(). The fold took the stronger of the two columns as though strength were the question; it is not, the measurement is, and the row it replaced already had it right. Back to "preset read-back", which is the maintainer's own wording. Nothing catches this: no test parses the Verified-by column. The section comment above the preset-library entries still said the scratch_preset fixture exists "but no test uses it yet". #42 landed the test that uses it, and the two entries immediately below the comment both describe that test. Four consecutive lines gave two answers to the same question. Also corrected: "drive it" in section 11 pointed at the 4.1.0 measurement rather than at the messages, because that measurement is the nearest antecedent. Dismissed: the merged row spelling the redo half `{redo: true}` rather than `{UPDATE, redo}` - the action is carried by the first half and this is the pre-existing row's shorthand. Raised for later, not here: nothing pins the test doubles' request() signature against protocol.transport.Transport.request, which is why #42 could widen three doubles and stay green. That wants a source-reading test of its own.
jonathanstokes
force-pushed
the
fix/six-follow-ups-from-42
branch
from
September 22, 2026 17:55
2fd623d to
349c194
Compare
jonathanstokes
marked this pull request as ready for review
September 22, 2026 18:01
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
The six non-blocking findings from the #42 re-review, applied rather than left
to the contributor. #42 was approved and its hardware suite passed on the 4.0.1
unit; none of these blocked it, and none of them changes what the library does.
Rebased onto
mainnow that #42 has merged.What is corrected
Two of the six refreshed
UNMARKED_OPERATIONSreasons were not true.wait_for_listingsaid thescratch_presetteardown uses it to confirmdeletion. It appears nowhere under
tests/hardware/, and_release_scratchcalls
delete_presetand nothing else. The reason it replaced was correct, soit is back.
delete_presetclaimed the teardown confirms absence; the fixture'sown comment explains that a listing guard was tried there and removed on
purpose, and
delete_presetreturnsNonerather than raising when the unitdoes not answer. Neither is caught by anything: the marker test checks only that
a reason is longer than 20 characters.
docs/protocol.mdcarried two undo/redo rows in one table. The maintainer'srow was already there and said the operations "land in PR #42"; #42 added a
second row beside it. After a merge a reader gets two answers and one says the
work has not landed. The two are folded into the original's position, keeping
every fact from both: the 2026-09-03 4.0.1 measurement, the contributor's 4.1.0
run, the bypass edit reversed and reapplied on disposable preset copies, and
empty-history behaviour being unknown. The same stale forward reference in the
section 11 prose now names
undo()andredo()instead of promising them.Three test stand-ins kept a
request(match=)the real transport no longerhas, two with the behaviour implemented in full. The offline suite replaces
Transportwith stand-ins, and a stand-in that accepts more than the real onestops telling the truth: a later
request(msg, match=...)inclient.pywouldpass all 2,968 offline tests and raise
TypeErroragainst a unit. Intest_client.pythe same edit also setself.last_matchon everyrequest(),which would overwrite the predicate
await_broadcaststores there. Nothing inthe repository compares the stand-ins against the real transport, which is why
this stayed green; that gap is now issue #79.
The read-twice comment in the
inhibited_moduleshardware test said thesecond read covers a dropped first READ.
_read_stategoes toawait_broadcast, which raises with no retry. The loop stays; the comment nowsays what it is for.
profiles.pystarts a sentence lowercase, from the semicolon #42 split.Triage
/triage-pron the first commit found two more of the same kind, fixed in349c194. The merged undo/redo row claimed "read-back + on-unit"; the table'slegend defines on-unit as confirmed on the unit's own screen, and no record here
says anyone watched the screen. Section 11, both docstrings and the hardware
test all say read-back. The fold took the stronger of the two columns as though
strength were the question. Back to the maintainer's own "preset read-back". And
the section comment above the preset-library entries still said
scratch_presetexists "but no test uses it yet", which #42 made false, two lines above two
entries describing that very test.
Dismissed there: the merged row spelling the redo half
{redo: true}ratherthan
{UPDATE, redo}, since the action is carried by the first half and this isthe pre-existing row's shorthand.
Not fixed here, and why
undo,redoandinhibited_modulesare absent fromQuadCortex41.VERIFIEDwhile their docstrings claim 4.1.0 confirmation. That is the defect #42 fixed
for
set_device_namealone. Promoting them needs 4.1 hardware evidence thisrepository does not have, and it is the contributor's to give. Raised on #42 for
him to settle.
Validation
Offline suite
2968 passed, 7 skipped; mypy clean across 46 source files;git diff --checkclean.Hardware suite on this exact head,
349c194, Quad Cortex CorOS 4.0.1 / d14e:No changelog entry: nothing here is visible to someone who installs the package.
docs/STEERING.md,CLAUDE.mdanddocs/ADR.mdare untouched. Nothing herechanges a decision, an owned path or an architectural rule; it corrects records
and test stand-ins. ADR-0020 mentions #42 as historical context for the
two-version decision, which stays accurate.
🤖 Generated with Claude Code