Skip to content

The doctor promises nothing of a reboot that a reboot was not seen to do - #172

Open
AliceLafox wants to merge 2 commits into
mainfrom
fix/doctor-reboot-claims
Open

AliceLafox wants to merge 2 commits into
mainfrom
fix/doctor-reboot-claims

Conversation

@AliceLafox

Copy link
Copy Markdown
Member

Closes #171.

What happened

A rebooted RC-5 still read MEMORY1 / MEMORY2 at 0x25a / 0x259 with differing documents, and a hand-copied PCM file in slot 30 came through that reboot still there and still unindexed (2026-11-05). The doctor had promised that a reboot reconciles the pair and that a reboot would discard the file. Both sentences came down from rc5cat's README: the first was an inference (the pedal heals MEMORY2 from MEMORY1 only when MEMORY2 fails validation), the second an over-reading of PR #45, where the 16-bit take lost its slot at re-index but stayed on the card.

The change

  • One vote for the pair. BankPair / inspectBanks / newestBank in Commands.hpp: the reader, the doctor and the connect line in operations.log take the newest bank from the same rule. A bank whose model keeps an operation closed keeps its vote and loses only its text, so a mutation is refused in the gate's words when the vote names it, never run on the older bank instead.
  • A pair one step apart, either order, is no finding. The warning for a pair more than one step apart stays. Two documents under one write generation, a state neither the app nor the pedal leaves, is a warning with the counter in it.
  • The connect line. PedalSnapshot::banks carries the pair's description from the scan, and the card-up branch of applySnapshot writes it once per card: connect: memory banks 0x25a / 0x259, reading MEMORY1.
  • The doctor judges slots by the bank the reader picks, so a save that landed in the newer bank is no longer "not indexed yet" according to the older one.
  • The non-float finding keeps its facts and loses the reboot clause: slot 30 has LOOPA.wav, which the pedal cannot index: it plays 32-bit float only. Re-push it through LooperCat to convert it.

Doctor sentences about boot, audited

Kept, each with its dated hardware reading in the comment beside it:

  • not indexed yet - reboot the pedal to index it (float32 only): rc5cat 0.1.0, 2026-07-19, README "Boot-time indexing" and the push --no-config story; golden.json's pedal-authored slot params are what that indexing wrote. That the pedal takes float32 only is PR Only float32 goes to the pedal untouched (fixes #44) #45 (2026-08-19).
  • AppleDouble junk (pedal may refuse to boot): rc5cat 0.1.0, 2026-07-19, README "The macOS landmine".
  • trailer is malformed - causes LOOPER DATA READ ERR: rc5cat 0.1.0 (2026-07-19) and 0.7.0 (2026-08-08, rc5cat Disconnect the pedal on app quit #9).

Reworded or dropped:

  • MEMORY1 and MEMORY2 differ ... a reboot reconciles the pair: dropped; the fact goes to operations.log at connect.
  • ... a reboot would discard it, not index it: dropped from the non-float warning.
  • Comments that said the same in Commands.hpp (the reader, restore) and tests/bank_pair_tests.cpp now say what the 2026-11-05 reading showed.

Tests

tests/doctor_claims_tests.cpp, written from the issue: the one-step pair is silent in either order and across the counter wrap; two steps warn either way; the log line names both generations and the bank read, with unreadable / no trailer for a bank that cannot vote; the scan's slots come from the bank the line names; the doctor judges slots by the newest bank; the non-float warning carries no "reboot" or "discard"; the gate falls on the bank the vote names; and a sweep over every kind of finding fails on any sentence about boot that the test does not list with a dated reading, on a listed sentence the doctor no longer says, and on a listed sentence whose comment in the source carries no date. Three mutants against the suite were each caught (the reboot promise restored, a withheld newest bank replaced by the other, the equal-counter warning dropped).

Local: 85 of 85 suites and 8 of 8 harnesses pass on macOS, UTF-8 lint included.

Hardware check to close

Disconnect, power-cycle the pedal, Connect: operations.log should show connect: memory banks 0x25a / 0x259, reading MEMORY1 (or whatever the card carries), no pair banner, and the slot 30 warning without the reboot clause.

Follow-ups, not in this PR

  • rc5cat's README still says "the pair reconciles on the next boot"; the repo this change corrects now contradicts it.
  • A half-written app write (first bank written, second refused) leaves a pair one step apart with the first bank ahead. The doctor cannot tell it from a save on the pedal and says nothing; the failed write's toast and log line are the record, as the old "normal right after a save" banner never was.

A rebooted RC-5 still read MEMORY1 / MEMORY2 at 0x25a / 0x259 with
differing documents, and a hand-copied PCM file sat in slot 30 through
that reboot, still there and still unindexed (2026-11-05). The doctor had
told the player that a reboot reconciles the pair and that a reboot would
discard the file. Both sentences came down from rc5cat's README: the first
was an inference (the pedal heals MEMORY2 from MEMORY1 only when MEMORY2
fails validation), the second an over-reading of PR #45, where the 16-bit
take lost its slot at re-index but stayed on the card.

The pair's vote now lives in one place (BankPair, inspectBanks,
newestBank): the reader, the doctor and the connect line in operations.log
all take the newest bank from it. A pair one step apart, either order, is
no finding; the warning for a pair further apart stays; the generations
and the bank read go to operations.log at connect ("connect: memory banks
0x25a / 0x259, reading MEMORY1"). The doctor judges slots by the bank the
reader picks, so a save that landed in the newer bank is no longer "not
indexed yet" according to the older one. The non-float finding keeps its
facts - the pedal plays 32-bit float only, re-push to convert - and loses
the reboot clause. Every doctor sentence that still says what the pedal
does at boot names its dated hardware reading in a comment, and
doctor_claims_tests lists those sentences with their readings: a new boot
promise fails the suite until it is listed with one, and a listed sentence
the doctor no longer says fails it too.

Issue #171.
…eans one document

Three things the review found. Gating each bank before the vote let a
mutation on a card whose newer bank is a model that keeps it closed fall
back to the older bank and run there; now a refused bank keeps its vote
and loses only its text, so the database is refused in the gate's words
when the vote names it, as reading the newest bank and gating it always
did. Dropping the "differ" finding also silenced two documents under one
write generation, a state neither the app nor the pedal leaves and one the
reader resolves to MEMORY1 without a word; that one is a warning again,
with the counter in it. And the sweep's readings lived only in the test:
it now reads the doctor's source and fails unless a dated reading sits in
the comment above each sentence it lists, so the banner and the comment
cannot drift apart.

Also: the comment claimed the pair is inspected once, but the doctor
inspects it again on its own; the rc5cat reading is of a WAV, not a
float32 WAV, and the float32-only part is PR #45, now cited as such.

Three mutants against the suite, each caught: the reboot promise restored
as an info finding, a withheld newest bank replaced by the other, the
equal-counter warning dropped.
Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

Doctor promises what a reboot does to the memory pair and to a stray WAV; the pedal does neither

1 participant