| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
Nine tasks, TDD throughout. The ordering mutation in task 4 is the
load-bearing check: indexing the new path before removing the old one is
what preserves a message's tags, and the wrong order passes every other
test while silently destroying them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 103's measurement is done, so its section carries the finding and the
three constraints that decide whether to open the spec, rather than the
design inline.
Item 118 is blocked on 103 and is the first action that would destroy mail
with no undo, which is why it is filed separately rather than folded in.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Measures what Delete does today (a notmuch tag and nothing else, verified
against the tag-to-flag table and a probe on a throwaway database) and
specifies the move-to-trash behaviour that replaces it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
The reconciliation against the user's own notes found ZERO unrecorded
entries, the first clean pass: the 2026-08-16 sweep added items 98 to 104 and
those absorbed the whole current "Not done yet" list.
Closed this session, sections moved to the closed-items file: 98, 100, 102.
Dropped:
- 78, at the user's request. Never a defect. Items 85, 23 and 81 already give
the whole journey (right-click a value, search it, save the query, make a
rule from it); this was only a shortcut across it, and the entry had already
said to gather usage evidence first. That evidence never appeared.
- 116, the same day it was raised, and its section is kept for the process
failure rather than the non-bug. Copy image was reported as copying markup
instead of pixels. Two explanations were eliminated by real evidence, and
the conclusion drawn was that something more interesting must be wrong; the
actual answer was that the measurement distinguishing them was broken. A
wl-paste reading taken minutes after the copy showed text flavours only, was
explicitly labelled unreliable in the entry, and was then reasoned from
anyway. Run immediately after a copy it reports image/png and 30 more, and
pasting into GIMP immediately works. A caveat that does not stop the
reasoning it qualifies is decoration.
Opened:
- 112, Toggle unread on a whole thread cannot reach "all unread" on a
partly-read thread. The direction comes from notmuch's UNION over the
thread, so one unread message anywhere makes the action pick "mark read" and
no input reaches the other branch. Third defect from that union after 110.
- 113, view source as our own plain-text dialog.
- 114, Save image is offered and does nothing: no downloadRequested handler
exists anywhere. The user corrected the first proposal, which would have
refused remote images on security grounds; once remote content is granted
the bytes are already fetched, so saving them is a local copy and blocking
it protects nothing.
- 115, no confirmation when a copy succeeds.
- 117, the pane offers no Select all. NOT caused by item 100: verified against
a build with that filter reverted. Three wrong theories preceded that
measurement, and the lesson is one item 100 had already written down: a menu
built by hand proves nothing about the menu Chromium builds.
The changelog's Unreleased section gains Important-as-a-toggle, the rules
Note column, the menu fix, and two Upgrading notes.
|
|
A thread's card has rendered one message since item 66, but every tag
action still acted on the entire conversation. Delete, Archive,
Important, Mark spam and Toggle unread now act on the message the card
shows; the whole-thread versions move to a "Whole thread" submenu in the
Message menu and the thread list's context menu, on Ctrl+Alt+<key>.
Closes items 87, 88, 105, 106, 107, 108, 109, 110 and 111.
The defects fixed along the way, several found by reading rather than by
report:
- threadAt(current.row()) answered about the wrong thread for a reply
row, because a tree numbers rows per parent. The audit found four live
sites, not the one reported: Delete and Toggle unread each chose their
DIRECTION from an unrelated thread, and the tag dialog counted the
wrong thread's tags. threadFor(index) replaces them.
- A message-scoped write made no optimistic model update and no reply
row carried a doomed cue, so acting on a reply moved the pending-edit
count and changed nothing on screen.
- Both toggles read the state of a reply's THREAD, which a
message-scoped write never changes, so they were one-way: the second
press re-sent a tag the message already had.
- flushHeldEdits() re-sent only thread-scoped edits, so a tag change
made on one message during a sync was applied to the row, counted as
unsynced, and then dropped without ever being written.
- applyTagChange() updated a thread's summary but not its loaded
replies, leaving an expanded thread's rows describing a state the
database no longer held.
- A thread's first message is not among its children, so both
message-scoped lookups missed it: acting on a root card repainted
nothing and emptied the message pane's chip row.
- ThreadSummary::tags is notmuch's union over the thread, so a card
standing for one message drew tags belonging to its siblings. The
worker now reads that message's own tags in the walk that already
finds its id, so the split is known before a row is ever opened.
The card shows both tiers: its own message's tags at full size, the rest
of the conversation's smaller and muted, so nothing appears to vanish
when a row is selected.
Auto mark-read is message-scoped as a result, and now arms for a reply,
which it never did. With maildir.synchronize_flags on, the old
thread-wide write reached the server for mail that had never been
displayed.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Double-clicking any row drills into its thread: the list becomes that thread
alone, expanded, and the pane shows the double-clicked row's own message. A
reply therefore opens its WHOLE thread with itself selected, never itself alone,
which is what the user asked for and is not the obvious reading of "open it by
itself".
This is recoverStaleThread() triggered by a gesture. That function already ran
thread:<id>, expanded the thread when the row arrived, selected the target
message once the replies landed, and fell back to the root when the message had
gone; all three cases are existing paths through it, so the new code resolves a
row to a thread id and a message id and hands both over.
The row is reached through the INDEX and never through index.row(): a tree
numbers rows per parent, so threadAt(row) on a reply answers about an unrelated
thread. That is item 88's trap, avoided here by construction.
The first click of a double-click arms the mark-read timer, and the handler
cancels it, because a gesture that navigates must not mutate mail. The timer is
armed again for whichever row the recovery lands on, so only the arming for the
row being left is cancelled. Its test asserts the timer was active beforehand,
so it cannot pass by the timer never having been armed at all.
The expander keeps its own double-click: ThreadListView::mousePressEvent accepts
a press inside its rect and returns, so Qt never pairs one into a double-click
there.
Nothing is built for getting back. The filter buttons already are that, per the
user.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
The entry said a double-click runs a query naming the row, an id: for a message
and a thread: for a thread. That is wrong for a reply, which the user wants to
drill to its THREAD with itself selected, not to itself alone. The view is
always the whole thread, expanded; only which message the pane shows changes.
It also assumed a thread: query would show the conversation. Nothing in the tree
auto-expands, so it lands on one collapsed card and the replies still need a
click.
Both are already solved by recoverStaleThread(), which runs the query, expands
the thread, selects the target reply when the replies arrive, and falls back to
the thread when the message has gone. Item 91 is that mechanism triggered by a
gesture rather than by the stale-thread notice, so the approach is to reuse it
rather than write a second selection-after-query path.
Recorded alongside: a double-click delivers a single click first, which arms the
mark-read timer, so the handler must cancel it rather than marking a message
read that the user only passed through.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
91 needed one decision, how the user leaves a drill-down, and the answer is that
nothing is built for it: "I didn't think of a back action, usually I'd go back
to a known list like unread or inbox at that point." The filter buttons already
are that and are one click away in every view, so the Back action, the history
stack and the restore-the-previous-query scheme are all unnecessary. That also
retires the undo-stack concern: a drill-down clears the stack exactly as a typed
query does, which is the behaviour the user already expects from the query bar.
The item is now fully specified and ready to build at S.
92 is postponed at the user's request: "I don't see the utility, so I don't
really know how to answer." The clarification that preceded it named the
per-message version, which needs provenance nothing records, costs a format
change across both repos against a hook running on real mail every ten minutes,
and stays blank on every message already tagged. A feature whose requester
cannot say what it is for should not be built. The cheaper substitute is
recorded beside it for if the question ever turns out to be "why does this
message carry this tag", which rules.json can answer at read time with nothing
stored.
Section moved to the closed-items file on this commit, per CLAUDE.md.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Both were "open, unspecified" and could not be planned from the backlog alone.
91 is not a second window. The user meant a drill-down to the selected thread
or message, the same result an `id:` query gives, reached by double-clicking
instead of typing. That removes the constraint the entry was mostly made of,
since no second QWebEngineView and no extra render process are involved, and
takes it from ? to S. What is left to decide is how the user gets back, and
that runQuery() clears the undo stack, which a gesture-triggered query would do
silently.
92 is the expensive half of the two it held: the user wants a per-message hint,
rule-written against hand-applied, and doubts its utility in the same sentence.
Nothing records that today, so it is a format change across both repos before
it is any pixels, and it stays blank on every message already tagged. Recorded
beside it is the cheaper question that may be the real one, "which rule would
tag this message", answerable from rules.json at read time with no stored
provenance and no hook change.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
An edit made while a sync is running is held rather than sent, because the
worker's read-write open blocks on notmuch's exclusive lock. At sync end
onExternalSyncStateChanged() refreshed the list first and flushed the held
edits afterwards, so the refresh read a database that still carried the old
tag, reconciled it into the model, and overwrote the optimistic update the hold
had deliberately left applied. The flush then wrote the tag correctly.
The database ended up right and the list ended up wrong, with nothing scheduled
to re-read it, which is why it looked like the edit had been lost. Reported by
hand: a message read during a sync went back to unread when the sync finished.
The flush moves ahead of the refresh and keeps both properties it already had.
It stays outside the Idle branch, so edits held when /proc/locks becomes
unreadable are not stranded waiting for an Idle that never comes, and it stays
after the status-bar retire, so its own "N held changes sent" message survives.
Both orders leave identical end state, so the first version of the test passed
against the defect: after the handler returns the queue is empty and the write
has been sent whichever ran first. flushGenerationForTesting() stamps the query
generation at flush time, which is what separates them, and the test fails
against the old order with Actual: 3, Expected: 2.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
runAutoSync() returned without rescheduling when a sync was already in flight.
The comment defending it argued the edits were not lost, because they reached
the mail store at edit time and the running sync was "very likely" to carry
them. Very likely is not always: an edit made after mbsync has already passed
that account's mailbox is not carried by it, the timer had fired, nothing
re-armed it, and the pending count sat non-zero until a manual sync or the next
cron run.
Skipping is unchanged and still required by item 71: the cron job holds the same
lock and mbsync fails on a second concurrent run. What changes is that the skip
schedules another attempt. scheduleAutoSync() re-checks the delay, the sync
command and the pending count on the way in, so this cannot arm a sync for
nothing, and against a long external sync it re-arms once per debounce interval,
which is a timer rather than a sync.
The test fires the timer by hand and asserts it is active again afterwards, at
the configured interval rather than a shorter one, with the pending indicator
still showing. It fails against the old skip path.
Item 89's other half is dropped rather than built. The list churn it described
is a tag-defined view working as intended: a thread that loses `unread` leaves
the Unread view, and the user resolved it by living in the Inbox view instead.
Three designs were drafted before asking and none is worth building.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Running a query blanks the message pane but left m_currentThreadId,
m_currentMessageId and m_currentMessageThreadId naming the thread that had
been showing. Both selection handlers compare a newly selected row against
those to decide whether it is already on display, so a result containing that
same thread was recognised as "already showing" and onThreadSelected() was
never called. The card painted as selected, the status bar reported one
thread, and the pane stayed on the placeholder.
This is why it looked like an `id:` query defect. The id is copied out of the
details dialog of the message being read, so that thread is current at the
moment the query replaces the view. Any query returning a different thread
hides the fault entirely.
Filed as the unverified half of item 66 and assumed to be the same
empty-MessageIdRole failure. It is not: 66's fix was correct and this
reproduced against it, so it is recorded as item 96. Four hypotheses were
eliminated by measurement first: the row does carry the message id, the
account-scoped query does return it, MimeParser parses the reported message
(ok, 40701 bytes of HTML), and both real ids resolve bare and quoted.
The regression test's first query must open the SAME thread the second one
returns; with two different threads it passes against the defect, which is how
the first version of it was green. Reverting the fix fails it.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Nothing loaded a translation before this: no QTranslator, no .ts file and
no build rule, so every string was English whatever the locale said. The
language now comes from the environment, LANG=it_IT.UTF-8, and any other
locale runs in English as before.
The audit found that the tr() discipline was largely holding, and found
eight strings that could never be translated into any language. kFields[]
in tagrulesdialog.cpp declared the rule-builder field labels with
QT_TR_NOOP inside an anonymous namespace, where lupdate reports "tr()
cannot be called without context" and extracts nothing, while the use site
calls TagRulesDialog::tr() on them at runtime. From, To, Cc, Subject, Tag,
Folder, Attachment and Date: the whole vocabulary of the rule builder,
absent from every translation file that could ever exist. The source
compiles and reads correctly; only lupdate reveals it.
Q_DECLARE_TR_FUNCTIONS is not the fix for that case, though it is the fix
for a free function calling tr(). Measured against lupdate: a class
carrying the macro beside the array still extracts 0 strings, because the
context must be attached to the literal itself. QT_TRANSLATE_NOOP names it
explicitly and matches the tr() that already reads them, so the use site
needed no change.
Twenty configuration and keybinding warnings were not translatable either.
They are user-facing, reaching the status label and the "Configuration
problems" dialog. Config already had the tr() macro; KeyMap needed it.
Translating the filter labels then broke startup_query, found in hand
testing: a filter's name is a translated label, so `startup_query = Inbox`
matched nothing where the filter shows as "In arrivo". The application
opened a different view and reported the user's own working config as
invalid. Resolution matches the generator as well now, which is stored in
queries.json and identical in every locale; the translated name still
works. The regression test installs a real QTranslator rather than a stub,
since the bug lives in the gap between the stored string and the displayed
one, and it writes a queries.json because the warning it asserts on is
guarded by a non-empty saved-query list: without one the branch never runs
and the test passes against a broken check.
main.cpp's --help and --version stay bare printf, as they run before
QApplication exists and no translator could serve them.
Verified per the backlog's own standard, that lupdate output is the
evidence rather than reading: 355 strings extracted with zero context
warnings, where before there were 327 with eight; lrelease reporting 355
finished and 0 unfinished; the built .qm loaded in a standalone probe
printing "From -> Da" and both Italian plural forms; and the install rule
placing it where main.cpp looks. test_translations guards it and was
mutation checked, failing on an emptied translation and naming the defect
when QT_TRANSLATE_NOOP is reverted.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 93 ships four built-in filters composing with the account dropdown, and
absorbs item 90. Item 95 is the overflow-menu defect found while hand testing
it: pre-existing, not caused by 93, and recorded as its own item rather than
folded in.
Two traps go to CLAUDE.md because they are still true of the code. Qt emits no
triggered for a QAction owning a menu, which is why every entry in the saved
query menu had always been inert. And a generator must be asked for one
account's query rather than having its all-accounts query wrapped, since the
wrap returns the right rows by accident of path: being hierarchical and a
row-count test cannot tell the two apart.
A third goes beside the existing rendering-probe warnings: visualRect reports a
real height for a row scrolled out of the viewport, so a non-empty check passes
while the pixel loop measures nothing and blames the wrong defect.
The Upgrading note covers what a user sees: the row starts with four shipped
buttons, a duplicate name means one of theirs is now beside a built-in, and
their migrated Sent entry is unpinned for them rather than deleted.
|
|
The user's end state for the query row is built-in filters only, with every
saved query living in the menu. At that point SavedQuery::pinned has nothing
left to decide.
Blocked on 93 and deliberately separate from it: the four buttons have to be
lived with first, and if one of them is wrong, pinning is the escape hatch,
which has to still exist to be used.
Recorded as a user-visible removal rather than a cleanup. pinned shipped in
0.18.0 as a checkbox in SaveQueryDialog and a right-click action, so removing
it is a minor bump with an Upgrading note. The stored field is a separate
decision from the UI, and leaving it in queries.json unread is both cheaper
and reversible.
The spec for 93 gains the ordering rule this resolves: filters first in fixed
order, the user's pinned queries after them, and nothing configurable, since
the mixed row exists only until 94 lands.
|
|
Explaining item 90 to the user produced a reframing rather than a fix. The
buttons and the "more queries" menu are two different kinds of thing sharing
one mechanism: a filter narrows whatever the user is looking at and should
compose with the account dropdown, while a saved query is a self-contained
destination entitled to set the account itself.
Nothing ships as a default today, so the buttons are whatever the user pinned,
which the queries.json migration did to every [queries] entry. That drift is
the defect.
Item 93 ships four built-in filters, Unread, Inbox, Flagged and Sent, as
generated entries in the closed kQueryGenerators set that already exists for
Sent. The user's own pinned queries are unpinned rather than deleted once the
buttons are confirmed working.
Three findings from reading the code, all in the spec. A generator must answer
per account rather than having its all-accounts query wrapped in a scope, or
Sent becomes path:"a/**" and (path:"a/Sent/**" or path:"b/Sent/**"), which
returns the right rows only because path: is hierarchical. Sent is flat and
the other three are not, so the four match in scope and not in view mode. And
m_accountBox has no signal connected to it, which is now a decision rather
than an omission: changing the account runs nothing, the button is the verb.
Item 90's section moves to the closed file, kept in full because its cause and
the rules preview that motivated the reset are still true of the code.
|
|
Item 74. runQuery() set the status bar once and only queryFinished cleared
it, so the bar kept claiming a query was running for the whole walk while
rows were visibly arriving behind it. Measured cold against a 1.1 GB index:
the first batch reaches the model at 642 ms and the walk finishes at 5714 ms,
so five seconds of a slow query read as a frozen one.
onThreadsReady now sets the bar from the model's own row count after each
batch, which is the number of rows the user can actually see. No timing
changes; this only stops the bar from lying.
The refresh branch returns before the new line, so a background refresh stays
silent exactly as onQueryFinished already keeps it. That silence has its own
test, which fails when the write is moved above the guard.
beginRefreshForTesting() is a new seam: refreshCurrentQuery() returns early
without a worker and a bare window has none, so a test cannot otherwise reach
the refresh path.
|
|
The 2026-08-15 pass over ~/Documents/Obsidian/note/notes on qtmaildir.md
found four entries with no item here, and one already-closed item with a
case that was never verified by hand.
89: runAutoSync() returns without re-arming the timer when a sync is
already in flight, so an edit made after the running mbsync passed that
account waits for a manual sync or cron. The second half of the same
complaint, rows leaving the list mid-interaction, is a design question
and is recorded as one rather than guessed at.
90: runSavedQuery() clears the account box for a query that names no
account. The line is deliberate and its comment says why, so this needs
a decision rather than a fix.
91 and 92 are unspecified, as the notes themselves say.
Item 66's row gains the one case its fix should already cover: a
single-message id: query whose card would not open is the same empty
MessageIdRole failure, unverified against an id: query.
|
|
Item 66 turns out not to have been the defect it was filed as. The pane
was never blank: an unexpanded thread root rendered the CONVERSATION,
and the same click rendered one message once the thread had been opened,
because the model learned the root's message id only when the replies
arrived. The user's step-by-step account is what separated the two
halves; two probes against a real database had failed to reproduce the
blank pane because there was none.
Closed by carrying firstMessageId in the query and removing the
conversation view, which the user asked for after being told the stubs
not expanding was itself a defect and that the feature was being judged
in a broken state.
Two defects came out of it and are open. 87: auto mark-read still marks
a whole thread, coherent while a root rendered the conversation and not
any more. 88: threadAt(current.row()) answers about the wrong thread for
a reply row, because a tree numbers rows per parent. 87 is blocked on 88
and the entry says why: a fix for 87 was written, mutation-checked,
shipped and reverted the same evening after it marked an unrelated
message read.
CLAUDE.md gains the row-number trap as its own entry rather than leaving
it implied by the item 20 note, plus the rule that a test for a write
path must exercise the reply case: the reverted fix was green because it
asserted on a root selection, the one case where row() is correct. The
cid-prefixing note is corrected to say every caller now passes one
message while explaining why the prefixing stays.
The changelog carries a Removed entry and an Upgrading note, including
that mark_read_delay_ms accepts a negative value to disable auto
mark-read entirely, verified against config.h.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 36's section moves to the closed file and its three traps go into
CLAUDE.md, which is where they will be read: the worker is unreachable
by findChild, rowCount on an unexpanded thread row is 0 by design, and
currentThreadId reports intent rather than content.
The claim that this class of bug cannot be reproduced in test_mainwindow
is now false, so it is corrected rather than left standing beside its
replacement. One in-test reference to item 36 as a permanent limitation
is reworded: bare-window cases still have no worker, but that is now a
choice per case rather than a property of the binary.
Item 66 stays open with the simple case ruled out. The negative result
sharpens this entry's own candidate rather than contradicting it: the
test drives setCurrentIndex, which updates the selection model
synchronously, while the suspect guard turns on a real click not having
done so yet. Two cheaper conditions are named as still unexcluded.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 84's section moves to the closed file, recording that the fix was
a split rather than a suppression, and that its mutation check
reproduces the original hang at exit 124 rather than merely failing.
Item 36 is specced and no longer "on demand": item 66 needs it. Two
findings while writing it. The deliverable is a RED reproduction of 66,
with the fix deliberately excluded, since that defect has never been
isolated and designing a fix beside a hypothesis is how a wrong one gets
locked in. And the item is smaller than it has read since 2026-08-04:
wireWorker() already builds the worker from a config key, so a test
writes a config pointing at the fixture and nothing in src/ changes.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Section moved to the closed-items file on the commit that closes it,
per the backlog's own rule. It records the two decisions that are not
recoverable from the code, and that the plan under-counted the
signature change: three test files drive these signals, not one.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Ordered so each task leaves the suite green: the grammar first, then
the SearchMode enum alone, then the four signatures as one mechanical
change with no behaviour difference, then the guard, then the menus.
Two traps carried from item 85 into the task that inherits them. The
details dialog's mutation check hangs rather than fails, and a hung
binary is re-run stale by a later ctest. The mutation revert is by hand,
never from a backup file, which silently reverted a fix last session.
Construction order verified rather than left to the implementer:
registerActions() runs after buildUi(), so both pointers are live and
no null guard belongs in the lambda.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Two decisions the user made, both recorded with their rejected
alternatives so they are not revisited.
Excluding from an empty query would mean the whole Maildir minus one
value. The menu entry is greyed rather than hidden when the query bar is
empty, so the feature stays visible to someone exploring a fresh window,
and SearchTerm::exclude returns empty for that case as a second layer
against a caller that forgets the guard.
The menus cannot see the query bar, so MainWindow pushes the fact down
through MessageView::setHasQuery from the textChanged lambda it already
runs for the Save button. A callback was rejected as an indirection with
one implementation; silently doing nothing was rejected because a live
menu entry that does nothing is worse than a greyed one.
The backlog entry loses its inline approach and points at the spec,
carrying the three constraints that decide whether it can be picked up.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
The 2026-08-14 reconciliation against the user's notes found one entry
with no item here: the right-click search offers "search for this" and
"add to search" but no way to add negatively.
Cause verified in the code rather than copied from the note. Item 85
shipped the two operations as a single bool, built identically in
messageview.cpp:566 and messagedetailsdialog.cpp:92 and branched on in
mainwindow.cpp:1654, and SearchTerm has no exclusion form at all. So
this is not a missing menu entry over an existing capability; there is
no third state for an entry to select.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Five surfaces in the message pane offer a search built from what they show,
replacing the query or narrowing it. The details dialog became rows along the
way, which the user wanted independently of this feature.
Item 78 is narrowed to the rule shortcut alone and drops to S: item 85 built
the menus and item 81 the seeded dialog, so both halves already exist. Its
approach text is corrected too, since it claimed the thread list holds a usable
sender and notmuch_thread_get_authors returns a display summary, not an address.
Three traps recorded in CLAUDE.md: a modal dialog must close before the action
it asked for runs, Qt::RFC2822Date validates the weekday against the date, and
every query goes through SearchTerm so five surfaces cannot grow five quoting
rules.
|
|
Qt::RFC2822Date validates the weekday against the date, so 'Thu, 14 Aug 2026'
parses as invalid: that day is a Friday. Task 2 hit it and Task 6 carried the
same wrong data. The failure is indistinguishable from the timezone-comment
trap the date parse exists to handle, so the plan now names it.
|
|
Eight TDD tasks. SearchTerm and the date parse first, since both are pure
logic testable without a widget; then one surface per task, the window's
wiring, and the suite check.
Two findings while planning, both now in the spec. The Date: header parse
already exists inside a file-local function in mimeparser.cpp, complete with
the fix for Qt::RFC2822Date rejecting a trailing timezone comment, so it is
extracted rather than rewritten. And queryRequested is the right precedent
but the wrong signature: these actions carry a replace-or-narrow flag, so a
second signal sits beside it and the placeholder links keep their gate.
|
|
Item 78 asked for a tagging rule built from something visible in a message.
Brainstorming narrowed it: a saved query can already be promoted to a rule,
so the road from "I see something interesting" to "a rule tags it" exists as
search, save, promote. Searching is the missing step and the safe one, since
a query costs nothing when it is wrong while a rule runs unattended against
real mail.
The search half splits out as item 85. Five surfaces gain a context menu with
Search for this and Add to search: subject, date and From/To/Cc in the header,
tag chips, body selection, and every header per message in the details dialog,
which is rebuilt as rows rather than one text box.
Item 78 stays open carrying the rule shortcut alone.
|
|
Cause verified by attaching gdb to the hung process rather than
inferred: showWarnings() raises QMessageBox::warning from the MainWindow
constructor, and nothing offscreen can dismiss it, so any config problem
in a test's fixture blocks the constructor forever.
Not a defect in the application. The modal is right for a person and the
code says why; the defect is that a test cannot dismiss it and the
resulting failure is a silent hang rather than an error naming the
cause.
Corrects item 81's closing note, which blamed the missing maildir key
itself.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Also records item 83's fix and the warning banner in the changelog,
which the earlier commit did not.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Six tasks, TDD, one commit each. Two things the spec did not anticipate
and the plan has to carry: the rules dialog is non-modal and
single-instance, so a second request must seed the dialog already open
rather than being dropped, and the seeded id is uniqued against the
working list rather than the file, since that list can hold unsaved
rules whose ids collide just as hard.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 81. A context-menu action on a stored saved query, seeding the rules
dialog with the query and a sanitised id, tags left empty and focused.
Three decisions worth the record. The backlog's own proposal, a checkbox
in the Save query dialog, is rejected: it would make one dialog write both
queries.json and the shared rules.json, and SaveQueryDialog is
deliberately pure UI that writes nothing. Generated entries are excluded,
since their query is composed from the accounts at runtime and a rule made
from one would freeze a snapshot that goes stale when an account is added.
And the empty tags are load-bearing rather than an omission: validate()
refuses a rule that tags nothing, so the one field the user must supply is
the one the dialog opens on.
This also turns out to be a single-repo change. The rule it creates is an
ordinary one, so mailrules.py is untouched; the backlog's note that item
81 spans two repos was about the file it lands in, not the work.
Item 78 becomes a second caller of the same seeded-dialog path.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
A rule named "justeat orders" in the field labelled Name was written to
rules.json correctly and then dropped by every reader, because load()
required ^[a-z0-9][a-z0-9-]*$ and the save path validated nothing. The
rule stayed in the file, invisible in the dialog, never applied by the
post-new hook, and the next save from the dialog would have deleted it
outright.
The asymmetry was the defect, not the pattern. TagRules::validate() is
now the single predicate: the dialog refuses to save against it, and
load() uses it to repair rather than drop, so a rule that fails is
visible and fixable instead of silently discarded.
- The typed name is sanitised into an id when the field is committed,
so the field shows what will reach the file. uniqueId() suffixes a
collision, since sanitising is many-to-one and can manufacture the
duplicate that load() then drops.
- An already-legal id is never rewritten, including one like "a---b"
that sanitising would otherwise collapse. Rewriting valid ids would
churn a file mailctl also reads.
- A bad id loads repaired, with the warning kept: what is on disk is
not what the hook runs until the file is saved back.
Deliberately not mirrored into mailrules.py. The hook tags real mail
unattended every ten minutes, where silently renaming an id is worse
than dropping the rule; the file converges as soon as the dialog saves.
No format change, so no version bump and no two-repo commitment.
The load warning was not missing: it had been showing "1 rule could not
be read and was skipped" on every open, in the same font and colour as
the intro prose two lines above it, and read as more explanation. It is
now a red banner beside Save, with an icon and a dismiss button, and it
says the rules need attention rather than that they were skipped, which
is no longer true. Dismissal is per-appearance only; a persistent one
would re-hide the problem that went unnoticed for a session.
Both new dialog tests were confirmed to fail with the sanitiser
reverted, and the banner's styling, position and dismissal each fail
under mutation. 20 of 20 suites green, 34 tests in test_tagrules.
Closes item 83.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 82. Saving a query worked and nothing else did: changing one field meant
retyping the whole query under the same name, and deleting one meant editing
the file by hand. An action that creates something the UI cannot then change or
remove is incomplete, and the user hit it within minutes of the first hand test.
Right-clicking a saved query, on its button or its menu entry, now offers Edit,
Move to menu / Show as a button, and Delete. Every path funnels through one
replaceSavedQuery(), which matches on the name the dialog was OPENED with rather
than the one it returns, so a rename replaces the entry instead of leaving the
original behind beside a new one, and which merges the stored entry's unknown
fields in a single place rather than in three. Delete confirms first: the rule
against confirmation dialogs covers tag mutations, which the undo stack can take
back, and this writes user config that it cannot.
Two cases the item did not anticipate. A generated entry has no query to edit,
so the dialog shows its composed query read-only rather than offering a field
that changes nothing, and carries `generated` and `flat` through an edit rather
than letting it decay into a plain entry holding a snapshot of what it resolved
to today. And the overwrite notice had to learn to ignore the entry being
edited, since warning that "Inbox" already exists while editing Inbox is noise.
This also fixes a defect that predated it and was already reachable from the
save path. rebuildSavedQueryRow() called deleteLater() on the old row, which
defers destruction to the event loop, so the stale row went on answering
findChild() and every lookup after a rebuild reported the state from before the
edit. Nothing looked wrong on screen, which is why it surfaced only as three
tests failing against a row that had in fact been rebuilt correctly.
Five tests, three mutations. Matching on the returned name fails two, never
writing the file fails three, and dropping the unknown-field merge fails one.
That last one initially proved nothing: it drove UNPIN, which copies the stored
entry and so carries `unknown` along by itself, and passed with the merge
deleted. It now goes through the edit path with a replacement that has none,
which is what the dialog actually returns.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
The spec asked for "a Save query button beside the search bar" and what shipped
was a menu entry and Ctrl+S. The user went looking for the button where the
design said it would be and did not find it.
Saving is a thing you decide on while looking at the results, so it belongs
where the results came from rather than behind a menu or a remembered chord.
The button takes the action through setDefaultAction rather than a second
connect, so it inherits the text, icon, tooltip and enabled state and cannot end
up offering to save an empty query while the menu entry correctly refuses. The
mutation that replaces it with a plain clicked() connection fails the test.
Also records item 82: a saved query cannot be edited, unpinned or deleted from
the UI. Item 23 specified saving and nothing else, and that is exactly what was
built, so the only way to unpin a query is a text editor or retyping it in full
under the same name. An action that creates something the UI cannot then change
or remove is incomplete, and this was found within minutes of the first hand
test. It is filed as a defect rather than an enhancement, and the spec now says
so where a reader would otherwise take the design for complete.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 23 had grown past what a backlog entry should hold: a storage format, a
migration, a dialog and a layout change. This document's own rule says a fully
specified item moves to specs/ and leaves behind the two or three things that
decide whether it can be picked up, the way items 53, 63 and 76 went. The entry
now carries the observation, the three deciding constraints and the relation to
item 10, and points at the spec for the rest.
The spec pins what was still loose. The JSON is an ordered array, since the
ordering is the whole reason for moving off [queries], and nothing may sort it
on load. A query's account scope stores the account KEY, the INI group suffix,
rather than the maildir path, so it does not duplicate config that already lives
in one place and go stale when the user edits it; the scope then composes
through Account::scopedQuery(), whose parenthesisation is load-bearing for the
same reason it is in the rules hook, an unparenthesised disjunction escapes its
scope and matches every account.
Migrated entries are pinned, so the query row does not silently empty on the
first run after upgrade, and migration order is alphabetical because that is
genuinely all the INI knows. Sent stays out of the file: it is generated from
allSentQuery() rather than stored, and folding it in would mean writing a
per-account path query into stored config, which is the duplication the
account-key decision just rejected.
The testing section is written against the traps already recorded in CLAUDE.md.
The migration test asserts the INI file is byte-identical rather than re-reading
it through QSettings, which would pass against a rewrite that preserved values
while dropping comments; the round-trip test asserts order, which is the
property the INI could not provide; and the dialog is left to a hand test,
because the offscreen platform cannot assert sizing at all and a Cancel goes
through done(int) rather than closeEvent.
Every code reference in the spec was checked against the files rather than
copied from the entry.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
The user proposed managing saved queries the way the tagging rules are managed,
in a JSON file rather than in the INI. It is a better answer than either option
the entry had been weighing, a per-query pinned flag or a [general]
pinned_queries list, because those each solved one problem and this solves
three.
Order is the one that could not be solved any other way. [queries] is read
through childKeys(), which returns keys alphabetically rather than in file
order, so the saved-query buttons appear alphabetically today and there is no
way to arrange them; config.cpp already carries a comment saying a hand-rolled
parser would be needed to change that. A JSON array is ordered intrinsically. On
top of that the document has room for the pinned flag the two tiers need, and
for the per-account scope the save dialog wants, which SavedQuery has nowhere to
put: it is {name, query} and nothing else.
The entry takes the shape of rules.json but explicitly not its machinery.
rules.json is JSON because two independent implementations have to agree on it,
this repo and mailctl's mailrules.py, and the unknown-field preservation and
version handshake exist to keep them from destroying each other's writes.
Queries have one reader, so only the versioned-document-with-unknown-fields part
is worth carrying over.
Migration reads [queries] once when queries.json is absent, writes the JSON, and
leaves the INI section in place. Stripping it would mean rewriting a hand-edited
file with QSettings, which drops comments and key order across the whole file
and is the exact loss this decision was made to avoid; leaving it costs a few
stale lines and keeps a downgrade working. Reading both forever was rejected as
two sources of truth for one thing.
This also retires the open question the entry had carried since 2026-08-04,
where the write should go. Both of the original answers were poor, one
machine-writing the user's hand-edited config and the other filing user intent
as window state under ~/.local/state. A machine-written JSON document beside the
hand-written INI is the cleaner split, in ~/.config so it lands in a config
backup.
Two constraints recorded that the build would otherwise meet late. startup_query
names a saved query and must keep resolving, and its "first entry" fallback
quietly changes meaning from alphabetically-first to first-in-the-user's-order,
which is user-visible and belongs in the changelog. And the README documents
[queries] in three places, one of which explains the alphabetical button order
this change removes.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 23 said only that a query could not be saved from the UI, and left the
presentation as a one-line sketch. The user described what they actually want:
a Save query button beside the search bar, opening a dialog that takes a name
and an account scope; saved queries split into two tiers, with a few kept
visible as buttons and user-made ones behind a menu; and the button row moved
onto a row of its own once it no longer holds everything.
Two things came out of writing it down. There is no built-in default query set
in the code at all: every entry in [queries] is user-written and renders
identically, and Sent is the lone exception because it is built from
allSentQuery() rather than living in [queries]. The two tiers therefore need a
mechanism that does not exist yet, either a per-query pinned flag or a
[general] pinned_queries list, and that config format choice is the one decision
left on the item. The design also settles the question the entry had left open:
a query the user named and scoped in a dialog is intent rather than machine
state, so it goes in qtmaildir.conf beside the hand-written ones, at the cost of
QSettings reformatting a hand-edited file on first save. The README has to say
so.
The save-as-a-filter half is split out as item 81. A saved query is a view and
costs nothing if it is wrong; a rule is applied to real mail by the post-new
hook every ten minutes and lives in the rules file that this repo and mailctl
implement independently. Folding it into 23 would make a presentation change
carry a two-repo commitment, so 23 can now ship without it. Item 81 records the
constraints it will hit: a stored query carries no scope and the hook
parenthesises it, which matters more here than usual because a query saved for
a view is often a disjunction, and the hook refuses to remove unread or inbox,
so the dialog must say so rather than failing silently.
Item 23's relation to item 10 firms up as a result. Account scope in the save
dialog answers item 10's remaining half as a side effect, so the entry now says
that outright, and says not to reopen item 10 to do it: the user postponed it
and asked that the rest not be proposed unprompted.
Also corrects a stale reference. The entry claimed a SavedQueryBar class shows
the saved queries; no such class exists or ever has, and the buttons are built
inline in MainWindow.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 73. The backlog kept every item's full Observed/Cause/Approach section
forever, including the sixty-eight that are closed, and had reached 5056 lines:
past the point where it could be read in one pass, and past the point where a
tool could open it at all.
The closed sections move to 2026-08-03-post-0.1.0-usability-closed.md, taking
the backlog to 570 lines. The status table stays where it was and remains the
index of all 80 items, so a closed item keeps its row, its date and its outcome
beside the open ones; only its evidence moved. Nothing was renumbered and
nothing was deleted, which the item required: the numbering is cited from commit
messages, from CLAUDE.md and from the specs, and both files share one sequence,
so item 42 is `## 42.` in whichever file holds it.
The split was done by script and verified by set difference rather than by
reading: every non-blank line of the original appears in one of the two files,
zero missing, and the only lines not in the original are the new file's header.
All 80 numbers resolve, every open item has its section in the backlog, every
closed one in the archive, with no duplicates and no orphans.
Two things the item's own approach did not anticipate. Three cross-references
said "see below" and their targets had just moved, so rows 60 and 75 and the
header's note on item 20's parked branch now say where the entry went. And the
cause was never the fifty done sections, it was that nothing moved a section on
the day its item closed; doing this once buys a few months and then item 73
returns. The rule in "Adding to this document" now requires the move on the
closing commit, and CLAUDE.md tells a future session that grepping the backlog
for a closed item's evidence will find the table row and nothing else.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 77. The dialog could say how many messages a rule matched and not
which ones. A Preview in list button now runs the selected rule's query
in the main window; the dialog stays open, since comparing the rule
against its results is the point.
Two constraints from the backlog entry, both now asserted and both
mutation-checked.
The query runs exactly as stored, with no tag:new and no wrapping
parentheses. The post-new hook adds those when it applies a rule, and a
preview that copied them would match nothing outside a sync window,
since tag:new is set only on mail that has just arrived.
The account selector is cleared first. runQuery() wraps the bar's text
in the selected account's scope, and a rule query usually names its own
path already, so previewing one with an account selected would scope it
twice and show an empty list, which reads as "this rule collects no
mail".
The second mutation only fails once the test's config has an account to
select: with the default empty config the selector sits on "All
accounts" anyway, and asserting that a preview leaves it there passed
against the mutation. Recorded in the test.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 80. A rule with eight From conditions left the list showing about
one and a half rows.
The list was added with stretch 1 and the form below it with none, which
looks decisive and is not: a stretch factor only distributes space above
each widget's minimum, and the form's minimum grew with every condition
row, so each row came straight out of the list. The builder asked for
120px with one row and 414px with eight.
A QSplitter now divides the list from the editor, so the balance is the
user's and is saved beside the column widths, and the condition rows sit
in a QScrollArea capped at 190px so the editor cannot grow without bound
however the splitter is set. The scroll area is what text mode hides;
hiding the builder inside it would leave an empty frame.
Three measures were tried in the test before one told the bug and the
fix apart, and two passed against broken code: the dialog's
minimumSizeHint does not track form rows and read 580 either way, and a
qMin against the scroll area's own hint read small whether or not the
cap was set, since an uncapped maximumHeight is QWIDGETSIZE_MAX. What
survives mutation is the editor pane's minimum inside the splitter, plus
the cap read directly, and both are asserted. A row's size hint is
invalid until the event loop runs, so the test calls processEvents after
selecting a rule or it measures the same height twice.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 75 shipped claiming the rules window remembers its size. It does
not, and no code here can make it.
Hyprland tiles the window to fill its slot, so the size dragged belongs
to the tile. saveGeometry stores frameGeometry beside normalGeometry and
restoreGeometry restores the normal one, which stays at whatever
resize() last set it to. Decoded from the real state file after a hand
test: frame 2248x806, normal 760x664. The dialog restores 760 correctly
and still opens tiled.
Three diagnoses were tried before this one and each was disproved by a
probe rather than argued away: that restoreGeometry rejected the blob as
off-screen, that the layout overrode a geometry applied before the first
show, and that a test could tell the broken and fixed versions apart.
The last one matters most: the offscreen platform returns an identical
frame for both, so a size assertion passed against the bug and a
mutation restoring it left the suite green. That assertion is not
reinstated.
The column widths, which are what actually works, keep their test. The
changelog and the backlog entry are corrected to say what ships, and
CLAUDE.md gains both the tiling-compositor trap and the rule that the
offscreen platform cannot test window sizing at all.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
The geometry was saved from closeEvent, and neither dialog button sends
one: Cancel calls reject(), Save calls accept(), and only the window
manager's X button produces a QCloseEvent. So the size and the column
widths were kept for the one route out of three that a user almost never
takes, and a resize followed by Cancel came back forgotten.
The save moves to a done(int) override, which both buttons funnel
through and which QWidget::close() also reaches.
The test that covered this passed against the bug because it asserted
with close(). It now drives all three routes rather than trusting one to
stand for the others, and shows the dialog before the close leg:
close() on a widget that was never visible returns early without
reaching done(), so that assertion would otherwise prove nothing.
Both traps recorded in CLAUDE.md, since neither is specific to this
dialog.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Item 75. saveGeometry() and the rule list header's saveState() go to
uistate.conf under keys of their own, written on closeEvent so a size
survives Cancel as well as Save. The 760x520 resize stays as the
first-run fallback.
The backlog's approach was wrong on one point and a test caught it. It
said to drop the resizeColumnToContents calls once a saved header state
exists, which fixes the restore and leaves the original defect standing:
with nothing saved, a width the user had just dragged was still
discarded by the next add or delete. Each column is instead auto-sized
once, on its first fill, after which the width belongs to the user
however it was set. Two flags, because the count column is filled later
by a reply from the worker.
The window stays a QDialog. Making it a top-level window needs the
unsaved-edit story that being modal currently sidesteps, and that is its
own decision rather than part of this item.
Both tests redirect XDG_STATE_HOME as well as XDG_CONFIG_HOME, so they
cannot write the real uistate.conf. The geometry is asserted on the
stored value rather than the reopened frame, per item 46: the offscreen
platform does not honour a resize.
Also corrects setFolders' doc comment, which still described the folder
list as coming from Config.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
|
|
Ticking "Edit as text" was a one-way trip: the only way back to the rows
was closing the dialog and reopening it. The checkbox was parented to the
builder widget and sat on the match row, and switching to text mode hides
that widget, so the toggle disappeared along with the rows it governs.
Move it to the query row, which is visible in both modes.
The existing tests all passed against this, because they drove the toggle
through setChecked and then asserted on the checked STATE. A hidden
checkbox reports its state perfectly well, so every one of those
assertions held while the widget was unreachable. The new test asks the
question that matters, whether the toggle would be on screen, and it uses
isVisibleTo since nothing is isVisible on a dialog that was never shown.
Worth recording how close the mutation check came to endorsing this too.
Reparenting the checkbox alone left it in the query row's layout, so it
stayed visible and the test still passed. Only restoring the full shipped
shape, parent and layout together, reproduced the fault and failed the
test. A mutation that does not reproduce the original bug proves nothing
about the test that is meant to catch it.
The spec's layout sketch carried the same error and is corrected, with
the reason, so the next reader does not reintroduce it.
|
|
The first pass of the field repair restored the query and the tags and
stopped there. The note was also blank, which the user noticed: every
sibling account rule carries an identical note and only the damaged rule
had none.
The reason it was missed is worth keeping. The shell backup was read for
the tagging command, and the note comes from the comment block above it,
which the migration had given to all five account rules alike. The
handler at fault writes every field of a rule, so every field is equally
exposed, and a repair that checks only the fields that first drew
attention will leave some of the damage in place.
Restored from the four siblings, which are byte-identical, and the whole
file re-audited: no rule now has an empty query, id, note or tag list.
|
|
Opening the tagging rules dialog and pressing Save destroyed the first
rule in the list, without any editing. The rule lost its query and its
tags, then vanished entirely on the next load, since a rule with an empty
query is dropped as malformed.
Reproduced against the released tag rather than the branch, in a
throwaway worktree at 9585674 with a two-rule fixture: constructing the
dialog and running its save path left one rule of two.
onSelectionChanged blocked signals for the note widget only, while
m_enabled::toggled two lines later reached applyEditsToCurrentRule, which
writes every field from widgets the loader has not filled yet. The
existing comment there shows the hazard was known for one widget and not
extended to the other.
The fix landed with the builder work: the reloading flag now covers the
whole load, and switchingRulesDoesNotLeakRowsBetweenThem is the
regression test, verified by mutation to fail without the guard.
The live rules file had one casualty, the account rule sitting first in
the list, with both its query and its tags empty while every sibling was
intact. Restored from the shell backup that the earlier migration kept
and verified through mailctl's own reader. The rule had stopped tagging,
but only one message had arrived meanwhile; that message is now tagged
and the account is complete again at 14969 of 14969.
|
|
The spec listed from:(((( among the queries the parser must reject, and
the plan's Task 6 asserted that rejection. Probing the built parser shows
it accepts the query as a From row whose value is the literal text, and
compiles it back byte for byte.
That is correct behaviour, not a leak in the strictness rule. notmuch
reads those parens as characters to search for rather than as grouping,
so the query is meaningful and the row displaying it tells the truth.
Rejecting it would buy nothing and would push a representable rule into
text mode.
The distinction the documents were missing: a parenthesis inside a VALUE
is not a shape question at all, only a parenthesis in grouping position
is. Restate both documents accordingly, and replace the assertion with a
round-trip one, which is the property that actually matters here.
|
|
The draft compile() quoted every Is/IsNot term, which contradicted the
same task's own assertion that a negated tag compiles to .
The implementer resolved it in the direction the tests specify, and the
resolution is right: notmuch reads tag:inbox and tag:"inbox" identically,
counting 5322 either way against the live index, so quoting a tag would
change the stored string without changing what it matches. That breaks
the byte-for-byte round trip this type exists to guarantee.
Restate the comment as the rule rather than as a note about what a test
expects, correct the plan's draft so the remaining tasks do not inherit
the contradiction, and warn the parser task that a quoted tag must not be
read back as a quoting operator.
|
