| Age | Commit message (Collapse) | Author | Files | Lines |
|---|
|
The action carried mail-mark-important, which Breeze and several other themes
draw as an exclamation mark rather than a star. The filter button for the same
tag has used `starred` since a15505d, where the comment records the user asking
for a star when item 57 renamed the action, and the two were allowed to differ
on the reasoning that a query-row icon reads as a category while an action icon
reads as a verb.
That reasoning held only while the action appeared beside its own label. Item
189 put it on the icon-only message bar, where the icon IS the control, and it
read as an info glyph. Both are `starred` now.
The label stays "Important" and the tag stays `flagged`; only the picture
changes. `starred` sits under status/ rather than actions/ in the icon spec,
which needs no fallback: an unresolved name already leaves the action with text
alone, and it resolves in the user's own theme, verified.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NY6poqw199LfFaXe5BHKNe
|
|
Follows items 185 and 186, which established the pane's bar as where actions
on the displayed message live. Star and Archive are both selection-scoped and
fit that rule with nothing to decide; Archive leaves the main toolbar the way
Delete did, since the same icon in two places reads as two controls when the
toolbar is icon-only.
Ordered by what they do rather than by where they came from: answering the
message, then filing it, then destroying it, so the destructive button is not
between two that are not.
Mark all read deliberately stays on the main toolbar, at the user's decision.
It is the one action in this window that ignores the selection and acts on
every row in the view, so a bar whose every other entry acts on the one
displayed message is exactly where it must not be.
Item 140's toolbar test named archive as an example of a list-wide action.
That was never true of it, only untested, and this item reclassifies it: the
test now asserts archive LEFT the toolbar and keeps its guard on
mark_all_read, which is the action that genuinely is list-wide.
Closes item 189.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NY6poqw199LfFaXe5BHKNe
|
|
The pane's bar offered Reply and Forward on a message the user had thrown
away, which are the two things a trashed message is least likely to want,
while Restore and the purges lived only in menus.
The bar now has a third branch, asked before the draft one: a deleted draft
must come out of the trash before it can be edited. It is keyed on the
SELECTION being in a trash folder, the same predicate the menu entries use,
rather than on the trash VIEW, which disagree on mail reached from a search.
It carries Restore, Delete permanently and Empty trash, and only Restore is
tinted: the two purges are one act at two scopes and need no colour to tell
them from each other, only from the one action that gives mail back.
Delete moves here from the main toolbar in the same change (item 186). It
acts on the displayed message, like Reply and Forward, so it belongs on the
pane's bar by the rule items 139 to 141 settled for those two. It stays in
the Message menu and the context menu.
Delete permanently is new. It is Empty trash scoped to the selection, the
same purgeMessages() call with the ids resolved from the selection rather
than from a query, so it inherits both of that action's safeguards: it
confirms, naming the count, and it carries no default shortcut. One combined
thread:/id: query resolves a mixed selection, so a conversation and a reply
selected together still ask once.
The bar is refilled when the conversation digest arrives as well as on
selection, since a conversation's trash-ness is not known until every path
has been reported.
Closes items 185 and 186.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01NY6poqw199LfFaXe5BHKNe
|
|
Item 125, the half that was genuinely missing.
Most of this item was already built and the row was stale. The exit-75
branch in onSyncFinished() predates this session and does what the entry
asks: the spinner clears, the skip is reported as neither success nor
failure, m_lastSyncFailed stays put, the log pane is not raised, the lock
latch is handed back to the external monitor, and the sync-on-exit case
has its own dialog. Item 174 then added the external half, a `skipped`
state a run the application did not start can be seen to have produced.
What nothing covered was the RE-ARM, and it is the symptom the item was
filed for. runAutoSync() re-arms when it declines to START, which is item
89 and covers a sync skipped before launching. A run that LAUNCHES, finds
the lock held and exits 75 reaches onSyncFinished() instead, and that
branch armed nothing: the edit stayed pending with nothing scheduled to
carry it, waiting for a manual sync or the next cron tick. That is "a
held edit waits for a completion that never comes".
scheduleAutoSync() in the skip branch. It re-checks the delay, the sync
command and the pending count on the way in, so it cannot arm a sync for
nothing, and against a long external run it re-arms once per debounce
interval until the lock clears.
This is the half the status file could not reach, and the distinction is
worth keeping: that file says what a run DID, this is what the
application does NEXT.
The test records a real pending edit first, since runAutoSync() correctly
declines when there is nothing to carry and a fixture without one would
arm nothing for a legitimate reason. It failed before the fix and is
mutation-checked.
Suite: 43 of 44, with undoMovesTheMessageBack failing as it does on
master (item 136).
|
|
Item 174, and half of item 125.
The premise was corrected before any code. The note asks for an external
`notmuch new` to clear the pending count; it must not. That count means
tag mutations not yet known to have reached the MAIL STORE, which is the
server: an edit is in notmuch the moment it is made, and what is
outstanding is mbsync pushing the renamed Maildir files. `notmuch new`
re-indexes local files and pushes nothing, so clearing on it would tell
the user their work was safe to quit on while it was still local. The
entry's own proposal to watch notmuch_database_get_revision() was
rejected for the same reason: a revision moves when mail ARRIVES too, and
in neither case does it say anything about the server.
What was actually wrong was the reporting channel. The application
inferred a finished run from an inode in /proc/locks and from grepping
the log for its RUN END banner, which made a human-readable line into
wire format and could not say WHICH channels a run carried. The local
sync path has always narrowed its clear to the accounts it carried; the
external path could not, and cleared everything, so an edit to an
account a run never touched was reported as delivered.
So the script reports instead of leaving evidence to be inferred. It
writes ~/.local/state/qtmaildir/syncstatus.json atomically at the end of
every run, including a skip, naming the channels, both exit statuses and
a state of ok, failed or skipped. MailSync::readStatus() reads it,
MainWindow prefers it over the log banner and narrows the clear through
Account::syncChannel(). A skipped run clears nothing, which is item 125's
first half: the application can now see that a run happened and carried
nothing. The log banner and lastRunOutcome() stay as the fallback for a
missing file, which is what a first run after upgrading looks like.
This is the user's own framing of the scope: the script was written for
another system and adapted, and is now qtmaildir's only consumer, so it
serves the application rather than the reverse. Two facts made it safe to
act on: their crontab runs mailsync.sh and nothing else touches mail, and
~/bin/mailsync.sh is a symlink into this repo, so an edit is live on the
next tick.
Two bugs found while wiring it in, both recorded in the closed item.
A test read the developer's real sync state, twice: a [sync] section
naming only `log` leaves syncStatus() defaulting to the real file, so two
tests asserting that a FAILED run leaves the count alone read the last
real cron run, found ok, and cleared. Pinning only `status` has the
mirror problem. noSyncTestReadsTheRealSyncState() is the guard, modelled
on noTestCanSeeTheRealLockTable().
And Qt::ISODate carries no milliseconds. The status file is preferred
only when it describes THIS run, compared against when the lock appeared,
so a stale success cannot outrank a fresh failure; but the script writes
date -Iseconds, and a round trip of "now" comes back 329 ms behind,
measured. A fast sync's own file therefore parsed as stale and fell back
to the log, with nothing failing to say so. One second of slack matches
the precision the format carries.
Design: docs/superpowers/specs/2026-08-29-sync-status-file-design.md
Suite: 43 of 44, with undoMovesTheMessageBack failing as it does on
master (item 136).
|
|
Item 182, found by hand: a thread of 9 messages with 5 unread, marked
read while a sync was running, reported "<subject>: mark as read" and
then reported the same work again when the sync finished. The user read
it as double reporting.
Not a double write, and the mail was correct. It is one action reported
twice because the FIRST report was the wrong one.
A sync holds notmuch's exclusive write lock and the worker's read-write
open blocks on it rather than failing, so an edit made during a sync is
held and sent when the lock frees. All three hold branches say exactly
that, in a label chosen deliberately: NOT transient, because it
describes state lasting until the sync ends, and a message that expired
would leave rows showing a tag the database has not got and no
explanation of why.
That label never survived. Every caller announced the action itself a
line later through showTransientStatus(), which overwrote it, so the
user was told the write had happened and the hold was never mentioned.
The flush at the end of the sync then reported the same work again and
read as a duplicate rather than as its completion.
announceAction() asks whether a sync holds the lock and, when one does,
sets a non-transient label naming the action AND the wait. The action is
still named because that announcement is what stands in for the
confirmation dialog this project rules out: it is how a user tells that
something larger than they meant has just happened, so the hold is added
to it rather than replacing it. The flush message is untouched and is
the only signal that held work actually landed, whose absence was item
106.
The test drives toggle_unread, the route the user took, and asserts both
halves: the text mentions the sync, and it still says what is waiting.
Asserting only the first would pass against an announcement that dropped
the action entirely. Mutation-checked by forcing the non-held branch,
which fails with the exact text the user reported.
The new string is translated, since one that misses the Italian ships as
English inside an otherwise Italian UI: lupdate found it with no context
warnings, lrelease reports 552 finished and 0 unfinished.
Suite: 42 of 43, with undoMovesTheMessageBack failing as it does on
master (item 136).
|
|
Item 181, from the user's notes: "the thread dashboard doesn't update
live with the modifications applied to the list pane. If I mark the
thread as read, the dash still reports N unread".
ThreadDashboard draws a ThreadDigest, which the worker builds from the
index and which reached the pane only when a conversation was selected.
A tag write updated the model optimistically and repainted the card
beside it, and nothing touched the digest, so the pane went on reporting
the unread count, the progress bar and the Waiting-for-you list the
conversation had when it was opened.
Reachable from the dashboard's own Mark all read button, which is the
worst version of it: the number sits directly above the button that
fails to move it.
refreshDashboardDigest() re-asks the worker for the digest of the
conversation on display, and returns at once when the pane is showing
anything else. It bumps m_digestGeneration like any other request, so
the guards in onThreadDigestLoaded() discard a reply that arrives after
the user has moved on. No placeholder digest, unlike the selection path:
the pane already holds this conversation, and blanking it to re-fill it
would flicker the whole dashboard for a change to one number.
Called from onTagsApplied(), where a write is CONFIRMED, and not from
the two write funnels. The first attempt put it beside the optimistic
model update by analogy with every other optimistic repaint, and that
analogy does not hold here: the digest is rebuilt from the index, so a
refresh queued beside the write reaches the worker before the write does
and answers from the state before it. The test failed identically to no
fix at all.
Every write rather than a chosen subset, at the user's decision:
narrowing it to the writes that change what the dashboard happens to
draw today is a list the dashboard can outgrow silently, and this costs
a round trip only while a conversation is on screen. Re-requested rather
than edited in place, because the digest is a derived summary and
recomputing it here would be a second place that has to agree with the
worker about what a write did.
The test is worker-backed over a real two-message conversation and is
driven through the mark_all_read action rather than the private funnel,
which is the path the dashboard's own button takes. It asserts the pane
carries the unread state before the write, so the assertion after it
means something.
Suite: 42 of 43, with undoMovesTheMessageBack failing as it does on
master (item 136).
|
|
Item 178. everySelectedRowIsInATrashFolder() read
ThreadSummary::firstMessagePath for any row that was not a message row.
That was correct while a thread row MEANT that message (item 108) and
stopped being correct when item 177 made it mean the conversation. A
conversation is in the trash only when ALL of its messages are, so a
partly trashed thread answered on whichever message the query returned
first: Delete could be hidden on a conversation that still had mail
outside the trash, and Restore offered on one that mostly did not.
Not data-affecting. Both actions are no-ops in the wrong direction:
Delete on already-trashed mail takes moveMessages()' already-there
branch, and Restore on mail that was never trashed finds nothing to
move.
qtmaildir cannot produce such a thread itself, since Delete is absent on
a reply row and Restore is thread-scoped. Two things outside it can:
another client trashing a single message, and a reply arriving after the
conversation was trashed.
ThreadDigest already walks every message of the selected conversation
for its sender counts, and a filename is served from the index like
everything else in it, so the paths ride along on a request the
selection already makes rather than costing a walk on every query.
ThreadDigest::messagePaths is relative to the mail root, for the reason
firstMessagePath records: an absolute path matches no account and
silently resolves every row to none. MainWindow keeps them beside the
dashboard's thread id and clears them when the dashboard is left, so a
late digest cannot answer about another row.
One limit, stated in the code rather than hidden. The digest is
requested only for a single selected conversation row, so that is the
only case with a real answer; any other selection falls back to the
summary's one path. That fallback IS the pre-177 answer and is wrong in
exactly the same partial case, which is the point: a multi-row selection
is left no worse than it was, rather than given a second, differently
wrong rule of its own. Making it exhaustive costs a per-query walk over
every message, which is what this avoids.
Two tests, both mutation-checked. The worker test puts its two messages
in different folders, since two in one folder answer identically
whichever way the code resolves them. The window test asserts both
directions, so a fix that simply hid Delete everywhere would fail it,
and sets totalCount explicitly: a summary left at the default is a
message row, and the test would otherwise exercise the other branch and
pass for the wrong reason.
Suite: 42 of 43, with undoMovesTheMessageBack failing as it does on
master (item 136).
|
|
The expander pill read "N replies" while the row stood for the
conversation: a thread of one message and four replies said "4 replies"
over rows that listed all five messages. The user's model is messages, so
it now reads "5 messages". A thread of one still shows nothing: its row is
the message, the pill is the expander, and there is nothing to open.
ReplyCountRole becomes MessageCountRole and CardLayout::Input::replyCount
becomes messageCount, so the names stop lying about what they carry. The
label is now translated under a CardLayout context, with Italian
"messaggio"/"messaggi" shipped; %n's untranslated fallback on this Qt does
not pluralise, so the two forms are separate entries. The card's densest
geometry test needs 460px rather than 400 now that the pill is one
character wider.
|
|
setThreadMessages() dropped nodes.first(), which was correct while a thread
row MEANT its first message: listing that message under itself would have
shown it twice. Item 177 made the row stand for the conversation and render
a dashboard instead, so the drop left the first message with no row
anywhere: the user reported the list starting at the second message with
the first unreachable.
A conversation row now keeps every message, including the first; a thread
of one keeps the old rule, since there it IS its message and must not be
listed beneath itself. The choice reads what actually ARRIVED rather than
summary.totalCount, which counts duplicates and can lie about whether a
thread really is a conversation.
Reply-scoped tests pointed at child 0, which was the first reply and is now
the first message; they read child 1 instead, and two asserted a child
count that grew by one. Two new model tests pin both halves, and both
directions are mutation-checked.
|
|
A thread row has no message to render, so the pane shows the conversation
instead. A thread of one message still opens its message on one click, and
the automatic mark-read is not armed for a row that displays nothing.
|
|
Senders, unread messages and an activity histogram for the dashboard, as a
plain value struct over a queued signal. Everything comes from the index, so
no message file is opened; the unread list is capped and unreadTotal carries
the real number.
|
|
Closes item 176. applyTags reports the messages whose tags really moved, and
a command stores that rather than what it asked for, so undoing a mark-read
no longer marks the whole conversation unread.
|
|
A widget over a ThreadDigest: header, tags, counts, the unread list capped
with a link to the rest, and an activity sparkline, scrolling under a pinned
action strip. It invents no colours.
ThreadDigest::unread is QVector<MessageNode> rather than QVector<MessageRef>.
The dashboard draws rows for messages it never opens, which is what
MessageNode exists for; MessageRef carries only what the pane needs once a
message is already open and has no subject, sender or date. The struct's
no-file-opening contract still holds, since all four of those fields are
served from the notmuch index.
|
|
Closes item 170 under item 177. A conversation belongs to a view while any
of its messages match, so reading one message of a thread no longer takes
the conversation out of the Unread view. The current row is never evicted,
and an automatic write defers its eviction until the selection moves.
|
|
Item 112 hid the toggle whenever the selection disagreed, because a union
is not a state and no honest label existed for it. That was affordable
because the "Whole thread" submenu sat beside it carrying two absolute
entries, which worked whatever the mix.
Item 177 deletes that submenu: the row decides the scope, so a second set
of actions is a second answer to a settled question. Hiding the toggle
then leaves the commonest conversation in the mailbox with no key at all.
The rule is a catch-all instead. Any unread message, a mixed conversation
included, reads "Mark thread as read" and marks every message read; only a
fully read selection reads "Mark thread as unread". Two presses therefore
reach either state from anywhere, which is what makes one key enough.
The write direction moves with the label. Computing it from
everySelectedRowHasTag() while the label promised "read" would mark a
mixed conversation unread, which is the item 112 report happening again
from the other end; the mutation putting that back fails the new test.
The three-valued selectionTagPresence() is unchanged and still asked, since
Every and Mixed differ for other callers. Only this label collapses them.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01PEK9z5D3oa1nVmJ6xpQhBs
|
|
The five *_thread actions and their submenu are gone: the row's identity is
what decides the scope, so a second set of actions was a second answer to a
settled question. mark_thread_unread went with them, being the sixth entry in
the same submenu. tagSelected() loses its TagScope parameter, and
everySelectedRowHasTag() its own, so the direction and the write ask the same
question of the same object. ThreadListModel::scopeFor() and messageScopeFor()
are deleted; scopeForSelection() is the one resolver.
Labels name the scope. Archive, Delete, Restore, Spam, Important and the
unread toggle all say "thread" on a conversation row, and Delete, Restore and
Archive are ABSENT on a reply: a single reply cannot be removed from a
conversation.
Compose follows the same rule. Forward, Save, Reply-all and Reply without
quoting disappear on a conversation row, which shows no message to act on, and
Reply becomes "Reply to this thread": reply-all, quoting nothing, threaded off
the conversation's NEWEST message so the answer lands at its end rather than
forking the discussion at its opening post. That id is not in the model, since
an unexpanded conversation holds no nodes for its replies, so it comes from
resolveThreadMessages(); resolveQuery() states its newest-first sort rather
than inheriting notmuch's default.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_012iDeN6C7y97nHYPvP6ST4L
|
|
One resolver replacing the scopeFor/messageScopeFor pair. The caller no
longer chooses the scope, which is what let one gesture mean two things.
|
|
Nothing is a sibling any more: a conversation row draws the thread's tags
and a message row draws its own.
|
|
Items 110 and 111 reconciled a card that showed one message with a row that
was a thread. The row is the conversation now, so the union is simply what
it means: the first-message substitution, PillOwnCountRole and the seeded
first node all go.
|
|
One predicate for the question every scope, label and membership decision
in item 177 keys on. It repeats hasChildren()'s rule deliberately: an
expander and a conversation are the same fact, including that a loaded
thread trusts its children over a count that included duplicates.
|
|
Item 171. A forward carried only the plain-text version of the original,
so formatting was lost; and an original with no plain-text part at all
(30 of 342 sampled inbox messages, ~9%) forwarded as an empty quote with
its content silently gone.
A forward now sends ONE part chosen by the Send-as-HTML toggle: the
original's markup when on, the text quote when off. Not a
multipart/alternative, at the user's decision: a forward's shape is
already decided by that toggle, and sending both hands the choice to the
recipient's client. The toggle is honoured even for an HTML-only
original, which then forwards as a text fallback.
HtmlSanitiser strips remote content from the forwarded markup, checked
by default with a per-forward opt-out. This is the security-critical
part: the markup leaves this process and is rendered by the recipient's
client, where none of MessageView's protections apply, so forwarding a
tracking pixel forwards the tracking. It is an ALLOW-LIST, unlike
HtmlBuilder::namespaceCids(), because a missed rewrite is a broken image
while a missed strip is a beacon reaching the recipient.
An HTML forward does not seed a text quote into the editor. The first
build did, then subtracted it when building the HTML part, so the user
could edit a quote whose edits were discarded; what the composer shows
must be what gets sent. The forwarded message appears in a read-only
pane beside the editor instead, a QSplitter at 60/40 with a toggle in
the Format menu. A plain forward is unchanged.
ComposeContextBuilder::quoteBody() renders htmlBody down to text when
there is no plain part, so the plain path never emits an empty quote.
Design in docs/superpowers/specs/2026-08-27-forward-html-design.md.
Two tests repaired for the splitter: the 60/40 assertion reads stretch
factors rather than pixels, since the offscreen platform gives the
splitter no width and reports 49/49 whatever the code asks; and
theComposerSplitsItsToolbarByScope looked for the body directly in the
composer's column.
Not yet hand-tested in this arrangement.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AtUzfNjMD8fiYfamDd3ywW
|
|
DraftStore::write() was called with "D", and it uses the flag string
verbatim, so every draft this application wrote landed as :2,D. With
maildir.synchronize_flags on, notmuch tags any message lacking the S
flag `unread`, and a draft the user authored is seen by definition.
The symptom heals itself: the next sync of that folder round-trips the
file, adds S, and the tag goes away. Only the newest draft in a folder
that has not synced since shows it, which is why it read as
intermittent and why measuring an older draft finds nothing wrong.
TestComposeWindow::aSavedDraftIsFlaggedSeen() asserts both flags on the
written filename, verified failing first against "D".
TestMainWindow::anAutosaveWritesADraftAndClearsTheDirtyFlag() asserted
endsWith(":2,D"), pinning the whole flag set where its own comment said
the point was the draft flag "not left bare", so it failed against the
corrected behaviour. It checks for D within the flag set now.
Also reconciles the backlog with the user's notes: records the
forwarded-HTML defect as item 171, closes item 169 (shipped last
session, its row still read open and its section was still in the open
file), and records this fix as item 172.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01AtUzfNjMD8fiYfamDd3ywW
|
|
380px of dialog for three rows left most of itself empty. The height is
asked of the layout now, capped so a long list scrolls rather than growing
past the screen and floored so a single row does not collapse it.
A background role on the scroll viewport was tried in the same pass and
reverted: the dialog renders semi-transparent under the developer's
compositor, and painting a Base-coloured layer under the list made that
worse rather than better. The transparency is the desktop's own doing, which
is the trap this project has already recorded for window geometry.
Not covered by a test. The offscreen platform returns an identical frame for
a correct size and a broken one, so an assertion there would pass against
both; CLAUDE.md records that measurement. Confirmed by hand instead.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P88Q3MCSCSQxKDy7pmXh9F
|
|
Item 119, and item 146 which is the same request recorded again. The status
bar's count answers "is my work safe to quit on" and could not say what the
work was.
The label opens a read-only list on a click. A QLabel has no clicked signal,
so the press is taken by MainWindow's existing event filter rather than by
replacing the label with a flat QToolButton, which would have brought the
style's button metrics into a status bar the label already sits correctly
in. The pointing-hand cursor is the affordance, since a status-bar label has
room for nothing else.
The layout is the user's own: a message appears once with its actions
beneath it. PendingChangesDialog::rowsFor() does the grouping over a run of
rows sharing an id, which the snapshot has already ordered, so the actions
under one message keep the order they were made in.
Read-only, deliberately. Retrying or discarding a change from here would be
a new mutation path with its own undo question, and the count exists to be
understood rather than edited.
Three rules the tests pin, each of which is a way the list could disagree
with the count it was opened from:
- Grouping must not collapse: two actions on one message are two rows.
- A thread row stays thread-scoped and reports how many messages it covered.
- An id the index no longer holds still opens a run of its own, showing that
its subject is unknown rather than folding its actions under the message
above it. This is why the row carries startsMessage rather than inferring
it from a non-empty subject.
The queued call carrying QStringList, QList<bool> and QList<int> is covered
by a test that drives it across a real thread, since a container whose
metatype does not resolve is dropped at runtime and the slot runs with a
default. Both survive on Qt 6.11; the test is what says so, and what would
fail if that changed.
Italian ships with it: five new strings, lupdate clean, lrelease 522
finished and 0 unfinished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P88Q3MCSCSQxKDy7pmXh9F
|
|
Item 119, second half of the data: the step that turns the snapshot's ids
into something worth showing.
resolvePendingSubjects() takes the rows' ids in order, each flagged as a
thread id or a message id, and answers positionally: one subject per input,
plus the thread's message total for a thread id and -1 for a message id.
Positional rather than set-based, and that is load-bearing. The caller has
already decided what its rows are and in what order, and one id can
legitimately appear on several rows: a message with two outstanding actions
is two rows carrying one id. A combined query returns a set, which loses
both the order and the duplicate, so the walk is one lookup per row instead.
The cost is bounded by what the user did by hand since the last sync, which
is not a query-sized number.
A missing id answers with an EMPTY subject rather than being dropped. The
dialog still shows that row, because the count the user clicked has to equal
the list they are shown, and dropping a row breaks that agreement in exactly
the case where the user is most likely to notice. An index that cannot be
opened answers the same way, one empty subject per row, so the list still
shows the changes with only the subjects missing.
The thread count is taken at snapshot time and says so: a held thread edit
applies when the sync ends, and a reply arriving in between makes the real
number larger. The row describes what the user is looking at, not what the
write will touch.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P88Q3MCSCSQxKDy7pmXh9F
|
|
Item 119, first half: the data the list behind the unsynced-changes count
is built from, with no dialog and no worker, so the rules it has to follow
are testable on their own.
pendingChangeSnapshot() gathers the three queues the count sums into
PendingChange rows. Two properties are the whole point.
Scope follows the ACTION, not the storage. A held thread edit stays one
thread row, because a `*_thread` action made it and reporting its messages
instead would claim the user acted on each one; a netted tag edit and a held
move are message rows. The queues already encode that distinction, so
nothing is expanded and nothing is escalated.
The rows are grouped by id, so a message with several outstanding actions
appears once with its actions beneath it, which is the layout the user
asked for. The sort is stable, so those actions keep the order they were
made in; QHash has none of its own, and without it the list would reshuffle
between openings.
A snapshot, taken once and frozen. Subjects are empty here and filled by the
resolve step to come.
m_pendingTagEdits gains the action name beside the direction it already
kept. The direction alone was enough to count with; a list has to say what
each change was, and only the action that made it knows. It is carried from
TagChange::description rather than derived from the tag, so there is no
second table of tag names to labels to drift from the first.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P88Q3MCSCSQxKDy7pmXh9F
|
|
Item 119's stated blocker, removed by finding out what it held: nothing.
pendingEditCount() summed four sources, three of which can name the messages
they hold and one of which was a bare int. That int counted confirmed
changes carrying no message ids, on the reasoning that an edit which cannot
be netted must still register rather than be lost. It was what made the
count impossible to open and list, since a dialog would have shown three
groups and then owed the user a remainder it could not describe.
The remainder is empty. NotmuchWorker::applyTags() is the only emitter of
tagsApplied(), and its first statement returns on an empty id list, which is
the exact condition the counter required. applyTagsToThreads() resolves
threads to message ids through a query and errors out when that comes back
empty, so it can only ever hand applyTags() a non-empty list.
Measured rather than read. A qFatal in the branch fired in 4 of 70
test_mainwindow cases, all four building a TagChange by hand and invoking
the slot directly with no worker involved; an assertion before the worker's
own emit never fired across the whole suite, worker-backed tests included.
The worker's guard stays and is pinned where it lives, by
applyTagsWithNoIdsDoesNothing() in test_notmuchworker. The MainWindow test
that asserted the deleted branch is replaced by one for the consequence: a
change reaching the indicator names its messages, and an edit with its
inverse nets back to nothing, which is the property a growing-only counter
could never have. Three tests that leaned on the counter to show the
indicator now carry message ids, as a real edit always does.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P88Q3MCSCSQxKDy7pmXh9F
|
|
Restore moved the file back to the inbox folder and left it invisible: the
message carried no `inbox` tag, so the Inbox view could not see it, and the
user reported restoring a message and losing it.
Delete strips `inbox` so a deleted message leaves that view, which makes
restoring it the other half of the same change. restoreResolvedMessages()
already meant to add the tag back, and the comment above the branch
describes exactly this failure, but the comparison deciding it read
`origin`, which four lines earlier had been reassigned from the bare folder
name to the finished tag. `deleted-from:Inbox` never equals `Inbox` however
an account spells its inbox, so the branch was dead and the tag never came
back.
The destination folder is taken from the move's own key instead, which is
what the surrounding code already builds and what the comment says is being
compared.
The existing test passed against this throughout. It asserted the file
moved, the origin tag came off and `deleted` came off, all of which were
true; nothing asserted the tag that decides whether the user can see the
message afterwards. It does now, and fails against the old comparison.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P88Q3MCSCSQxKDy7pmXh9F
|
|
Hand-testing item 169 found four defects, three of them visible on every
card.
The initials were taken from whatever the card's first line held, which is
the raw From header on a reply row and notmuch's comma-joined author
summary on a thread row. A naive space split therefore gave `T<` for
`tsujan <notifications@github.com>` and one letter each from two different
people for `Standreas, tsujan`, and a separator counted as a word, so
`INE - Expert IT Training` drew `I-`. Avatar::initialsFor() now normalises
first: the angle-addr and any quoting go, a comma takes the first entry
unless the name is quoted, a bare address is not a name, and a word has to
carry a letter or a digit. Avatar::fillFor() uses the same normalisation,
so an address in the name's place no longer reads as a person.
The two-tone fill built its gradient axis as a radius from the centre, so
the 0.5 colour stop landed on the squircle's edge and one hue filled almost
the whole face. The axis spans the diameter now.
The fade ran left to right, which put its hard stop at 60% of the card and
read as a slab rather than a wash. It runs right to left: opaque at the
card's right edge, where the only hard stop is the card's own boundary, and
gone before it reaches the accent bar that already states the account.
And the flat views hashed the user's own address on every row, so every
Sent and Drafts card shared one pattern. ThreadSummary::firstMessageRecipient
rides the recipient fold, which already parses the To header, and
SenderAddressRole prefers it, falling back to the sender when there is no
usable To.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01P88Q3MCSCSQxKDy7pmXh9F
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Item 68, which turned out to be three things once its premise was
measured. The note asked to extend a "passed" subject rule to "Fw:";
there was no subject rule, and the correlation it rested on did not
exist. What did exist was a gap nobody had reported.
Reply and forward now flag their source. The Maildir R and P flags,
which every other client sets and notmuch reads back as "replied" and
"passed", had never been written here: measured on the developer's
index, all 317 "replied" and all 6 "passed" came from other clients.
ComposeWindow emits sourceMessageAnswered after a successful send and
MainWindow routes it through sendMessageTagChange, message-scoped and
off the undo stack, for the reason auto mark-read is: the flag records
that the mail went, and the send cannot be undone.
ComposeContext carries sourceMessageId rather than reusing inReplyTo,
which is deliberately empty on a forward so the recipient's client does
not file it under the thread it left. Keying on it made the "passed"
half dead code that compiled and never fired. A resumed draft is
excluded: its kind records how the file was opened, not what the user is
doing, so flagging on it would set R from a guess.
A received forward gets its own mark. Derived from the subject at paint
time, storing nothing and reaching no server, because "passed" means "I
forwarded this" and setting it from a guess would assert something false
on 222 existing messages. subjectIsForwarded() shares forwardSubject()'s
prefix table so the two cannot disagree, strips a Re: chain first, and
takes extra locale spellings from [general] forward_prefixes, which
extends the built-in table rather than replacing it.
A mutation survived the first round and corrected a claim in the code:
QRegularExpression::escape already makes a punctuation prefix inert, so
the word guard is not about pattern validity. It stops a configured "-"
matching "-: x". The comment and test say that now.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01LXCZFLXbAii5n5wtovpdhh
|
|
Item 168, found by the user while hand-testing 118: Delete could be
triggered on a message already in the trash. Not dangerous, which is how
it survived. moveMessages() finds the file already in the destination
and takes its early-return branch, so the message is reported as moved,
an unsynced change is counted, and nothing happened. Restore had the
mirror of the same problem, added unconditionally to both menus and so
offered on mail that was never deleted.
Each is now hidden where it has no meaning, which is the rule item 112
established for the unread entry. The question is about the PATH, never
the deleted tag: a message trashed by another client carries no such
tag, which is why the trash view is path-based, and asking the tag would
hide Delete on exactly the mail a trash view is full of.
Delete also removes unread now, at the user's request on the same
tangent. It travels inside the same sendMove() call rather than as a
second write, so one undo returns the folder and the tag together. This
rewrites the Maildir filename, because maildir.synchronize_flags is
true, and so reaches the server: the same mechanism the post-new hook
refuses to touch, and the difference is that the hook acts unattended on
arriving mail while this is an explicit gesture on a message in front of
the user.
A mutation survived the first round and found a real hole: comparing the
prefix without its trailing separator passed every test, because no
fixture had a folder whose name starts with the trash folder's. Under it
Delete silently vanished from mail in acct/trash-old, which is not the
trash. The fixture carries that row now and all three properties are
mutation-checked. The suite is 37 of 38, the failure being item 136 on
an unrelated path.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HFuRPtzFrSxCQjFk6tq7gD
|
|
Item 118, unblocked by 103. Message > Empty trash..., scoped to the
account selector, with no default shortcut.
purgeMessages() is a separate worker entry point from moveMessages()
rather than a flag on it, because the two look alike and only one can be
undone. It takes named ids, never a folder sweep, so the blast radius is
what the dialog enumerated and the user confirmed, and it deletes every
file of a message: notmuch deduplicates by Message-ID, so leaving one
behind leaves the message alive in the folder the user emptied.
It confirms, naming the count and the account, defaulting to Cancel.
That breaks CLAUDE.md's no-confirmation rule deliberately and the rule
now records it as its single exception, in the same paragraph: a purge
has no inverse to push onto the undo stack, so the protection the rule
provides has to come from somewhere, and the dialog is where.
Two defects found rather than reasoned. The count claimed messages whose
files were already gone, overstating an irreversible action; an absent
file is correctly not an error, but that is not the same as destroyed.
And the user's hand test found the list still showing mail that no
longer existed: a purge removes rows rather than changing them, so there
is no optimistic update to apply and nothing was connected to
messagesPurged at all. It re-runs the current query now.
Verified against the live index after the user emptied one real
account's trash: zero files on disk, zero in the index. The suite is 37
of 38, the failure being item 136 on an unrelated path. Ten new strings
translated, lrelease reports 0 unfinished.
Item 168 is filed from the same hand test, on Delete being offered on
mail already in the trash.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HFuRPtzFrSxCQjFk6tq7gD
|
|
Item 112, and 99 and 147 with it: the user's note is one design across
all three. A union is not a state. ThreadSummary::tags is notmuch's
union over the conversation, so a thread holding even one unread message
answered "unread" and the thread toggle always chose "mark read". There
was no input that reached "mark thread unread" on a mixed thread, which
is the thread a user wants it for.
The thread toggle becomes two absolute actions, mark_thread_read and
mark_thread_unread. Neither takes a default chord, at the user's choice:
Ctrl+Alt+U meant whichever direction the union picked, and since item
132 a shortcut is a chosen subset rather than a requirement. It is now
unbound.
The message-scoped toggle stays a toggle, because one message has a real
two-valued state, and its label now names the direction it will go. On a
selection with no single state the entry is hidden rather than labelled
wrongly, chosen over disabling it; the thread submenu is the route then,
and its entries are absolute.
selectionTagPresence() is the three-valued predicate that needed to
exist. everySelectedRowHasTag() delegates to it and keeps its two-valued
answer, which is all a direction needs; a label needs the third value.
The refresh is keyed on the model's dataChanged as well as on the
selection, so a write moves the label without reselecting and none of
the six optimistic-update call sites has to remember.
Three mutations fail: restoring the union predicate reports the user's
original symptom, showing the action on a mixed selection, and dropping
the dataChanged refresh. The suite is 37 of 38, the failure being item
136 on an unrelated path. Four new strings translated, lrelease reports
0 unfinished.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HFuRPtzFrSxCQjFk6tq7gD
|
|
The version alone cannot tell one build of an unreleased X.Y.Z from
another, and the user rebuilds and hand-tests unreleased builds daily.
They chose a counter over a git description: what they want to know is
that the binary is newer than the one they were running, not which
commit it came from.
QTMAILDIR_BUILD_NUMBER is a cmake option, ON by default, that runs
cmake/BuildNumber.cmake as a build step to increment a counter and write
buildnumber.h. It had to be a build step: configure_file runs once per
cmake run, so a counter interpolated into version.h.in would sit still
across exactly the rebuilds this exists to distinguish, which is why
version.h.in includes a second generated header rather than carrying the
number itself.
Two macros, and the split is load-bearing. QTMAILDIR_VERSION stays a
clean X.Y.Z and keeps the window title, applicationVersion and anything
that might ever compare versions; QTMAILDIR_VERSION_DISPLAY carries the
number and goes to the three surfaces the user picked, --version and
--help, the About dialog, and the placeholder pane. The window title was
offered and declined, since the number would then be in every
screenshot.
The counter lives in the build directory and is not tracked, so it
cannot conflict on a pull or leave the tree dirty; a fresh build
directory restarts at 1, which is honest, because it is a different
build tree. A release passes -DQTMAILDIR_BUILD_NUMBER=OFF and the header
is written empty. The SlackBuild in the my-slackbuilds repo needs that
flag and is a separate commit there.
Verified by running it, since none of this is reachable from a C++ test:
three consecutive builds reported build 2, 3 and 4, and a separate
Release configure with the option OFF reported a clean 0.27.0. Passing
the flag to a tree that does not have the option yet is an unused-cli
warning and exit 0, so the SlackBuild change is safe before 0.28.0
ships. The suite is 37 of 38, the one failure being item 136 on an
unrelated path.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01HFuRPtzFrSxCQjFk6tq7gD
|
|
A read-only notmuch handle is a Xapian snapshot taken when it is opened,
so it never observes a write made by another process afterwards. The
worker opened one handle and kept it for the process lifetime, which made
the sync script's `notmuch new` invisible: every query after startup was
answered from the index as it stood when the application launched.
The symptom was mail arriving while the window was open and not appearing
until a restart. It was not confined to the post-sync refresh, which is
what made it hard to place: a query typed by hand also found nothing,
since it hits the same handle. Tag writes were unaffected throughout,
because applyTags opens its own read-write handle per call.
Reopen in openReadOnly() rather than at each call site: every read path
begins by asking for the handle. A reopen failure is deliberately not
fatal, since the existing handle is still usable and answering from a
slightly stale index beats refusing to answer.
The suite could not reproduce this before: the test helper builds a fresh
worker per query, so it opens a fresh handle every time. The new test
holds one worker across two queries and indexes between them from a
second process.
Item 104.
|
|
Item 163. mbsync renames an uploaded file to add its `,U=<uid>` infix,
and the model's `MessageRef::filePath` was captured when the query ran,
so a row loaded before that sync names a file that no longer exists.
MimeParser then honestly reports a message it cannot open.
MaildirName::resolveRenamed() answers the filesystem question: returns
the path unchanged when it still exists, otherwise looks in that one
directory for the file whose unique stem matches. mbsync preserves the
stem (`<stem>:2,D` becomes `<stem>,U=5:2,D`), which is what makes this
safe to do by filename at all. It never recurses, never crosses a folder
boundary, and refuses an ambiguous match rather than guessing, since
opening or moving the wrong message is worse than reporting none.
It lives in MaildirName because that namespace already owns the `,U=`
infix and is a pure-value unit testable without a widget. A file that
changed FOLDERS is a different question that only the message id can
answer, and NotmuchWorker::moveMessages() re-resolves that way already.
Three call sites, all of which held a stale path:
- The message pane, which reported "(unreadable message)" over a file
that was on disk and readable. Cosmetic and self-repairing.
- Reply and Forward, refused outright, so the user could not answer a
message that was sitting there.
- The draft reopen, and this is the half that costs data. The refusal
happens BEFORE any composer exists, so the user composes again into a
fresh window whose autosave has no previous path to unlink. The old
revision survives, each save mints a new Message-ID, and both files
reach the server. The unlink machinery was correct throughout and
never ran.
forDraft() seeds draftPath from the RESOLVED path, never the caller's:
seeding the stale one would let the reopen succeed and the unlink still
miss, which is the same fork arriving one step later.
Covered by five unit tests on the resolver, including the two that keep
it honest (a genuinely missing file yields nothing, and a neighbouring
message is never matched), and by an integration test that renames the
draft the way mbsync does and asserts the file COUNT, which is the shape
the fork actually takes. Both mutation-checked; the integration test
fails with the reported symptom when the resolution is removed.
The stable-Message-ID question is deliberately untouched: it is what
turns a stale path into two server-side messages rather than one
replaced file, and it wants its own item.
Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>
Claude-Session: https://claude.ai/code/session_01UUQS6n3cmsFrsjCNmwNtf8
|
