Files
lanework/Kanban/UI/Card/CardTitleEditSession.swift
T
rzen c27cc93ec1 The card window's title learns Edit mode — a sibling session rides the body's own doors
CardWindowView's header is now the title, live: a static Text in Preview, an editable
single-line TextField in Edit, both reading a new CardTitleEditSession's buffer rather
than the card's own snapshot value — the same "buffer outranks the snapshot" reason the
body surface already reads bodySession.text instead of card.body.

CardTitleEditSession is CardBodyEditSession's shape one field over: the one-isDirty write
gate, dirty-buffer-wins on adopt(diskTitle:), the ~700ms injectable debounce, flush() /
flushOrThrow() for DirtyBufferGuard, and beginEditSession()/endEditSession() with a
session-coalesced undo step (one per session, never per debounced tick). It rides the
body's own begin/end/flush doors rather than opening a second session boundary — title is
only ever editable while the body column is in Edit mode — because the two write through
different WriteOperations with different validation and merging them would conflate two
unrelated frontmatter keys behind one buffer.

BoardStore.commitCardTitle(inCard:title:) reuses the same private setTitle helper and the
same .rename WriteOperation the board's own inline rename commits through, so trimming,
empty-removes-the-key, unchanged-writes-nothing and banner enrichment are one code path,
not a re-implementation. It resolves through cardBodyTarget (spans lanes and the trash),
not boardItem (board only), because a card window's title field stays live through the
same dismissal-into-trash flush the body already gets — the one deliberate divergence
from the board's own rename, which treats a trashed target as vanished.
registerTitleEdit(inCard:priorTitle:newTitle🔛) mirrors registerBodyEdit, anchored by
card identity and the already-reserved ExpectedField.title, folding into the same
one-coarse-step-per-window-close undo model.

Every place the body's Edit buffer flushes, the title's now does too: mode exit
(bodyPresentation.flushEdits/beginEdits), window close and the dismissal path
(CardWindowSession.endSession()), raw-source entry (configureRawSource), the close-time
DirtyBufferGuard modal (attemptSave tries body then title), and the fast-path close gate
(closeAfterFlushing() now checks title.isDirty alongside body.isDirty).
holdsUnsavedContent and settlement carry the title too.

Tests: CardTitleEditSessionTests.swift mirrors CardBodyEditSessionTests.swift (write
gates, normalization and newline-stripping, dirty-buffer-wins, debounce, undo-step
coalescing). CardTitleWriteTests.swift covers commitCardTitle/registerTitleEdit:
byte-identity no-op, empty-removes-key, vanished, the trashed-card-is-still-writable
divergence, a readable-but-uneditable target refusing and bannering, and a read-only
board suspending quietly. CardSessionUndoTests.swift gains coverage that a title edit
folds into the coarse close step alongside a body edit and registers on the window's own
stack, never the board's. RawSourceTests.swift's hand-wired rig picks up the title
session configureRawSource now also flushes.

3148 KanbanTests pass, 0 failures.

Claude-Session: https://claude.ai/code/session_014PtZdPwqZuqEDLc6wZMtEy
2026-08-09 10:37:55 -04:00

258 lines
12 KiB
Swift

import Foundation
import Observation
// MARK: - CardTitleEditSession
/// The card window header's title field — **a body-edit session in miniature**
/// (`CardBodyEditSession`'s own phrase for `CommentEditSession`, applied one level up), scoped to the
/// single line the header shows in Edit mode.
///
/// ### What it borrows from `CardBodyEditSession`, verbatim
///
/// The buffer/disk pair, the single write predicate (*write if and only if the buffer differs from
/// disk*), dirty-buffer-wins on `adopt(diskTitle:)`, the ~700 ms trailing debounce, `flush()` /
/// `flushOrThrow()`, and the begin/end pair whose end registers the session's one undo step. **There
/// is no session boundary of this type's own** — the card's title is only ever editable while the
/// body column is in Edit mode (m6-card-body's rule, stated on `CardWindowView`'s header), so this
/// session's `beginEditSession()` / `endEditSession()` are called from the exact places
/// `CardBodyEditSession`'s are (`CardWindowHost.configureSession`'s `bodyPresentation.flushEdits` /
/// `.beginEdits`, and `CardWindowSession.endSession()`), never on a schedule of their own. That is the
/// literal reading of "title dirtiness rides the same gates/flush points as body".
///
/// ### What differs, and why it is a sibling rather than a merge
///
/// The write target is a different `WriteOperation` — a title commits through the same `.rename`
/// vocabulary the board's inline rename uses (`BoardStore.commitCardTitle`, `BoardStore.setTitle`),
/// never `.editBody`. Folding the two into one session would mean one buffer standing for two
/// unrelated frontmatter keys with two different validation rules (a title trims and empty-removes;
/// a body does neither), so a sibling object — driven by the same triggers, writing through its own
/// seam — is the shape that keeps each write's rules in one place.
///
/// ### The value is `String?`, so "not yet captured" needs its own bit
///
/// `CardBodyEditSession.sessionOriginBody` uses `nil` for "no save has landed this session" because a
/// body is never itself optional. A title's on-disk value legitimately *is* `nil` — a card can be
/// untitled — so this type keeps `sessionHasOrigin` beside `sessionOriginTitle` rather than reusing
/// `nil` for two different meanings.
///
/// ### Single-line, by construction and by filter
///
/// A plain `TextField` already turns Return into a submit rather than a newline (`CardWindowView`'s
/// header wires `.onSubmit` to `flush()`), which is the whole of "Return ends editing rather than
/// inserting" on the keystroke that matters. `edited(_:)` additionally strips `\n`/`\r` from
/// whatever arrives — typing cannot produce either, but a paste can — so the buffer this type shows
/// and writes is single-line even when the text came in some other way. Trimming leading/trailing
/// whitespace and turning an empty result into `nil` happen at commit (`normalize(_:)`), mirroring
/// the board's inline rename exactly (`BoardStore.commitRename`): a title of three spaces reads as
/// untouched, never as a real (if blank) name.
@MainActor
@Observable
public final class CardTitleEditSession {
// MARK: State
/// What the field is showing — raw, as typed, never trimmed. Preview's header reads this too
/// (`CardWindowView.bodyColumn`), for `CardBodyEditSession.text`'s own reason: a flushed buffer is
/// ahead of the snapshot by a reload, so the buffer is the truer of the two in both modes.
public private(set) var text: String = ""
/// What the last snapshot said the card's `title` key holds. `nil` is untitled, a legitimate
/// resting state rather than a missing reading.
public private(set) var disk: String?
/// Whether the buffer, once normalized the way a commit would, differs from disk. Comparing the
/// *normalized* text rather than the raw one is what keeps typing spaces around an untouched title
/// from reading as dirty — `commitRename`'s own "whitespace commits as empty" rule, read one level
/// earlier.
public var isDirty: Bool { Self.normalize(text) != disk }
/// Whether an Edit session is open right now — `CardBodyEditSession.isEditing`'s own bookkeeping,
/// kept here too because this type's undo step coalesces the same way body's does: one step per
/// session, captured from the first landed save to the session's end.
public private(set) var isEditing = false
// MARK: Seams
/// The debounce interval — body's own ~700 ms default, and settable for the same reason
/// (`DragSession.holdTimeout`'s precedent: production default on the property, a test dials it
/// down).
@ObservationIgnored
public var debounceInterval: Duration = .milliseconds(700)
/// Where a save goes — `BoardStore.commitCardTitle(inCard:title:)`, filled in by the window once it
/// has joined its board. Takes the *normalized* title (trimmed, empty turned to `nil`), which is
/// what the store's own write gate compares against the card's current title.
@ObservationIgnored
public var save: ((String?) -> CardBodyWriteOutcome)?
/// Where this session's one undo step goes, called at `endEditSession()` with the title disk held
/// when the session's first save landed and the title it holds now —
/// `CardBodyEditSession.registerUndo`'s seam, one field over.
@ObservationIgnored
public var registerUndo: ((_ priorTitle: String?, _ newTitle: String?) -> Void)?
/// Whether a save has landed this session — `CardBodyEditSession.sessionOriginBody`'s `nil` check,
/// spelled as its own flag because `String?` is a real value here, not a "not captured" marker.
@ObservationIgnored
private var sessionHasOrigin = false
/// What disk said before this session's first landed save, when `sessionHasOrigin` is `true`.
@ObservationIgnored
private var sessionOriginTitle: String?
@ObservationIgnored
private var pending: Task<Void, Never>?
/// How many saves have actually been attempted through `save`.
@ObservationIgnored
public private(set) var saveAttempts = 0
public init() {}
// MARK: - Disk → buffer
/// A snapshot arrived. Dirty-buffer-wins, `CardBodyEditSession.adopt(diskBody:)`'s exact shape:
/// `disk` always follows it; `text` follows it only when the buffer had nothing unsaved.
public func adopt(diskTitle: String?) {
let wasDirty = isDirty
disk = diskTitle
guard !wasDirty else { return }
let shown = diskTitle ?? ""
if text != shown { text = shown }
}
// MARK: - Buffer → disk
/// The field changed. Strips embedded newlines first (paste's own hazard — typing a single Return
/// never reaches here at all, since a plain `TextField` treats it as a submit), then restarts the
/// debounce, or cancels it when the change brought the buffer back to what a commit of it would
/// leave on disk.
public func edited(_ newText: String) {
let sanitized = Self.stripNewlines(newText)
guard text != sanitized else { return }
text = sanitized
guard isDirty else {
cancelPending()
return
}
scheduleSave()
}
/// Saves now if there is anything to save, cancelling the pending debounce first — the mode-exit
/// and window-close flush, `CardBodyEditSession.flush()`'s own reason.
@discardableResult
public func flush() -> CardBodyWriteOutcome {
cancelPending()
return saveNow()
}
/// The start of one Edit session, called alongside `CardBodyEditSession.beginEditSession()` from
/// the same door (`CardBodyPresentation.beginEdits`). Idempotent for the same reason: a
/// re-published focus value or a redundant menu validation must not renotify observers over an
/// already-open session.
public func beginEditSession() {
guard !isEditing else { return }
isEditing = true
}
/// The end of one Edit session, called alongside `CardBodyEditSession.endEditSession()` from the
/// same door. Flushes, then registers the session's one undo step when something of this session's
/// actually landed and left a net change.
@discardableResult
public func endEditSession() -> CardBodyWriteOutcome {
let outcome = flush()
if sessionHasOrigin, sessionOriginTitle != disk {
registerUndo?(sessionOriginTitle, disk)
}
sessionHasOrigin = false
sessionOriginTitle = nil
if isEditing { isEditing = false }
return outcome
}
/// Throws the buffer away and takes disk's word for it — the Discard branch of the save-or-discard
/// step, `CardBodyEditSession.discardBuffer()`'s exact shape.
public func discardBuffer() {
cancelPending()
text = disk ?? ""
sessionHasOrigin = false
sessionOriginTitle = nil
if isEditing { isEditing = false }
}
/// `DirtyBufferGuard`'s `attemptSave`, folded together with the body's own in
/// `CardWindowSession.init()`: the same flush, with a real failure raised instead of reported. The
/// three non-failures are `CardBodyEditSession.flushOrThrow()`'s own — written/unchanged land the
/// text or need not, vanished has nowhere to write, suspended is the standing read-only lock row
/// rather than a failure — and none of them is a reason to block the close.
public func flushOrThrow() throws(BoardWriteError) {
switch flush() {
case .written, .unchanged, .vanished, .suspended:
return
case let .failed(error):
throw error
}
}
// MARK: - Private
private func scheduleSave() {
cancelPending()
let interval = debounceInterval
pending = Task { [weak self] in
try? await Task.sleep(for: interval)
guard !Task.isCancelled, let self else { return }
self.pending = nil
_ = self.saveNow()
}
}
private func cancelPending() {
pending?.cancel()
pending = nil
}
/// The gate, and the one place `save` is called. A landing moves `disk` up to the normalized text
/// that landed, so the echo arriving a reload later finds the buffer already clean.
private func saveNow() -> CardBodyWriteOutcome {
guard isDirty else { return .unchanged }
guard let save else { return .vanished }
saveAttempts += 1
let priorTitle = disk
let normalized = Self.normalize(text)
let outcome = save(normalized)
switch outcome {
case .written:
if !sessionHasOrigin {
sessionHasOrigin = true
sessionOriginTitle = priorTitle
}
disk = normalized
case .unchanged:
disk = normalized
case .suspended, .vanished, .failed:
break
}
return outcome
}
/// Strips `\r\n`, `\n` and `\r` outright — the single-line field's own hard rule, applied to
/// whatever arrives rather than only to what a commit would write, so the on-screen text never
/// shows a line break a paste tried to bring in.
private static func stripNewlines(_ raw: String) -> String {
raw
.replacingOccurrences(of: "\r\n", with: "")
.replacingOccurrences(of: "\n", with: "")
.replacingOccurrences(of: "\r", with: "")
}
/// Trimmed, empty turned to `nil` — `BoardStore.commitRename`'s own reading of a typed title,
/// applied here so `isDirty` and the value handed to `save` agree with what the board's inline
/// rename would do with the same keystrokes.
private static func normalize(_ raw: String) -> String? {
let trimmed = raw.trimmingCharacters(in: .whitespacesAndNewlines)
return trimmed.isEmpty ? nil : trimmed
}
}