Jobs: the delete dialog and the chain-end states #176

Closed
opened 2026-08-23 19:46:05 +00:00 by nalum · 0 comments
Owner

Part of the jobs flow redesign. The design handoff (design_handoff_jobs_flow, held outside the repo) is the source; the facts it depends on are repeated here so this issue stands alone.

Delete is a hard delete with no undo, so it always asks, and the dialog names what survives it.

A plain job asks plainly. A job with sub-jobs counts them in the question and in the button (Delete all 13), calls out how many are already ticked, and states what happens to the points: the earner keeps what they banked, and the linked reward returns to the rewards list. The cascade is real — sub-jobs go in the same transaction (internal/services/item/delete.go) — and unlinking the unclaimed rewards has now landed on the server, so the copy is true.

Afterwards a notice rather than a snackbar, because there is no Undo to offer.

The chain-end states. Ticking a repeating job mints its successor, and that is the one other act that confirms first. The completed sheet then names the successor and offers it. If un-completion is attempted on a job that already spawned one the API refuses (domain.ErrChainMovedOn); surface it as Can't reopen this — Tue 16 Sep took over. Open that one instead. A bare refusal strands the reader.

A deleted repeating job ends its chain. A deleted completed job leaves its successor alone; they are separate records.


Update — the server side is done, in review

The dialog's second line ("the 22 points reward goes back to the rewards list")
was not true against the shipped server: delete left the reward pointing at a
dead row, and the eligibility filter then hid it for good.

That is fixed in PR #185 — unlinking joins the delete cascade for the job
and each cascaded sub-job, while claimed rewards keep their link as settled
history. Two integration specs cover it. The copy ships as written once that
merges; do not build this dialog against the old behaviour.


Two requirements added after #213 shipped

The job sheet went out to the family's cluster three times before it was right.
Seven visual defects, none caught by a suite: pickers that unfolded forms
instead of opening, three Save buttons on a commit-on-blur screen, (optional)
labels, a reward's name used as a field label, an unstyled add row, checkboxes
with a ghosted tick, and a create form with a browser spinner. Every one had to
be found by a person looking at a screen.

1. Screenshot goldens are part of this ticket

Record gallery (Playwright) and Roborazzi cases for every surface this ticket
builds or visibly changes, and treat them as a deliverable rather than a
follow-up. Three of #213's defects — the unstyled add row, the ghosted
checkboxes, the spinner in the create form — were pixel faults that a golden
would have caught before a deploy did.

#172 deferred its goldens on the reasoning that later tickets would re-record
them. That reasoning was sound then and is not now: the row grammar and the
pickers are settled, so an image recorded here stays valid.

Mechanics, from AGENTS.md: a new case rides in the same commit as the code
that adds it — a commit that adds a gallery or Roborazzi case without its image
is red on its own, because check / web runs npm run test:screens and
android / build runs verifyRoborazziDebug. A re-recorded existing case
goes in its own commit, so the reviewable diff stays code.

2. Check for leaked form styling before you finish

Every visual defect on #213 came from one place: the shipped app's form
fragments reused inside the new row grammar. They pass every test, because the
tests assert the pieces exist rather than that the screen reads right.

Before you call this done, grep the surfaces you touched and report what you
found and what you did about each:

  • a raw <input> / OutlinedTextField / BasicTextField outside the row and
    picker components — a form field dropped into a list of rows
  • type="number", which renders browser spinner arrows
  • btn, link-btn, btn-quiet or TextButton where a row or a quiet row
    verb belongs — underlined links reading as broken text
  • (optional) in any label — §8 names that pattern as the form admitting it
    should not have asked
  • any Save button on a surface that commits on blur, including Save changes and Save the points
  • a disabled-and-grey control where law 5 wants it absent
  • a record's title used as a field row's label — that column is for a
    field's name (Who, When, Worth), and a long title wraps in it

If your ticket is the first to land after this was written, add the check to
AGENTS.md beside the goldens rule, so it stops being a thing I remember to say.

Part of the jobs flow redesign. The design handoff (`design_handoff_jobs_flow`, held outside the repo) is the source; the facts it depends on are repeated here so this issue stands alone. Delete is a hard delete with no undo, so it always asks, and the dialog names what survives it. A plain job asks plainly. A job with sub-jobs counts them in the question **and in the button** (`Delete all 13`), calls out how many are already ticked, and states what happens to the points: the earner keeps what they banked, and the linked reward returns to the rewards list. The cascade is real — sub-jobs go in the same transaction (`internal/services/item/delete.go`) — and unlinking the unclaimed rewards has now landed on the server, so the copy is true. Afterwards a notice rather than a snackbar, because there is no Undo to offer. **The chain-end states.** Ticking a repeating job mints its successor, and that is the one other act that confirms first. The completed sheet then names the successor and offers it. If un-completion is attempted on a job that already spawned one the API refuses (`domain.ErrChainMovedOn`); surface it as `Can't reopen this — Tue 16 Sep took over. Open that one instead.` A bare refusal strands the reader. A deleted repeating job ends its chain. A deleted completed job leaves its successor alone; they are separate records. --- ## Update — the server side is done, in review The dialog's second line ("the 22 points reward goes back to the rewards list") was not true against the shipped server: delete left the reward pointing at a dead row, and the eligibility filter then hid it for good. That is fixed in **PR #185** — unlinking joins the delete cascade for the job and each cascaded sub-job, while claimed rewards keep their link as settled history. Two integration specs cover it. The copy ships as written once that merges; do not build this dialog against the old behaviour. --- ## Two requirements added after #213 shipped The job sheet went out to the family's cluster three times before it was right. Seven visual defects, none caught by a suite: pickers that unfolded forms instead of opening, three Save buttons on a commit-on-blur screen, `(optional)` labels, a reward's name used as a field label, an unstyled add row, checkboxes with a ghosted tick, and a create form with a browser spinner. Every one had to be found by a person looking at a screen. ### 1. Screenshot goldens are part of this ticket Record gallery (Playwright) and Roborazzi cases for every surface this ticket builds or visibly changes, and treat them as a deliverable rather than a follow-up. Three of #213's defects — the unstyled add row, the ghosted checkboxes, the spinner in the create form — were pixel faults that a golden would have caught before a deploy did. `#172` deferred its goldens on the reasoning that later tickets would re-record them. That reasoning was sound then and is not now: the row grammar and the pickers are settled, so an image recorded here stays valid. Mechanics, from AGENTS.md: a **new** case rides in the same commit as the code that adds it — a commit that adds a gallery or Roborazzi case without its image is red on its own, because `check / web` runs `npm run test:screens` and `android / build` runs `verifyRoborazziDebug`. A **re-recorded existing** case goes in its own commit, so the reviewable diff stays code. ### 2. Check for leaked form styling before you finish Every visual defect on #213 came from one place: the shipped app's form fragments reused inside the new row grammar. They pass every test, because the tests assert the pieces exist rather than that the screen reads right. Before you call this done, grep the surfaces you touched and report what you found and what you did about each: - a raw `<input>` / `OutlinedTextField` / `BasicTextField` outside the row and picker components — a form field dropped into a list of rows - `type="number"`, which renders browser spinner arrows - `btn`, `link-btn`, `btn-quiet` or `TextButton` where a row or a quiet row verb belongs — underlined links reading as broken text - `(optional)` in any label — §8 names that pattern as the form admitting it should not have asked - **any Save button on a surface that commits on blur**, including `Save changes` and `Save the points` - a disabled-and-grey control where law 5 wants it absent - a record's title used as a field row's `label` — that column is for a field's name (`Who`, `When`, `Worth`), and a long title wraps in it If your ticket is the first to land after this was written, add the check to AGENTS.md beside the goldens rule, so it stops being a thing I remember to say.
nalum closed this issue 2026-08-26 18:35:12 +00:00
Sign in to join this conversation.
No milestone
No project
No assignees
1 participant
Notifications
Due date
The due date is invalid or out of range. Please use the format "yyyy-mm-dd".

No due date set.

Dependencies

No dependencies set

Reference
eagraiclainne/app#176
No description provided.