[pikann]

Summary

Shows subtasks hierarchically in the Table (product backlog, sprints) and Timeline views, as requested in #622. Parent rows get a chevron and a done/total progress chip; expanding a row loads its children on demand, indented under the parent, with the view's filters and sort applied. Board cards show the same chip but stay flat.

Closes #622

What changed

API (services/api)

• Task list responses carry subtask_count and subtask_done_count (live direct children; children in a done-category status), computed in one batched query per page.
• New exclude_subtasks=true list filter: omits tasks whose parent is a live non-Epic task, so a subtask is not shown both nested and as its own row. Tasks that only belong to an Epic stay in the list. It also applies to the count endpoint.
• parent_task_id listing (already supported) keeps ignoring the sprint filter while status/assignee/sort filters apply; the web client relies on that.
• New view config field subtask_mode (collapsed | expanded | hidden; empty means collapsed), included in the per-user override merge/diff logic.
• docs/api/http-design.md updated.

Web (apps/web)

• TaskRow gets the chevron, progress chip and indentation (capped at 3 levels, tighter on mobile); SubtaskRows lazily fetches children with loading, error/retry, load-more, "N of M match the current filters" and an inline "Add subtask" row.
• The Timeline rows get the same chevron/chip and expand to their child tasks.
• View settings (Table and Timeline only) gets a Subtasks option: Nested, collapsed (default), Nested, expanded, Hidden (flat list). It is saved with the view like other settings (only for me / for everyone). Rows opened or closed by hand are remembered per user and per view in localStorage.
• Board cards show the progress chip inline with the other field chips.
• Translations added in all 9 locales.

Behaviour to be aware of

• While nesting is on, a child whose parent is hidden by a filter or sits in another group is not shown at the top level either (it appears when its parent does). Showing the parent as a dimmed context row is a possible follow-up.
• New projects have no Subtask system type, so subtasks are ordinary tasks linked by parent_task_id; nesting works off that link.

Type of Change

• Other (feature)

Testing

• Go: new repository tests for the subtask counts and exclude_subtasks.
• Web: subtask-rows.test.tsx, roadmap-nesting.test.tsx; full suite, tsc -b and biome pass.
• e2e: new features/projects/subtask-hierarchy.feature and tests/projects/subtask-hierarchy.spec.ts (24 tests, passing on chromium and mobile-chrome against the e2e stack); the existing interaction-views, view-settings and timeline specs still pass.
• Checked manually in the browser on the Backlog Table, Timeline and Board views.

Checklist

• The change is focused and scoped.
• Related documentation is updated.
• New structure or direction is explained clearly.
• I avoided unnecessary detail or premature abstraction.

🤖 Generated with Claude Code

[pullfrog]

Important

The new subtask_mode setting cannot be saved back to "Nested, collapsed" as a personal override while the shared view is "Nested, expanded" — the choice is silently discarded. One other lower-severity concern inline, plus a cache-invalidation note below.

Reviewed changes

Reviewed the full PR (33 files, 1 commit) end to end.

• API subtask counts and filter — ListTasks now attaches subtask_count / subtask_done_count per task through one batched query, and a new exclude_subtasks filter omits tasks whose live parent is a non-Epic task; applied to the list and count endpoints.
• View config subtask_mode — new collapsed / expanded / hidden field threaded through the entity, DTOs, and the per-user override merge/diff.
• Web nested rows — TaskRow gains a chevron, progress chip, and indentation; new SubtaskRows lazily fetches and recurses over children with the view's filters/sort and an inline "Add subtask"; the Timeline gets the same via RoadmapChildRows.
• View settings and i18n — a Subtasks selector for Table/Roadmap views, translations in 9 locales, and per-user/per-view expansion remembered in localStorage.
• Tests — Go repository tests for counts and exclude_subtasks, web unit tests, and a new e2e feature + spec.

ℹ️ Nested rows and the parent chip aren't refreshed by the task-detail modal

The new nested children query (childTasksInfiniteQueryOptions, key ["projects", id, "tasks", parentId, "children", filters]) is only invalidated through the broad ["projects", id, "tasks"] prefix from SubtaskRows.handleAdd, the list's field/status edits, and realtime. The task-detail modal's subtask callbacks still invalidate only the exact subtasksInfiniteQueryOptions(...).queryKey, so editing or adding a child there leaves the nested rows and the parent's done/total chip stale until a realtime task.* event (or the 15s staleTime) refetches them. If the acting client doesn't receive its own realtime event, this is user-visible.

Technical details

# Detail-modal subtask edits don't invalidate the nested child query

## Affected sites
- `apps/web/src/components/projects/interactions/task-detail/index.tsx:464-487` — `onSubtaskUpdate` / `onSubtaskCreate` invalidate the exact key `["projects", projectId, "tasks", task.id, "subtasks"]`.
- `apps/web/src/lib/interaction-api.ts:903-916` — the nested query uses the distinct key `[..., parentId, "children", filters]`, which the exact key does not match.
- `apps/web/src/components/projects/interactions/interaction-layout.tsx` — the parent's `subtask_count` / `subtask_done_count` chip lives on the top-level list (`allTasksQueryOptions`), also not matched by the exact key.

## Required outcome
A subtask change made from the detail modal should refresh both the nested rows and the parent's progress chip without relying on a realtime round-trip.

## Suggested approach
Change the two modal invalidations to the prefix `{ queryKey: ["projects", projectId, "tasks"] }`, matching the pattern already used by `SubtaskRows.handleAdd`.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes

Reviewed the delta from 24b1c1c to 8943b40; the prior review's three findings (one important, two informational) are all addressed, and no new issues were introduced.

• Persisted an explicit "Nested, collapsed" personal override — a new pickSubtaskMode helper stores the literal "collapsed" when the shared view is "expanded"/"hidden" (and only leaves it unset when the shared default is already collapsed), fixing the silent discard, with unit tests in pick-subtask-mode.test.ts.
• Scoped the new parent-based reads to the project — exclude_subtasks now requires sp.project_id = tasks.project_id, and attachSubtaskCounts joins the parent on a matching project_id; the repository tests add a cross-project child to cover both paths.
• Widened the detail modal's subtask invalidation — onSubtaskUpdate / onSubtaskCreate now invalidate the ["projects", projectId, "tasks"] prefix, so the nested rows and the parent's progress chip refresh without waiting on a realtime round-trip.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

Important

The new Epic exclusion in attachSubtaskCounts zeroes subtask_count for Epic parents, which is exactly the signal the Timeline (Roadmap) view reads to render the chevron/chip and to expand an Epic into its child tasks. Because the Timeline's default view lists Epics only, its nested-child feature is now unreachable — and the PR's own e2e scenario "An Epic shows its child count and expands to its tasks" would fail.

Reviewed changes

Reviewed the delta from 8943b40 to 3327b82 (Go repository/service/DTO/handler, the web subtask store and API, docs, and a migration). One must-address regression, flagged inline.

• Sprint-scoped exclude_subtasks — the NOT EXISTS now requires the parent to share the child's sprint (sp.sprint_id IS NOT DISTINCT FROM tasks.sprint_id), so a child whose parent is in another sprint stays in the flat list.
• Epics excluded from attachSubtaskCounts — the batched count query now skips Epic parents, and docs/api/http-design.md states Epics report no subtask_count.
• View config normalization — normalizeViewConfig drops unknown subtask_mode values on create, update, and the per-user override path.
• Filter/DB plumbing — exclude_subtasks is ignored when parent_task_id is set, a partial index on tasks(parent_task_id) is added, and the Go tests cover the new sprint scope and the Epic skip.
• Web — the inline "Add subtask" now inherits the parent's sprint; the manual expand/collapse record drops an entry once it matches the view default.

  ^{｜ Fix all ➔ ｜ Fix 👍s ➔ ｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}

[pullfrog]

✅ No new issues found.

Reviewed changes

Reviewed the delta from 3327b82 to d331f33; the prior review's must-address regression is fixed and no new issues were introduced.

• Restored Epic subtask_count on the server — the Epic predicate was removed from attachSubtaskCounts, so Epics again report their child counts and the Timeline's chevron/chip and child expansion keep working; the Go test now asserts an Epic count of 1.
• Moved the double-display guard to the Table client — ListGroup.renderTaskRow zeroes the counts for a row whose type is the system Epic, since an Epic's children are already listed flat there, so the Table shows no chevron/chip for Epics without touching the server count the Timeline depends on.
• Updated the API docs — docs/api/http-design.md no longer claims Epics report no subtask_count.

  ^{｜ View workflow run ｜ Using deepseek-v4.1-flash (free via Pullfrog for OSS) ｜ 𝕏}