Phase 5c slice A. Pulls the structural rules out of the Postgres
triggers into a Go-side validator. The trigger stays as defence in
depth; the validator is the human-facing error path.
- docs/plans/itemwrite-validation.md enumerates every rule the
triggers in 0001 + 0010 enforce, with the ValidationError.Kind
callers will see for each. Eleven rules total (two SQL-only safety
rails kept untranslated).
- internal/itemwrite/itemwrite.go: ValidationError + Input + Reader
interface + ValidateFormat (pure: missing fields, slug format,
status whitelist, self-parent) + ValidateAgainstStore (DB-aware:
unknown-parent, slug-collision under any common parent, cycle via
ancestor-closure DFS capped at 64 hops to mirror the trigger).
- Eight kind constants exported: missing-required, invalid-slug-format,
invalid-status, slug-collision, cycle, self-parent, unknown-parent,
unresolvable-path.
Tests cover every kind on both happy and reject paths: missing /
whitespace fields, slug containing dot / upper / whitespace, invalid
status enum, self-parent guard, unknown parent id, root slug collision,
sibling slug collision under common parent, cycle on ancestor closure,
and the "Reader returns ListAll error → validator returns nil" path
(callers see the infra error later, validator doesn't mask it).
No caller migrates yet. Same Go-linker DCE caveat as 5a/5b slice A:
`strings <binary> | grep internal/itemwrite` returns 0 until slice B
imports.
Task: t-projax-5c-itemwrite
Phase 5b slice A. Generic TTL cache that replaces the mechanically
identical dashboardCache + timelineCache in slices B/C.
- TTLCache[V] over map[string]entry[V] with sync.RWMutex.
- Get / Set / Invalidate(key) / InvalidateAll.
- Lazy expiry — a Get past the deadline removes the entry; no sweeper
goroutine (matches today's behaviour and stays simple at single-user
scale).
- Nil receiver is safe across all four methods — same defensive shape
the existing per-package caches use.
Tests cover empty Get, Set+Get, expiry on miss, overwrite,
keyed-Invalidate isolation, InvalidateAll, nil receiver, pointer
payload behaviour, and a -race-flag concurrent-access probe across
8 workers × 200 ops.
No web/mcp wiring yet — slices B/C migrate the callers. Same Go
linker DCE caveat as 5a slice A applies (strings | grep alone won't
fire on this slice).
Task: t-projax-5b-cache
Phase 5a slice D. The MCP timeline tool no longer depends on
*web.Server — it talks to *aggregate.Aggregator directly. The wrong-way
mcp → web layering that necessitated the TimelineBuilder interface is
gone.
- mcp/tools.go: TimelineBuilder interface deleted.
RegisterProjaxTools(s, st, agg *aggregate.Aggregator) now takes the
aggregator directly; passing nil keeps the timeline tool unregistered
(kill-switch contract unchanged).
- mcp/tools.go: TimelineArgs moved from web/ to mcp/ since it is the
MCP-facing input shape. The timeline tool runs the full pipeline:
store.ListByFilters → in-mem timeline-exclude + has-link narrowing →
agg.All(...) → Result.ToTimelineRows() → aggregate.BuildTimelineDays
→ timelineView. No web/ import in the timeline path.
- internal/aggregate/rows.go: new Result.ToTimelineRows() helper that
projects the typed rows into the flat TimelineRow sum-type both
web/timeline.go and mcp/tools.go consume. Single source of truth for
the Date-anchor choice across kinds.
- internal/aggregate/timeline_days.go: FormatPERDate lifted from web/
so timeline-row builders outside web/ can render PER strings without
re-importing web/.
- web/timeline.go: BuildTimelinePayloadFromArgs + TimelineArgs deleted
(no remaining callers — slice D inlined the MCP path).
- cmd/projax/main.go: pass srv.Aggregator() into RegisterProjaxTools.
MCP tree-filter parity note: the move to store.ListByFilters narrows
status to a single value (first of args.Status) and AND-matches
management (vs the web TreeFilter's OR). m's documented MCP uses
(tag + default status) round-trip identically. Logged as a footnote in
docs/plans/aggregator-refactor.md.
All mcp + web + aggregate tests green.
Task: t-projax-5a-aggregator
Phase 5a slice A: a new package that concentrates the "fan out across
linked items" pattern web/dashboard.go, web/timeline.go and mcp/tools.go
each had separate copies of. No callers touch it yet — slices B/C/D
migrate them in turn.
- Aggregator with five methods (Todos/Events/Issues/Docs/Creations) plus
All convenience for the MCP timeline. Each method takes a *store.Item
slice and (optionally) a Window, returns typed Row slices.
- Row types embed the underlying caldav.Todo / caldav.Event / gitea.Issue
so existing html/template field accesses (.Todo.UID, .Event.Summary,
…) keep resolving via Go field promotion in slices B/C.
- TimelineRow sum-type wrapper (with pointer slots per Kind) plus the
flat template-friendly fields. Lifted-but-untouched from web/.
- BuildTimelineDays + SortTimelineRows + EventStartLabel +
EventDurationHint lifted near-verbatim from web/timeline.go.
- CalDAV/Gitea/Store interfaces in the aggregator so unit tests stub IO
cleanly. Real *caldav.Client / *gitea.Client / *store.Store satisfy
by method set.
- Per-source error handling preserved: log at WARN + skip the bad
fetch, return surviving rows.
Tests cover empty inputs, fan-out call counts, per-source error
recovery, window narrowing for todos, issue-cache hit path, doc/creation
allow-list filtering, BuildTimelineDays asc/desc order, sticky pills,
far-future fade, within-day sort.
Plan doc captures the slicing strategy + design decisions:
docs/plans/aggregator-refactor.md.
Task: t-projax-5a-aggregator
