c9aab19a97
- Scan history browser (history.js, GET /api/db/sessions, get_sessions(), get_session_items(ref_scan_id)) — review any past session without rescanning - User-scoped viewer tokens (#34) — scope by individual employee across M365 and GWS; autocomplete from Accounts list; dual-email support - Fix: GWS scan never marked finished (end_scan → finish_scan) and emitted wrong SSE event (scan_done → google_scan_done), excluding GWS items from all exports - Fix: file scan begin_scan called with wrong keyword args (TypeError swallowed), so local/SMB items were never written to DB - Fix: Graph sendMail reported failure on success — _post() now returns {} on empty 202 response instead of raising JSONDecodeError - Fix: Graph error hidden behind generic "No SMTP host" message when both Graph and SMTP were unavailable - Fix: Gmail vs Google Workspace SMTP error messages distinguished by username domain; Workspace errors point to admin console, not personal security settings - Docs: update README, MANUAL-EN, MANUAL-DA, CLAUDE.md, TODO.md, CHANGELOG.md Co-Authored-By: Claude Sonnet 4.6 <noreply@anthropic.com>
88 lines
6.6 KiB
Markdown
88 lines
6.6 KiB
Markdown
# TODO — Pending features and sustainability
|
||
|
||
Quick overview of what's still to be done.
|
||
|
||
---
|
||
|
||
## Recently completed
|
||
|
||
### Memory exhaustion during large M365 scans ✅
|
||
Six root causes fixed in `scan_engine.py` and `document_scanner.py`:
|
||
- Email body HTML stripped at collection time (`body` key deleted from each message dict before it enters `work_items`; plain text stored as `_precomputed_body` instead)
|
||
- `work_items` list converted to a `deque` before processing so each item is released immediately after `popleft()`
|
||
- `del content` added in file-processing branch as soon as raw bytes are no longer needed (before NER/PII counting)
|
||
- `del body_text` added after email body is fully consumed
|
||
- PDF OCR page images (`PIL.Image`) nulled out one by one after OCR instead of holding all pages in RAM
|
||
- Memory guard using `psutil` skips file downloads when < 300 MB RAM is available
|
||
|
||
**Still open:** The collection phase itself is still a "gather all, then process" loop. For very large tenants (>500k emails) the pre-extracted plain text in `work_items` could still be significant. The complete fix is to process each user's emails/files inline as they are fetched (generator/streaming pattern) rather than accumulating them into `work_items` first — estimated 1–2 days of refactor.
|
||
|
||
---
|
||
|
||
## Pending
|
||
|
||
### #15 — Scan profiles ✅
|
||
Named, reusable scan configurations. Full spec in SUGGESTIONS.md §15.
|
||
**Size:** Large · **Priority:** High
|
||
|
||
### #23 — Google Workspace role classification + cross-platform identity mapping ✅
|
||
Full spec in SUGGESTIONS.md §23.
|
||
**Size:** Large · **Priority:** Medium
|
||
|
||
### #27 — Migrate i18n format from `.lang` to JSON ✅
|
||
Full spec in SUGGESTIONS.md §27.
|
||
**Size:** Medium · **Priority:** Low
|
||
|
||
### #29 — Rename `skus/` → `classification/` ✅
|
||
Full spec in SUGGESTIONS.md §29.
|
||
**Size:** Small · **Priority:** Low
|
||
|
||
### #33 — Read-only viewer mode with PIN/token URL ✅
|
||
A shareable URL (token-protected) or numeric PIN that gives a DPO, school principal, or compliance coordinator read-only access to the results grid — with disposition tagging but without scan controls, credentials, or delete access. Full spec in SUGGESTIONS.md §33.
|
||
**Size:** Medium · **Priority:** Medium
|
||
|
||
### OneDrive 404 errors — investigate and handle appropriately ✅
|
||
404 on `drive/root/delta` during delta scans was being broadcast as a red `scan_error`. Root cause: `_get()` hit `raise_for_status()` for 404s, which fell through to the generic `except Exception` handler in `_scan_user_onedrive`. The full-scan path silently swallowed the same 404 via `except Exception: return` in `_iter_drive_folder_for`.
|
||
|
||
Fixed by adding `M365DriveNotFound(M365Error)` exception, raising it from `_get()` on 404, and catching it explicitly in `_scan_user_onedrive` with a lower-severity `scan_phase` broadcast ("OneDrive (user): not provisioned — skipped") instead of a red error card.
|
||
|
||
---
|
||
|
||
### #34 — User-scoped viewer tokens ✅
|
||
Viewer token scope extended to `{"user": ["m365@…", "gws@…"], "display_name": "Alice Smith"}`, filtering `flagged_items` by `account_id IN (list)`. Lets a single employee see only their own flagged files across both M365 and Google Workspace.
|
||
|
||
**Implemented:**
|
||
1. Scope format — `user` is a list of email strings (one per platform); `display_name` stored for UI display. Legacy single-string format coerced to list automatically.
|
||
2. Token creation UI — scope-type selector (`All` / `Role` / `User`) reveals either the role select or a searchable name autocomplete. Autocomplete filters `S._allUsers` by display name or email; rows show name + both emails for dual-platform users. Selected user's full name fills the input; both emails stored in the scope.
|
||
3. `GET /api/db/flagged` — filters `WHERE account_id IN (scope.user set)`, covering items from both platforms.
|
||
4. Viewer header — `#viewerIdentityBadge` shows `scope.display_name` (full name); `#filterRole` hidden.
|
||
5. `POST /api/viewer/tokens` — validates all entries in `scope.user` contain `@`; rejects combined `role`+`user` scope.
|
||
6. Token list — shows display name badge; falls back to emails joined with `, `.
|
||
|
||
**Size:** Small · **Priority:** Medium
|
||
|
||
---
|
||
|
||
### Scan history browser ✅
|
||
Review results from any past scan session without running a new scan.
|
||
|
||
**Implemented:**
|
||
1. `gdpr_db.py` — `get_sessions(limit=50, window_seconds=300)`: groups `scans` rows into 300 s windows (same logic as `get_session_items`), returns newest-first list with `ref_scan_id` (highest scan_id in group), timestamps, sources set, flagged count, total scanned, and a delta flag.
|
||
2. `gdpr_db.py` — `get_session_items(ref_scan_id=N)`: when `ref_scan_id` given, anchors the 300 s window to that scan's `started_at` instead of the latest scan.
|
||
3. `GET /api/db/sessions` (new endpoint in `routes/database.py`) — returns the sessions list; viewer-mode sessions share the same `GET /api/db/flagged?ref=N` endpoint with scope enforcement intact.
|
||
4. `static/js/history.js` (new module) — `loadHistorySession(refScanId)`, `openHistoryPicker()`, `closeHistoryPicker()`, `exitHistoryMode()`, `invalidateHistoryCache()` all exposed on `window.*`. Session cache (`_sessions`) invalidated by all `*_done` SSE handlers so the picker stays fresh after a new scan.
|
||
5. History banner (`#historyBanner`) — shows session date/time, sources, item count; "Sessions" button opens picker dropdown; "Latest scan" button appears only when not already viewing the latest.
|
||
6. Auto-load on page load — `results.js` calls `window.loadHistorySession?.(null)` when the SSE watchdog detects `!status.running`; `null` resolves to the latest completed session.
|
||
7. Live→history transition: clicking a session in the picker sets `S._historyRefScanId` and shows the banner. History→live transition: `startScan()` calls `window.exitHistoryMode?.()`.
|
||
|
||
---
|
||
|
||
### Gmail SMTP error message when App Password already in use ✅
|
||
The `535` auth error from Gmail fires for wrong app password, revoked app password, spaces in the 16-char code, and wrong username — all indistinguishable at the SMTP level. The old message unconditionally told users to "create an App Password", which is unhelpful when they already have one. Both the `smtp_test` and `send_report` error handlers now emit a Gmail-specific message that lists the three common causes and links to the App Password page for regeneration.
|
||
|
||
---
|
||
|
||
### #32 — Windowed mode for Profiles, Sources, and Settings ✗ Won't do
|
||
The workflow is sequential (configure → scan → review), not parallel — there is no realistic scenario where a modal and the results grid need to be open simultaneously. The Sources panel is already visible in the sidebar. Option A (the least-work path) still loads the full 3800-line JS stack twice. Closed.
|
||
|