Second half of the envelope fix: ITrackPayload is {name, properties, profileId} — sending
'event' is rejected with 400 "payload/name Invalid input: expected string", which the
module swallows by design. Verified against packages/validation/src/track.validation.ts
and the SDK's track(name, properties) signature.
Two blockers found while wiring OpenPanel for real:
- frontend/Dockerfile declared ARG for VITE_OPENPANEL_URL/CLIENT_ID/SENTRY_DSN but not
VITE_OPENPANEL_SCRIPT_URL, so passing it as a build arg was silently ignored and the
loader compiled out (import.meta.env value undefined).
- backend/app/services/analytics.py posted a flat {type, event, properties} body; the
Track API expects {type, payload:{event, properties, profileId}} (verified against the
SDK source in packages/sdks/sdk). The API answered 400 and the module swallows every
error by design, so all server-side events were being dropped without a trace.
Verified after deploy: the bundle carries the client id + op1.js URL + the queueing shim,
op1.js serves 200, and a track event posted with the real envelope lands in OpenPanel.
analytics.ts called window.op?.(…) but nothing ever defined that global — no script tag
in index.html and @openpanel/web was never imported — so every call was a silent no-op
and product analytics would have stayed empty no matter how the build args were set.
Now, when VITE_OPENPANEL_SCRIPT_URL is configured, the loader installs OpenPanel's
documented queueing shim (window.op becomes a queue) and appends op1.js async; the
existing op("init"/"identify"/"track") calls are queued and drained when it loads, so
ordering cannot break them. Without the URL the behaviour is unchanged (no-op), which
keeps local dev and air-gapped builds silent.
Build args: VITE_OPENPANEL_URL (API base, e.g. https://stats.<host>/api),
VITE_OPENPANEL_CLIENT_ID, VITE_OPENPANEL_SCRIPT_URL (e.g. https://stats.<host>/op1.js).
The client *secret* is deliberately not a frontend arg — it belongs to the API only.
Categories are kept only as the label each supplier carries (suppliers.category_id
+ the material_categories list it points at). Everything that treated category as
a shared key across materials, inventory, quotes and unit conversions is removed —
including two features that were never populated (0 rows on every volume).
Removed (backend):
- routers category_attributes, category_ref_units, unit_conversions + registration
- ?category_id= filters on /materials, /inventory and /quotes (the /suppliers one
stays — it filters suppliers by their own label)
- GET /inventory/materials/options: required a category and had no caller
- the material-payload category derivation (expand_service.category_by_material,
the inventory serializer lookup) and the material category on /pm/on-the-way;
the wire keys stay, now always null, so the payload shape is unchanged
- startup DDL: DROP the empty category_attributes / category_reference_units /
unit_conversions tables (idempotent) and their CREATE statements; seed.py
truncate list updated
Removed (frontend):
- the category-keyed conversion path: refUnitForCategory / findConversion /
refConverted, the categoryUnits + unitConversions store slices, api clients,
hooks and the "≈ converted quantity" line on quote detail and project pages
(it needed rows in those tables to render, and there were none)
- material.supplier_category and its read sites (the edit modal grew one
yesterday to restore labels; the label lives on the supplier now)
- the inventory page's category chips, the row filter and the «دسته» column,
plus the lookup's categories fetch and catName helper
Kept: supplier label + its list and picker, /suppliers?category_id=, the global
standard_conversions feature and the units page (material_units conversions).
Verified against a restore of the production dump: startup DDL drops exactly the
three tables, suppliers keep their label (3/3), material_categories still serves
its 8 rows, the removed endpoints 404, and materials/inventory/quotes/units/
standard-conversions all return 200 with quote line items intact.
The category filter is not used: remove it end to end — the toolbar's category
combobox (desktop + mobile drawer), its option list, the active-filter count and
the filterQuotes() clause it drove, plus the now-unused `category` field in the
Filters type. Supplier and status filters are unchanged.
The backend's own quote filtering by category is untouched (it filters on the
quote's supplier category when ?category_id= is passed), so the API keeps its
contract; only the dashboard UI stops offering the filter.
v2.7 dropped materials.category_id and the API started returning category: null for
every material, but the frontend still read material.category as the material's label.
Effect: the edit page's material field rendered blank for every row, the dashboard
category filter listed nothing (and matched nothing), the wizard's category suggestions
were empty, and the conversion lookups keyed by category (refConverted →
categoryUnits/unitConversions) silently returned null, dropping the per-reference-unit
price. Saving an edited quote matched materials on an always-true category term, so two
materials sharing brand+specs were interchangeable.
Auditing the dropped column also turned up four places still dereferencing it, one
request away from a 500 and hidden only because the inventory is empty:
GET /inventory (serializer), GET /inventory?category_id=…, GET /inventory/materials/options
and DELETE /material-categories/{id}.
Backend:
- expand_service: category_by_material() resolves a material's display category from its
FIRST linked supplier (oldest supplier_materials row — the v2.5 trigger's rule), used by
the material and line-item payloads; category_id is populated again when not expanded.
- inventory: list filter, row serializer and material-options filter match through
supplier_materials ⋈ suppliers, mirroring GET /materials?category_id=…
- material_categories: the delete guard blocks on suppliers (the RESTRICT FK) and says so.
Frontend:
- Material.category → Material.supplier_category (supplier-derived, documented); the
rename pushed every read site through tsc.
- Labels use the material's own name; the edit modal field is «متریال» and prefills by
name; formatMaterialTitle no longer falls back to a category.
- Identity matching (edit modal, wizard, material memory) is name + brand + specs via a
shared matchMaterial(), so same-brand same-spec materials are not conflated; legacy
composed names are still found through materialBaseName().
- Filters read the supplier's category (dashboard options + quote filter, consistent with
the backend's quote filtering); supplier categories now come from the supplier FK
instead of the removed JSONB column. Conversion keys use supplier_category.
- Removed dead code carrying the old logic: two uncalled topicCategories copies and
materialOptions' unused "cat" branch.
Verified: both images build clean (tsc -b && vite build); the API was exercised in-process
against a restore of the current production dump with an inventory row added — material
and line-item payloads carry the supplier category, the inventory endpoints that used to
raise return 200, and the category delete guard returns 409 instead of 500.
Supplier topic/categories were derived from quote line items' material
category and topic. The catalog v2.7 refactor decoupled materials from
categories, so that derivation now yields nothing and every supplier lost
its topic/category — the suppliers dashboard rendered without names,
because it groups topic → category → supplier.
The backend already returns the supplier's own category
(category_id FK + nested category{id,slug,name}); adaptSupplier simply
never read it. Now it is the primary source, with quote-derived values and
the local registry still taking precedence when present.
Verified against the live stack: all 7 suppliers return their category
from /suppliers, and the suppliers page renders names again.
Downloads require an Authorization header, which <iframe>/<img> cannot
send. The previous blob: workaround broke under download managers (IDM
hooks blob URLs, fails to resolve them, and leaves a 0-byte file) and in
Chrome's mobile emulation, which has no inline PDF plugin.
Backend:
- preview_token.py: HMAC-signed token bound to one (quote, kind, file)
with a 5-minute TTL, keyed off an existing secret
- GET /quotes/{id}/preview-url/{kind}/{attachment} mints the URL
- the two file GET routes accept a Bearer session OR the signed ?t= token
(they leave the global auth dep, so the other six file routes now carry
it explicitly — nothing is open by omission)
Frontend:
- api.files.previewUrl() replaces the blob helper
- shared FilePreviewModal + useFilePreview; FileList and the dashboard
quote card both use it
- PDFs render with the browser's native viewer on desktop and pdf.js on
canvas where no inline viewer exists (mobile); pdf.js is fed the bytes
in memory so no PDF URL is exposed for a download manager to intercept
- removed the per-attachment download buttons on the dashboard card and
the modal header
Ops:
- nginx: explicit MIME types (.mjs as text/javascript) — a module worker
served as octet-stream is refused and crashed the page
- frontend service bypasses the nginx entrypoint, whose apk manifest call
hangs when the package mirror is unreachable (502s)
prod_v2's inline preview modal and EditModal redesign conflicted with the
hardening branch's authenticated file API (urlOf/downloadUrl were removed
because <a href>/<img> cannot send the Authorization header).
- FileList: kept prod_v2's design (icons, preview modal, per-row actions)
and adapted it — previews now fetch the bytes with a Bearer token and
use a local objectURL (api.files.blobUrl), with revoke on close/unmount;
the row download uses the authenticated onDownload.
- EditModal: took prod_v2's tabbed redesign wholesale (it already carried
our globalSpecs, clean material name and target_project work) and
converted its FileList to onDownload + loadPreview.
- AcceptModal: converted the remaining urlOf sites and removed duplicated
file/date blocks the earlier merge had left outside the scroll body.
- api.ts: restored the global handle401 (definition + all four fetch
sites) alongside the new blobUrl helper.
Conflict in init_db.py resolved keeping the audit_log table ensure
function (new-one) — prod_v5 predates it and its side was empty.
Both sides' SQL lands: audit_log table plus the category-free materials
model.
Conflicts resolved keeping the hardened/current side:
- docker-compose.yml: kept new-one's db-backup service, hardened env
(ENVIRONMENT/DEV_LOGIN/CORS/issuer var) and build args; prod_v2's
8083/8085 dev port remaps dropped as they conflict with the hardened
Casdoor issuer expectation.
- AcceptModal.tsx: kept the per-item target_project model (43ff048) over
prod_v2's older projUnion candidate-project logic, kept the
authenticated file upload/download blocks, and adopted prod_v2's
sticky footer.
Bug: a multi-project quote accepted for «تست ۱» only also appeared under «تست ۲»'s پذیرفته شده column, because both the Projects page and /projects/{id}/stats classified quotes by their GLOBAL status over any project link.
Rule now: a quote counts as accepted/arrived for a project ONLY if at least one non-rejected line item's decision targets that project (item_decisions.project_id). A quote merely linked as a candidate — including one accepted for a different project — is categorized exclusively under پیشنهاد شده for that project. Legacy quotes without any decision targets keep the old global-status classification.
Backend: project_stats membership now spans all link kinds (legacy mirror, quote_projects, line_item_projects, decision targets) with the per-project decision-based bucketing above; bucket VALUES follow the same isolation (accepted/arrived value counts only items targeting the project; candidate items of a still-proposal quote stay in the offered bucket); the raw path string is normalized to UUID for the Python-side comparisons. /quotes?project_id=… now matches any relevant link kind instead of only the legacy mirror column.
Frontend: ProjectsPage gains acceptedForProject/hasTargets helpers; the three columns (پیشنهاد شده / پذیرفته شده / رسیده) and their value chips use per-project items (target-project items first, then candidates), so totals no longer leak other projects' items.
Verified end-to-end (2 projects × 1 quote × 2 items, accept item-1 for P1 only): P1 accepted=1 then arrived=1 after the PM receipt; P2 stays offered=1 / accepted=0 / arrived=0 at every step; /quotes?project_id=P2 still lists the quote. All fixture data removed; 7/7 checks pass. python compile + api/frontend builds clean.
Warehouses now belong strictly and exclusively to the project they were created under; cross-project reassignment is removed end-to-end.
Backend: PATCH /inventory/warehouses/{id} no longer applies project_id — any attempt to move a warehouse to a different project returns 400 with «پروژهی انبار قابل تغییر نیست — انبار برای همیشه به پروژهی والدش تعلق دارد» (sending the current project stays a harmless no-op). The WarehouseUpdate schema documents project_id as accept-only-to-reject; creation still requires project_id and delete is unchanged.
Frontend: the Organization page's per-warehouse project dropdown («انتقال به پروژه دیگر») and its moveWarehouse handler are removed — warehouses render statically inside their project card with only the dashboard link and delete; the warehouses.update API type no longer accepts project_id. WarehousesPage already had no move UI (rename/describe only).
Verified end-to-end against the running stack: rename 200; same-project project_id no-op 200; cross-project reassignment 400 with the guard message; warehouse remains in its original project; test data cleaned up. python compile + api/frontend builds clean.
1) Quote card header: an official/unofficial badge (رسمی / غیررسمی) renders next to the supplier name from quotes.is_official (the «فاکتور رسمی» checkbox) — green for official, muted slate for unofficial.
2) Conversion Factors sidebar: the material group and variant labels no longer truncate with ellipsis — whitespace-normal + break-words so long Persian names wrap cleanly to the next line.
3) Global fixed conversions are now seeded in the unified seeder (app/db/init_db.py — the same file that seeds categories, base units and the superuser path), NOT hardcoded in the frontend: تن→کیلوگرم (1000), کیلوگرم→گرم (1000), هزار عدد→عدد (1000). The supporting units (گرم, هزار عدد) were added to the idempotent BASE_UNITS seed; inserts use ON CONFLICT (from_unit, to_unit) DO NOTHING against the uq_standard_conversion key so every startup is safe. Verified live: 3 rows after first boot AND still 3 rows after an api restart.
4) Universal proforma attachments: the Edit modal now has a پیشفاکتورها section (list/upload/delete via the existing MinIO-backed files API) alongside the invoices section, active at every quote lifecycle stage — including accepted/arrived and receipts-locked quotes (the section is deliberately not gated by hasReceipts; the backend already permits uploads in any status). View/delete of existing proformas unchanged elsewhere.
5) EditModal fixes: (a) the supplier Combobox no longer opens on mount — Radix Dialog autofocus was triggering its onFocus; Combobox gains an openOnFocus prop (default true) and the modal passes false, so the dropdown now opens only on explicit click or typing; (b) the line-item material field binds to the material's own name (m.name) instead of its category relation, and the save path matches materials by name+brand+specs accordingly.
Validated: python compile, api + frontend image builds (tsc + vite) clean; seeding verified against the live local DB.
Materials are now global and category-free. The category lives only on the supplier (suppliers.category_id); every category-facing query goes through the supplier edge instead of the material row.
Schema (migration.sql + init_db DDL):
- drop the materials.category_id column, its FK, and the v2.5 first-link-wins automation (fn_derive_material_category + trg_supplier_materials_derive_category) — the DDL runs idempotently at API startup
Backend:
- GET /materials?category_id=… now filters via EXISTS(supplier_materials ⋈ suppliers.category_id)
- quote category filters (quotes list) use the quote's supplier category
- /pm/on-the-way derives the display category from the supplier of the quote
- supplier_materials links no longer derive/return a material category
- expand_service: material expand=category resolves to null (kept for contract compatibility)
- material create/update no longer validates or stores category_id; seed.py aligned
Frontend:
- material creation no longer sends category_id
- formatMaterialTitle reworked to a strict order with « . » separator: [Brand] . [Material] . [Spec 1] . [Spec 2], brand deduped when the stored name already starts with it; specs separator switched from « · » to « . »
Validated: python compile, api + frontend image builds (tsc + vite) clean; branch cut from the tip of prod_v4 (= origin/production, 5a9ef40 — the team's merge made them identical).
RUNTIME BUG (caught by live smoke tests): the global auth dependency
queried the request-scoped session, auto-beginning its transaction — any
endpoint opening 'async with db.begin()' afterwards crashed with
'transaction already begun' (POST /materials 500). get_current_user /
get_actor now resolve in a short-lived session (expire_on_commit=False
keeps the context usable detached); inventory/pm scopes keep passing the
request session explicitly.
Smoke suites rewritten against CURRENT contracts:
- assertions are behavior-based (shapes, envelopes, filters), no magic
seed counts; dynamic lookups instead of hard-coded names/ids
- rerun-tolerant: tolerate 409-on-create by reusing prior artifacts
- aligned with current rules: revert blocked after inventory receipts
(v2.6), attachments allowed in every status, termsheets optional at
quote level
- SMOKE_USER/SMOKE_PASSWORD env for login (default tester/123456)
Verified on a fresh volume: migration + audit_log bootstrap, Casdoor
seed (bcrypt), demo data, and 29+76+61 smoke checks all green.
- slowapi limiter on POST /auth/login (10/min per IP, Persian error msg);
edge-level coarse limits stay in Traefik
- mem_limit on every main-compose service (~4.4GB worst case, protects the
8GB shared host during rebuilds)
- monitoring images pinned to verified tags (prom v3.4.1, grafana 11.6.3,
loki/promtail 3.5.0, node-exporter v1.9.1, postgres-exporter v0.17.1,
cadvisor v0.49.1); exporter image name fixed to postgres-exporter
- db-backup sidecar: pg_dump -Fc of procurement+casdoor every
BACKUP_INTERVAL_SECONDS into BACKUP_DIR, retention via BACKUP_KEEP_DAYS
- audit_log table (migration.sql + runtime ensure_audit_table) with
model + transactional helper (utils/audit.record) — wired into user
create/role-change and inventory transaction creation
- get_actor: quote_events.actor now stores the VERIFIED user's display
name instead of the raw bearer token
A material that ends up referenced by nothing (no line_items anywhere, no warehouse stock) was previously left behind forever, accumulating dead catalog rows as offers were edited or deleted.
- new quote_service.cleanup_orphan_materials(db, material_ids): for each candidate it checks usage across line_items (all quotes) and warehouse inventory; only when both are zero does it remove the auxiliary rows (supplier_materials, brand_materials, material_units, material_attribute_values) and finally the materials row; returns the purged ids
- replace_quote_children (edit modal): snapshots the quote's material ids before the children are replaced, diffs against the incoming items, and purges whatever was detached
- delete_quote: collects the quote's material ids + attachment keys first, deletes the quote, purges orphans inside the transaction, and only then (post-commit) removes the MinIO objects — now both the recorded attachment keys and a recursive remove_prefix(quote_id/) sweep, so stray objects under the quote folder cannot survive; the response's cascaded block reports purged_orphan_materials
- line_items: PATCH that switches a line item's material, and DELETE of a line item, both clean up the previously referenced material
- storage.remove_prefix(prefix): best-effort recursive object deletion under a prefix (sync MinIO client wrapped in asyncio.to_thread, same pattern as remove_object)
Safety: usage checks run against the whole system before any delete, so shared materials — including those held only by warehouse stock — are never touched. Verified: python compile + api image build clean.
Wizard:
- failed saves now route to the exact step+field: a new data-field attribute on every required section (supplier_id, line_items, termsheets, project, quote_date, expiry_date, notes, proforma) plus resolveErrorTarget() that maps ApiError.field / message keywords (supplier, line item/material/unit/quantity, termsheet/schedule/milestone, project, dates, notes, file) to the offending step — the «رفع خطا» toast action scrolls/focuses the right control instead of guessing
- material records are no longer created on the fly while typing (rememberMaterial no longer fires addMaterial per blur); materials/specs/units are written once during the final save, avoiding half-created catalog rows
Accept flow:
- quote_service no longer rewrites LineItemProjects on accept — those candidate-project rows stay 100% untouched and only item_decisions.project_id is written, matching the per-item target-project model and the UI payload (item_projects: [])
Dates/time:
- JalaliDatePicker: hour/minute are now editable text fields (Persian digits, clamping, 3-digit retype handling) with a controlled useEffect that only rewrites the inputs when the numeric value actually changes — no more wiping in-progress typing; date-only detection tightened to a full YYYY-MM-DD match so datetime strings are not misread
Also: /api validation logging (RequestValidationError) now includes the path and error list for faster diagnosis.
Verified: api + frontend image builds (python, tsc, vite) clean on top of origin/production (incl. the team's mobile layout merges).
The specs combobox drew only from materials in the selected topic, further
narrowed by the row's category and brand, so a spec was offered only when
some material in that topic shared the same category AND brand. On a new
quote the user therefore lost specs recorded in earlier quotes.
- api.specsList(): new global source reading the attribute-definitions
registry (topic/category/brand independent)
- materialOptions('spec'): unions the registry, specs of all materials,
and specs typed in sibling rows — no topic/category/brand filter
- EditModal: same global registry added to its spec options
Verified the registry endpoint returns labels and that the union includes
specs absent from material attributes (e.g. 'مسی').
Adding a new item wrote the material name as 'brand + category + specs'
(e.g. 'پارس لوله لوله PE100 سایز ۱۶'), while specs were also stored
correctly in EAV attributes. The wizard then tried to recover the base
name by stripping brand and specs out of that string, but never stripped
the injected category — so the spec stayed baked into the cached material
name and the مشخصه appeared as part of the material name on later visits.
- api.ts / WizardPage / EditModal: build name from the material itself
only (brand goes to brand_ids, specs to attributes)
- materialBaseName: strip category too, never strip a part that would
empty the name, and reject a spec-only result; legacy rows whose name
cannot be disentangled fall back to the stored name rather than a guess
- WizardPage: match existing materials on category+brand so legacy rows
are still reused instead of duplicated
Brings in prod_v3 5084623 on top of production 9e8a236. Merged without
conflicts: prod_v3's base (7ead60c) is an ancestor of production, and
production's extra tab-bar refinements (17c7ae6) touch disjoint regions.
Quotes that already have warehouse receipts can no longer be edited or rolled back:
- backend: quote_has_receipts() guard — replace_quote_children and every backward status transition now return 409 with «این پیشنهاد بهدلیل ثبت رسید در انبار، غیرقابل ویرایش یا بازگشت است.»
- reopening to «offered» additionally clears item_decisions.project_id (the purchased-project link belongs to acceptance, not to a reopened offer)
- frontend EditModal: hasReceipts locks the whole form — disabled inputs and date pickers (JalaliDatePicker/NumberInput/MilestoneRow/MultiSelectCombobox gained disabled support), add/remove controls for items, termsheets and milestones hidden, status change blocked, amber warning banner; quote deletion now goes through a confirmation dialog (useDeleteQuote)
- date-only semantics: JalaliDatePicker keeps/emits YYYY-MM-DD when withTime=false (typing, «امروز», day selection) and toDateOnly no longer mangles pure date strings
Per-item target project (candidate projects stay intact):
- expand_service exposes the decision target project (decision.project + line_item.target_project)
- the accept payload derives item_decisions.project_id from that target project and sends item_projects: [] so line_item_projects keeps the original candidates untouched
- AcceptModal/EditModal select the target project per line item
Material titles:
- quoteUtils.formatMaterialTitle() composes one normalized, deduplicated «name + brand + specs» title, replacing materialLabel/materialSpecs/specsLabel usage across dashboard cards, quote detail and the modals
Verified: python compile + api/frontend image builds (tsc + vite) clean.
Brings in main_v2 17c7ae6 (tab bar sizing, inventory actions, date picker
proportions, stats lead time).
Conflicts resolved in favour of main_v2, which deliberately removed the
'تغییر موجودی' top button (nav lives in the tab bar / InventoryTopNav now)
and rewrote the proforma date field. Dropped the now-unused Button import.
Merges main_v2 a110741 (mobile UI optimizations) into production and
reinstates the mobile bottom navigation that f0e4243 had removed.
Conflicts in DashboardPage, QuoteCard and InventoryPage resolved in
favour of main_v2's mobile-first markup, keeping production's newer
design system (Button variant="primary" on InventoryPage).
f0e4243's removal was reverted in effect: MobileTabBar is restored and
remounted in App.tsx, and PageLayout regains max-md:pb-24 clearance so
content is not hidden behind the fixed bar.
Task 1 — PM 1:N + access-loss handling:
- projects.pm_user_id had no UNIQUE constraint (verified live + in schema); added a lookup index via init_db
- auth: UserContext.project_ids = every project of a PM (oldest first); project_id stays the first for compatibility
- GET /users/me/projects — assigned projects of the caller (drives the switcher + revocation validation)
- login / /users/me / /users list now return the full projects array per user
- /inventory/* and /pm/* scopes accept an explicit ?project_id=… and verify the PM manages it (403 otherwise); without the param they fall back to the first project; admin/manager use it as an optional filter
- POST /users/{id}/project is multi-aware: assigns one project (replacing that project's previous PM) and KEEPS the user's other assignments; null clears all
- PATCH /projects/{id} accepts pm_user_id (admin / procurement manager only)
- frontend: auth store holds projects + the active project (persisted via the existing localStorage store); RequirePmProjects re-validates on load, falls back when the active project was revoked, and shows a dedicated no-projects screen with logout (no redirect loop, no re-auth)
- Panels menu: «مدیریت انبار» gets a ChevronDown accordion listing the PM's projects with «✓ فعلی»; selecting one switches only the active project and reloads the section data
- inventory/pm API calls now send the active project_id explicitly
Task 2 — Organization page:
- new searchable ProjectMultiSelect combobox (chips with ×, live Persian filtering via normalize, assigned projects hidden) replaces the single select for PM users
- add = PATCH project pm_user_id=<user>; × = pm_user_id=null; table and project cards sync immediately; RBAC unchanged (role column admin-only, combobox editable for admin + procurement manager)
Verified locally: 17 scripted checks — multi-project assignment without unique conflicts, /users/me/projects, explicit project_id authorization (foreign project 403 on /inventory, /pm/on-the-way, /pm/receipts), PM blocked from pm_user_id changes, legacy assign endpoint keeps other assignments, revocation removes the project from the list, zero-project PM gets an empty list and a guarded inventory endpoint. All test data cleaned up afterwards.
Both download endpoints (invoice/proforma) put the stored file_name raw into the Content-Disposition header. HTTP headers are latin-1, so any non-ASCII name (i.e. every Persian filename) raised UnicodeEncodeError before the body was sent -> HTTP 500. Uploads were unaffected.
Content-Disposition is now built by a helper following RFC 6266 / RFC 5987: the UTF-8 name is percent-encoded in filename*=UTF-8''..., with a header-safe ASCII fallback in filename= (quotes and control characters stripped; a name that folds away entirely becomes attachment<ext>).
Verified locally after reproducing the exact production error (same latin-1 positions 22-27): Persian proforma + invoice downloads 500 -> 200 with filename*= carrying the full Persian name and bytes sha256-matching the MinIO object; ASCII filename regression 200; helper edge cases (CRLF/quote injection stripped, empty name -> attachment, pure-Persian png -> attachment.png); no UnicodeEncodeError in the api logs after the fix.
whqreza
ZCode