Django's ' escaping is decoded by the browser before hyperscript parses
a _="..." attribute, so a quote in a currency or account name closed the
string and the rest ran as hyperscript. Names now live in plain data-*
attributes and are read with @data-*. Spaces in currency prefixes and
suffixes are kept.
Refs GHSA-fh52-f8r4-hv6g
Currency, account and category names were rendered into inline <script>
blocks with |safe, so a name containing </script> or a quote ran as script
for every user whose net worth or insights page included it.
Fixes GHSA-fh52-f8r4-hv6g
Add /api/splits/ endpoints for events, participants, expenses and payments,
plus an event summary (balances and settle-up) and leave action. They share
the web UI's access rules and share calculation. MCP reaches them through
search_api/call_api, and its instructions now point to them.
Track expenses shared with other people (trips, flats, dinners) and settle
up with the fewest payments.
- Events with WYGIWYH users and people without an account, who join through
a share link and pick a password on first sign-in
- Equal, exact, percentage and share-weight splits that always add up
- Balances, a minimal settle-up plan and an activity feed showing what each
item means for you
- Link items to your own transactions; recording a payment marks planned
ones as paid
- Share menu: printable PDF report and PNG images of the summary, settle-up
and balances (adds modern-screenshot)
- Only the two people involved or the organizer can record a payment;
members can leave events they don't organize
Saved filter presets have to be picked from the menu every time a page
opens, so a view someone always wants, like hiding projected
transactions (#521), takes a few clicks on every visit or a bookmarked
URL.
A preset can now be set as the default from the presets menu, with the
star next to its name. All transactions and the Monthly overview then
open with it applied, relative dates moved to today, whenever their
URL has no filters of its own:
- `FilterPreset.is_default`, with a constraint allowing one default per
user. Starring another preset moves the default to it, and starring
the default again turns it off.
- Filters in the URL, like `/transactions/?is_paid=1`, take precedence
over the default, and Clear still empties the filter for the visit.
- The star stays visible on the default preset and shows on hover for
the others, like the delete button.
- Building a filter from a preset moves into
`TransactionsFilter.from_preset`, shared by applying a preset and
opening a page with the default one.
The export and restore forms now sit on the Export and Restore page
instead of an offcanvas. Exporting shows a progress line and a toast
when the download starts; restoring splits ZIP and CSV uploads into
tabs and shows errors in the form.
- An empty export warns instead of downloading the message as a file
- Export errors show a message instead of a stuck spinner
- The download link is kept long enough for Safari to save the file
- Non-superusers get a 403 instead of a redirect loop with the login
- The asset classes CSV can be uploaded on its own
- Drop the export form's unknown investment_classes layout field
Restore looked categories, tags and entities up through the admin's
filtered manager, so other users' transactions moved to the admin's
categories and lost their tags. They are now matched to the row's
owner (then to ones that owner can see) and created for that owner.
Export also read tags and entities through that manager and left other
users' private ones out.
Recurring transactions and installment plans now export account_owner,
which restore already read but never received.
Transaction exports ran a few queries per row and paged prefetched rows
with OFFSET, so large histories timed out. They now load relations up
front and read in cursor chunks: 200k transactions export in ~43s with
~200 queries (10k took 34s and ~21k queries before).
Fixes#691Fixes#692
Override django-oauth-toolkit's unstyled authorize template with the login
page's layout. It says who's signed in, what the app will be able to do, and
where the answer is sent, so a look-alike app with someone else's redirect
is easier to spot.
Editing a recurring transaction to end earlier kept its unpaid occurrences
after the new end date, counting them in projected balances, and left
last_generated_date past it. Remove them (both sides of a generated transfer;
a partner linked from elsewhere is unlinked and kept) and move the bookmark
back to the last remaining occurrence. Paid occurrences stay.
Also pin ENABLE_SOFT_DELETE=False on the test that asserts hard deletes, so a
.env that turns soft delete on no longer fails it.