These tests queried an Activa account (checking) as accountId, but
the totals chart only ever runs for Expense-class accounts, which
are always the target side of a spend booking. With a realistic
expense-account setup, the existing TargetAccountId filter behaves
correctly: it excludes bookings where the account is the origin
(a negative value can't be shown as a pie slice) rather than netting
them against the target-side booking.
Bookings like "Narbonne Car" and "Narbonne Food" should roll up into
one series/slice instead of staying separate, so grouping now uses
the text up to the first space via TransactionDataset.GetTextGroup().
Adds two charts to the per-account transactions view: a stacked bar
chart of monthly totals and a pie chart of overall totals, both
grouped by booking text, so recurring bookings (rent, subscriptions,
etc.) are visible at a glance. Chart components are keyed on the
account id so switching accounts remounts them instead of reusing a
stale Plotly instance.
ChartService summed raw transaction Value instead of converting via
GetValue(account.IsFxAccount), so FX expense accounts charted in their
foreign currency labeled as CHF. Subclass series also took their Name
from an arbitrary account in the group instead of the actual subclass
name.
Every repository, service, Blazor page/dialog, and test now uses
async/await. Single atomic diff; the codebase does not compile in
intermediate states.
- BaseRepository: LoadListAsync/LoadAsync/SaveAsync return Task<T>;
per-file locks use SemaphoreSlim so waiters can await; Save
serialises to a MemoryStream sync (XmlSerializer has no async
form), then File.WriteAllBytesAsync + sync File.Move.
- RepositoryCache.GetOrLoadAsync takes a Func<Task<List<T>>>.
- All 7 repository interfaces + implementations async.
- All service interfaces + implementations async (except vendor
IFxService and stateless IFxConverter / SettingsService).
- Every Blazor OnInitializedAsync switches to await base.
- Test suite fully async, 42 tests pass.
AccountRepository.EnsureAccountsFile keeps two .GetAwaiter().GetResult()
bridges because it runs from the constructor.
Null-render guard follow-up (folded in):
Blazor now renders the component once with fields at their initial
values while OnInitializedAsync awaits — so fields declared `= null!`
are actually null on that first render and things like
`accounts.GroupBy(...)` throw ArgumentNullException. Fixed across
Transactions, BalanceReport, BalanceSheetReport, ProfitLossReport,
DetailReport, Assets, Spendings, SpendingsOverTime, TransactionDialog,
TransactionSplitDialog, and BookingRuleDialog:
- Collection fields initialise to [] so first-render loops are empty.
- Single-object data fields become nullable; the razor wraps
consumption in `@if (field is null) { <p>Lädt…</p> return; }`.
- <PlotlyChart> guarded behind a null check on config/layout/data so
Plotly.Blazor's @bind doesn't see nulls.
- Header/footer strings initialise to "" instead of null!.
Architectural hygiene on a single-user local Blazor Server app: the
observed win is one File.ReadAllBytesAsync and one
File.WriteAllBytesAsync per Load/Save, and after PR D each file is
loaded at most once per SignalR circuit.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
Add Schaad.Accounting.Tests (xUnit + NSubstitute + Shouldly, net9.0)
wired into Accounting.sln with a direct Reference to
Schaad.Finance.Api.dll so it flows into the test binary.
Naming convention: file/class is <Subject>TestShould, each test method
is DoThisWhenThat. Reads as a sentence:
"ViewService test should sum activa and passiva totals separately when
getting balance view". All assertions use Shouldly (.ShouldBe,
.ShouldBeNull, .ShouldContain, ...) rather than xUnit Assert.*.
38 tests across seven files:
- ViewServiceTestShould: balance math (activa/passiva totals, per-
account balance from start balance + credits - debits, FX conversion
to CHF) and bank-transaction auto-matching (booking-rule text,
value-matching preference, same-accounts-last-month fallback,
open-transaction filter).
- TransactionRepositoryTestShould: FX round-trip against a temp XML
directory, unknown-id -> null, and the mutation-on-read regression
from PR A (a second Get on the same FX transaction used to divide
by FxRate again).
- FormattingTestShould: Swiss thousands separator, two-decimal
rounding, culture independence.
- RepositoryCacheTestShould: loader called once, per-key isolation,
invalidation forces reload, case-insensitive keys.
- AccountRepositoryTestShould: id assignment, in-place update,
currency defaulting, delete, bank-account lookup, bank-balance
update, and constructor re-run against an existing file.
- ChartServiceTestShould: honours settingsService.GetYear() without
mutating it (locks in PR E), account vs sub-class grouping
heuristic, skips empty accounts.
- FileServiceTestShould: CSV header + running balance for debit and
credit lines, ordering by BookingDate/ValueDate/Value, no in-place
value mutation (locks in PR A).
CLAUDE.md gains a `dotnet test` line. IMPROVEMENT_PLAN.md notes that
this landed as three PRs (G, H, I) and was squashed on request.
Also add `*.DotSettings.user` to .gitignore so Rider's per-user
solution settings don't get accidentally staged.
Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>
