Speed up CI and format and lint CLI commands

[tjementum]

Summary & Motivation

Three coordinated changes that cut CI critical path roughly in half and speed up local format and lint runs.

Test infrastructure

BackOfficeEndpointBaseTest and EndpointBaseTest<TContext> each constructed a fresh WebApplicationFactory<Program> per test instance. Combined with Meziantou.Xunit.ParallelTestFramework, every account-API test booted a complete ASP.NET Core host: full DI container, all singletons, test server. With 27 BackOffice classes and ~60 Account API classes, every CI run paid hundreds of cold-starts, and the cost grew with every test added.

Both test bases now share one host per test class via xUnit's IClassFixture<...>:

• BackOffice — new BackOfficeWebApplicationFactory + BackOfficeTestContext route per-test state (SqliteConnection, telemetry collector, MockStripeState) through an AsyncLocal. 26 derived test classes use the fixture via primary constructor + IClassFixture<BackOfficeWebApplicationFactory>. BackOfficeBlobProxyTests carries a derived factory to keep its IBlobStorageClient substitute.
• Account API — new AccountWebApplicationFactory + AccountTestContext mirror the same pattern (adds per-test IEmailClient routing). 60 derived test classes converted. CompleteEmailSignupTests carries a derived factory for its logger override.
• MockStripeState is re-registered as transient in the test host so per-request resolution reads the per-test instance from the accessor.
• TestServer.PreserveExecutionContext enabled so the AsyncLocal flows from the calling test into request handling.
• Env vars and the SPA-shell file copy moved into the fixture constructor (one-time setup).

Single Code Style workflow

A new code-style.yml replaces the per-SCS code-style work in account.yml, main.yml, app-gateway.yml. One workflow, four jobs:

• detect-scope — small first job that classifies the diff and outputs the backend scope and format mode for the downstream jobs (inline git diff, no third-party action).
• code-linting — backend inspectcode + frontend build + frontend lint (oxlint).
• code-formatting — backend cleanupcode + frontend format check (oxfmt).
• sonarcloud — runs once per push against application/PlatformPlatform.slnx (was previously running three times per push, once per SCS).

Backend lint and format are scoped to a single SCS via --self-contained-system <name> when the diff touches only that SCS; cross-cutting changes (shared-kernel, AppGateway, multi-SCS) fall back to the full solution.

Faster format and lint

• The developer CLI format command defaults to changed-only. Cleanupcode runs against .cs files diffed vs origin/main; explicit --all-files for full sweep. CI auto-flips to --all-files when application/dotnet-tools.json changes (JetBrains tool upgrades reset the format ruleset, so accumulated drift in untouched files needs a full pass).
• The developer CLI lint --changed-only flag is opt-in (default = full). Recommended for routine local runs; CI always lints the full solution because inspectcode has cross-file rules. The lint skill documents when to omit --changed-only (cross-cutting changes that affect untouched files).
• Removed the workaround that generated a temporary .slnf from each .slnx to feed cleanupcode — JetBrains 2026.1 accepts .slnx directly.

Database Plan reuses Build and Test artifacts

build-and-test uploads application/**/bin and application/**/obj as a <scs>-build artifact (only when staging is enabled). _migrate-database.yml downloads it into application/ and runs dotnet restore for the NuGet cache, skipping the redundant checkout + setup + dotnet build chain.

CI benchmark

Comparing real PR-sync workflow durations on a comparable sister branch (PR #888, n=8 per workflow) against this branch (n=3 per workflow, mean):

Workflow	BEFORE (PR #888 sync, median)	AFTER (this branch)	Δ
Account	9m00s	6m16s	-30%
Main	6m41s	5m47s	-13%
AppGateway	7m26s	1m07s	-85%
Code Style	-	4m40s	new — does not extend critical path
Critical path per push	9m00s	6m16s	-30%
Total CI minutes per push	23m07s	17m50s	-23%

The AppGateway delta is large because code-linting and code-formatting were removed from app-gateway.yml entirely — they live in the consolidated code-style.yml now. AppGateway PR runs are just build + test.

Note for downstream projects

Downstream projects with a substantial Main SCS will see a bigger lift than this benchmark suggests. The slowest single CI step used to be backend cleanupcode; it now only inspects changed .cs files. Evidence: Main and Account workflows now sit at roughly the same wall-clock (5m47s vs 6m16s) even though Account has materially more code than Main. The fixed-cost setup (npm + dotnet restore + build) dominates once the format pass is changed-only, so SCS size matters less than it used to.

Test-step benchmark (isolated)

To isolate the impact of the test-host refactor alone, I temporarily disabled the Code Style and Database Staging jobs and measured just Run Tests with SonarScanner Analysis (5 samples per side):

	Run 1	Run 2	Run 3	Run 4	Run 5	Mean	Stddev
AFTER (BackOffice + Account shared host)	2m18s	2m26s	2m21s	2m21s	2m21s	2m21s	2.9s
BEFORE (per-test host, baseline)	3m46s	3m51s	3m38s	3m41s	3m54s	3m46s	6.7s

Δ −37% on the test step alone, and stddev more than halves — the runs are markedly more stable too.

Checklist

• I have added tests, or done manual regression tests
• I have updated the documentation, if necessary

[linear]

PP-1209

[tjementum]

E2E test fix piggy-backed on this branch

Two e2e tests inherited from the just-merged feature-flags PR were failing on a default-seeded database:

• feature-flag-flows.spec.ts:262 — "Features" heading missing on /account/settings
• user-management-flows.spec.ts:473 — same root cause

Root cause: account-overview (TenantOwnerConfigurableFlag) and compact-view (UserConfigurableFlag) are both registered with isKillSwitchEnabled: true. The startup reconciler creates kill-switch flags with EnabledAt = null, so they're inactive on a fresh seed. The GetTenantConfigurableFeatureFlags / GetUserConfigurableFeatureFlags handlers filter inactive base rows out, the API returns an empty array, and FeaturesSection / PreferencesFeatureFlagsSection render null — heading and all. The original author's environment had these flags already activated.

Fix: Tests now activate the kill-switch flags via the back-office activate endpoint during setup. No production code touched, no seeder change, no conditional branches on flag state in the tests.

Result locally: 35/35 passed.

Strictly speaking this is a feature-flags PR follow-up, not a CI-perf change, but landing it on this branch unblocks CI green here.

[sonarqubecloud]

Quality Gate passed

Issues
13 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud