Elgato_dark/orca
1
0
Fork 0
mirror of https://github.com/stablyai/orca synced 2026-04-21 14:17:16 +00:00
Code Issues Projects Releases Packages Wiki Activity Actions 1

test(e2e): drop filter-narrows spec, keep only the AI layout check

Browse source 
The filter-narrows test was pure Playwright after the DOM-read refactor
and its regression coverage overlaps the default store-driven suite.
Remove it so the electron-stably project only runs specs where the AI
genuinely earns its latency cost — the populated-panel layout screenshot.
This commit is contained in:
brennanb2025 2026-04-20 13:35:02 -07:00
parent aceb8c0771
commit 14d4499df4
1 changed files with 15 additions and 115 deletions
Show stats Download patch file Download diff file Expand all files Collapse all files

130
tests/e2e/stably/source-control.spec.ts
View file

@ -2,22 +2,16 @@
* Stably-AI E2E tests for the Source Control panel.
*
* Why this suite lives under `./stably` and not alongside the default specs:
* - These tests exercise the *real* DOM path: opening the right sidebar,
* clicking "Stage" / "Unstage" / "Discard" on a row, and reading the
* panel back through AI extraction. The default suite drives
* `window.__store` directly (see tests/e2e/helpers/orca-app.ts), which
* is fast and deterministic but bypasses the React event handlers that
* most regressions land in.
* - `agent.act`, `toMatchScreenshotPrompt`, and `page.extract` hit
* Stably's hosted API, so this project only runs when STABLY_API_KEY is
* set (see tests/playwright.config.ts). Contributors without a key — and
* fork PRs whose secrets are not mounted — run the default suite only.
*
* Covers behavior that is currently only unit-tested:
* - SourceControl.tsx row actions (Stage / Unstage / Discard)
* - useGitStatusPolling.ts → gitStatusByWorktree hydration on tab focus
* - Panel-level visual: section headers, entry counts, and row icons
* render together without layout drift
* - `toMatchScreenshotPrompt` hits Stably's hosted API, so this project
* only runs when STABLY_API_KEY is set (see tests/playwright.config.ts).
* Contributors without a key — and fork PRs whose secrets are not
* mounted — run the default suite only.
* - The default suite drives `window.__store` directly (see
* tests/e2e/helpers/orca-app.ts); this suite renders the populated
* panel and asserts its structural layout through an intent-based
* visual, which catches regressions (section disappearing, filter
* input vanishing, row layout collapsing) that a store-level check
* cannot see.
*/
import { existsSync, writeFileSync } from 'fs'
@ -96,11 +90,11 @@ test.describe('Source Control (AI-driven)', () => {
await waitForActiveWorktree(orcaPage)
await ensureTerminalVisible(orcaPage)
// Why: open the right sidebar on the Source Control tab *before* running
// the agent so its first pass sees the panel. Also primes
// useGitStatusPolling, which only tops up gitStatusByWorktree when the
// window is focused — the headful Electron run needed here satisfies
// that precondition.
// Why: open the right sidebar on the Source Control tab *before* the
// screenshot assertion so the panel is visible in the captured frame.
// Also primes useGitStatusPolling, which only tops up
// gitStatusByWorktree when the window is focused — the headful
// Electron run needed here satisfies that precondition.
await orcaPage.evaluate(() => {
const store = window.__store
if (!store) {
@ -112,100 +106,6 @@ test.describe('Source Control (AI-driven)', () => {
})
})
test('typing in the filter input narrows the change list', async ({ orcaPage, testRepoPath }) => {
const { untrackedPath } = seedUncommittedChanges(testRepoPath)
try {
// Nudge the renderer's git status poll so the new files show up without
// waiting up to 3s for the natural interval. This is the same IPC the
// sidebar calls — we're just firing it eagerly on test startup.
await orcaPage.evaluate(async () => {
const store = window.__store
if (!store) {
return
}
const state = store.getState()
const worktreeId = state.activeWorktreeId
if (!worktreeId) {
return
}
const worktree = Object.values(state.worktreesByRepo)
.flat()
.find((entry) => entry.id === worktreeId)
if (!worktree) {
return
}
const status = await window.api.git.status({ worktreePath: worktree.path })
state.setGitStatus(worktreeId, status as Parameters<typeof state.setGitStatus>[1])
})
await expect
.poll(
async () =>
orcaPage.evaluate(() => {
const state = window.__store?.getState()
if (!state?.activeWorktreeId) {
return 0
}
return (state.gitStatusByWorktree[state.activeWorktreeId] ?? []).length
}),
{ timeout: 10_000, message: 'git status never hydrated for the active worktree' }
)
.toBeGreaterThanOrEqual(2)
// Why this test uses Playwright `.fill()` rather than `agent.act()`:
// earlier drafts asked the agent to find and click the filter input,
// but SourceControl.tsx exposes two "Filter files…" inputs — one
// persistent at the top and a second hover-revealed one inside each
// section. The agent consistently chose the wrong target (or worse,
// the terminal), so we use a deterministic locator. The AI value in
// this test is the assertion side — `page.extract()` reading the
// rendered section counts — not the typing mechanic.
const filterInput = orcaPage
.locator('input[placeholder="Filter files…"]')
.describe('Top-level Source Control filter files input')
.first()
await filterInput.fill('notes-')
await expect(filterInput).toHaveValue('notes-')
// Cross-check the visible effect: the store-level data did not change
// (both files still exist in gitStatusByWorktree) but the rendered
// counts next to section headers reflect only matching entries.
//
// Why a DOM read instead of page.extract: the count is plain text
// rendered in a sibling span (SourceControl.tsx → SectionHeader), so
// Playwright can read it directly without the ~5s latency and API
// cost of an AI extraction — and the result is the exact integer
// rather than an AI-parsed guess. The regressions this assertion
// catches (filter unbound, section label drift) are all covered by a
// text query over the section header buttons.
const countBySection = await orcaPage
.locator('button.uppercase')
.describe('Source Control section header buttons')
.evaluateAll((buttons) => {
const result: Record<string, number> = {}
for (const button of buttons) {
const spans = button.querySelectorAll('span')
const label = spans[0]?.textContent?.trim() ?? ''
const count = Number.parseInt(spans[1]?.textContent?.trim() ?? '', 10)
if (label && Number.isFinite(count)) {
result[label] = count
}
}
return result
})
// Only the untracked `notes-…md` file should match "notes-". The
// modified README row would be hidden. A regression where the filter
// stops narrowing (e.g. someone unbinds the onChange handler) would
// leave "Changes" visible with count >= 1.
expect(countBySection['Untracked Files']).toBe(1)
expect(countBySection['Changes']).toBeUndefined()
} finally {
cleanupSeed(testRepoPath, untrackedPath)
}
})
test('populated Source Control panel matches the expected layout', async ({
orcaPage,
testRepoPath