Agents - switch to using GraphQL for the PullRequestModel

[lszomoru]

No description provided.

[github-actions]

Screenshot Changes

Base: b3a73624 Current: 8e902a11

Changed (3)

chat/aiCustomizations/aiCustomizationManagementEditor/McpBrowseMode/Dark

Before	After

editor/inlineCompletions/other/JumpToHint/Dark

Before	After

agentSessionsViewer/NeedsInput/Dark

Before	After

[Copilot]

Pull request overview

This PR updates the Agents window GitHub PR integration to fetch a single “PR snapshot” via GraphQL (PR details + mergeability + review threads) and adjusts the reactive model, service usage, and tests accordingly.

Changes:

• Replace the PR fetcher’s REST/GraphQL mix with a single GraphQL snapshot query and mapping.
• Simplify GitHubPullRequestModel refresh behavior to one fetch + atomic observable updates (and remove refreshThreads).
• Update CodeReviewService and browser tests to the new refresh/fetcher shape.

Show a summary per file

File	Description
src/vs/sessions/contrib/github/browser/fetchers/githubPRFetcher.ts	Switch to GraphQL snapshot query and derive PR/mergeability/threads from one response.
src/vs/sessions/contrib/github/browser/models/githubPullRequestModel.ts	Refresh now fetches a full snapshot and updates observables in a transaction; refreshThreads removed.
src/vs/sessions/contrib/codeReview/browser/codeReviewService.ts	Use prModel.refresh() for initial load (instead of threads-only refresh).
src/vs/sessions/contrib/github/test/browser/githubFetchers.test.ts	Update tests to validate the new GraphQL snapshot behavior and return shape.
src/vs/sessions/contrib/github/test/browser/githubModels.test.ts	Update model tests to match the new fetcher contract and refresh behavior.

Copilot's findings

Comments suppressed due to low confidence (1)

src/vs/sessions/contrib/github/browser/fetchers/githubPRFetcher.ts:147

• The GraphQL snapshot query hard-limits reviews, reviewThreads, and per-thread comments to first: 100 without any pagination. This can lead to missing threads/comments in the UI, and (more importantly) incorrect mergeability blocker detection when there are >100 reviews or threads. Consider adding pageInfo { hasNextPage endCursor } and paging with after, or at least requesting the most recent items explicitly (e.g. last: 100) and treating truncated data as Unknown mergeability.

	'      reviews(first: 100) {',
	'        nodes {',
	'          state',
	'          submittedAt',
	'          author {',
	'            login',
	'          }',
	'        }',
	'      }',
	'      reviewThreads(first: 100) {',
	'        nodes {',
	'          id',
	'          isResolved',
	'          path',
	'          line',
	'          comments(first: 100) {',
	'            nodes {',

• Files reviewed: 5/5 changed files
• Comments generated: 2

[Copilot]

prModel.refresh().catch(...) is unlikely to ever run because GitHubPullRequestModel.refresh() swallows fetch errors internally (it logs but doesn’t rethrow). As a result, the PR review UI won’t transition to Error on refresh failures, and the log message also no longer matches what’s being fetched (it’s a full PR snapshot, not just threads). Consider either propagating refresh errors from the model (rethrow / return a result) or handling failures via a dedicated observable/error signal; also update the log message accordingly.

See below for a potential fix:

		// Start polling and trigger the initial PR snapshot refresh.
		void prModel.refresh();