@gitkraken/core-gitlens 0.6.0 → 0.7.1

package/src/plus/integrations/reads/ordering.ts

@@ -1,124 +0,0 @@
-import type { IssueShape, IssueSorting } from '../../../git/models/issue.js';
-import { defaultIssueSort } from '../../../git/models/issue.js';
-import { getIssueComparator } from '../../../git/utils/issue.utils.js';
-
-/**
- * How an issue read decides what order to request, and how to order a page it assembled itself.
- *
- * Its own module rather than more of `filters.ts` because it answers a different question. That file validates what
- * a caller asked to NARROW to; this one resolves what a caller asked to ORDER by — and unlike a filter, an order
- * doesn't only have to be expressible by the provider, it also has to survive the read's own merging. Every issue
- * read makes both decisions in the same shape, so they are resolved here once rather than at each read.
- */
-
-/**
- * Why {@link resolveIssueSort} refused a key, carrying what the refusal has to say — the same shape
- * `IssueSearchCriteriaRejection` uses for the filtered search, and for the same reason.
- *
- * The payload travels WITH the rejection rather than being re-read at the warning: the check already held both
- * values, so handing them back is what lets a caller word the refusal without re-deriving anything TypeScript
- * can no longer narrow (which is how this grew five non-null assertions), and it makes it impossible for the
- * message to name a supported list the check didn't use.
- */
-export type UnsupportedIssueSortRejection = {
-	reason: 'unsupported-sort';
-	/** The key the caller asked for. Non-optional, unlike the `sort` a successful resolve returns. */
-	requested: IssueSorting;
-	/** Every key this surface CAN express, for the message. `undefined` when the read can't be ordered at all. */
-	supported: IssueSorting[] | undefined;
-};
-
-/**
- * Resolves the order an issue read should request: validates a caller-supplied key against what one provider
- * surface can express server-side, and supplies this facade's default when the caller asked for none.
- *
- * Three outcomes, and each is a decision rather than a convenience:
- * - A key the surface can't express is REJECTED and the read refused whole, never resolved to the nearest
- *   supported key. Same reason a dropped filter isn't dropped: the provider serves at most a bounded window of
- *   matches, so a list ordered by something else is a DIFFERENT subset than the one asked for, and the cursor that
- *   comes with it describes that other subset — indistinguishable, from the result, from the order requested.
- * - An omitted key becomes {@link defaultIssueSort}, because leaving it to the provider is what makes the same
- *   read return different things on different hosts (Azure by creation, Jira by nothing at all). The default is
- *   this facade's promise; the SDK deliberately has no opinion.
- * - An omitted key on a surface that can't express even the default returns `undefined`, leaving that provider's
- *   own order. A default the caller never asked for must not refuse a read that works today.
- *
- * Takes the supported list rather than an `IntegrationIds`, unlike the filter validators, because which list
- * applies is the CALLER's decision — `supportedIssueSorts` for a repo-scoped or tracker read,
- * `supportedAccountWideIssueSorts` for an account-wide one — and that branch already exists at the call site
- * (`repos` present or not). Resolving it in here would duplicate it.
- */
-export function resolveIssueSort(
-	supported: IssueSorting[] | undefined,
-	sort: IssueSorting | undefined,
-): { rejection?: UnsupportedIssueSortRejection; sort?: IssueSorting } {
-	if (sort == null) {
-		return supported?.includes(defaultIssueSort) ? { sort: defaultIssueSort } : {};
-	}
-	if (!supported?.includes(sort)) {
-		return { rejection: { reason: 'unsupported-sort', requested: sort, supported: supported } };
-	}
-
-	return { sort: sort };
-}
-
-/**
- * Why a merged page can't honor the key it was given: the read fans out over several provider queries and no
- * normalized issue carries the field to merge on.
- *
- * Named for the same reason {@link UnsupportedIssueSortRejection} is, and kept separate from it because the two
- * refusals are different claims: that one says the PROVIDER can't express the key, this one says the provider
- * can but this READ's shape can't — the same key against the same provider succeeds on a single scope.
- */
-export type UnmergeableIssueSort = {
-	/** The key the caller asked for, which a single-scope read of this provider would have honored. */
-	requested: IssueSorting;
-};
-
-/**
- * What a read does with the order it resolved, once it also knows whether the page it is about to build MERGES
- * several provider queries.
- *
- * Both facts are bound here, at the one point a read knows them, rather than consulted separately at the refusal
- * and again at the sort. That is the difference this type buys: `merged` and the comparator were each tested twice
- * per read, and the two tests are the same decision — whether the requested order survives this page's shape.
- */
-export type IssueOrdering = {
-	/** The key to send to the provider, or `undefined` to leave the provider's own order. */
-	readonly sort: IssueSorting | undefined;
-	/**
-	 * Set when the requested key cannot be honored: this page merges several queries, and no normalized issue
-	 * carries the field to merge on. The read refuses rather than serving concatenated per-scope runs, which would
-	 * look ordered without being so. `undefined` for a single-query page, where the provider did the ordering.
-	 *
-	 * Carries the key rather than being a bare flag, for the same reason {@link UnsupportedIssueSortRejection}
-	 * does: the refusal needs it, and re-reading `sort` at the call site needs an assertion for an invariant this
-	 * type would otherwise not express.
-	 */
-	readonly unmergeable: UnmergeableIssueSort | undefined;
-	/**
-	 * Orders a merged page. A no-op for a single-query page (the provider already ordered it, and re-sorting could
-	 * only reproduce that order while hiding a provider that ignored the key) and for a read with no key at all.
-	 *
-	 * Returns a new array rather than sorting in place: the input is often the provider's own `values`, and a read
-	 * that mutates what it was handed is a surprise waiting for the next caller to share that reference.
-	 */
-	order(items: IssueShape[]): IssueShape[];
-};
-
-/**
- * Binds a resolved sort key to the shape of the page about to be built.
- *
- * `merged` is the caller's own count of provider queries reduced to the only thing that matters: GitHub sends one
- * search however many repositories are named, GitLab one per repository, Azure and the trackers one per project,
- * and every account-wide read is a union of several. Above one, the union is what gets published, and only a field
- * an {@link IssueShape} carries can order it.
- */
-export function toIssueOrdering(sort: IssueSorting | undefined, merged: boolean): IssueOrdering {
-	const comparator = sort != null ? getIssueComparator(sort) : undefined;
-	return {
-		sort: sort,
-		unmergeable: merged && sort != null && comparator == null ? { requested: sort } : undefined,
-		order: (items: IssueShape[]) => (merged && comparator != null ? [...items].sort(comparator) : items),
-	};
-}

package/src/plus/integrations/reads/paging.ts

@@ -1,379 +0,0 @@
-import type { CollectionMetadata } from '@gitkraken/provider-apis';
-import type { PagedResult } from '../../../utils/paging.js';
-import { PagingMode } from '../providers/models.js';
-import { mergeCollectionMetadata, parsePageCursor, toPageCursor } from '../providers/utils/providerPaging.js';
-import type { ProviderPagedResult, ProviderPageInfo, ProviderWarning } from '../results.js';
-import { appendDedupedWarning } from '../results.js';
-
-/**
- * The paging mechanics shared by every paged read on the provider facade: translating the SDK's `paging` into
- * the page-oriented shape consumers get, deciding what position and what continuation a page may honestly
- * claim, and walking a cursor-only read forward to a requested page number.
- *
- * Kept in one module because the value of these rules is that they are the SAME on every read — the contract
- * they implement is documented once on {@link ProviderPageInfo} and on `IntegrationManager`'s paging section.
- * Copies of them drifting apart is exactly the defect this consolidation removed.
- */
-
-/**
- * Whether a repo-scoped read for this provider ADVANCES on a page number, so the facade may synthesize a
- * page-number cursor, `page.currentPage` may echo the request, and a next-page number is a usable continuation.
- *
- * Only `PagingMode.Repos` (GitHub/GHE) is excluded: its read is cursor-only, so it accepts a page number and
- * ignores it, answering with page 1 — which is why {@link drainToRequestedPage} exists for that mode.
- *
- * Every other provider, including one with no declared mode (Bitbucket Data Center), reads `page` as a 1-based
- * page number. That last part is only true as of `@gitkraken/provider-apis` 0.54.0: before it, Bitbucket Data
- * Center consumed `page` as a raw `start` ITEM OFFSET, so a synthesized page 3 asked for `start=3` — a window
- * overlapping page 1 — and this facade had to withhold the synthesized cursor from it. The SDK now converts
- * (`start = (page - 1) * limit`) and reports `nextPage` as a page number, so the guard is gone and that host
- * can be advanced by number like the rest.
- */
-export function isPageNumberAdvanceable(mode: PagingMode | undefined): boolean {
-	return mode !== PagingMode.Repos;
-}
-
-/**
- * Encodes a 1-based page number as the opaque cursor the provider paging layer understands, via the same
- * {@link toPageCursor} the providers encode with (its `parsePageCursor` counterpart is what reads these back).
- * Page 1 needs no cursor: it is the read's own starting position.
- */
-export function pageToCursor(page: number | undefined): string | undefined {
-	return page == null || page <= 1 ? undefined : toPageCursor(page);
-}
-
-/**
- * Filters out the provider paging layer's empty-cursor sentinel. `providersApi` seeds its next-cursor with
- * `'{}'` and leaves it there when the SDK reports another page without an `endCursor`, so `'{}'` means "no
- * usable continuation" and must never be threaded back as one.
- */
-export function usableCursor(cursor: string | undefined): string | undefined {
-	return cursor == null || cursor === '{}' ? undefined : cursor;
-}
-
-/**
- * The terminal page a paged read returns when it refuses the request outright — the surface doesn't apply (an
- * issue tracker asked for a repo read), the target couldn't be resolved, a capability is missing, or a filter
- * set isn't expressible server-side. Every one of those is "no page was served": empty `items`, no continuation,
- * and the reason carried in `warnings`.
- *
- * `currentPage` stays a caller-supplied parameter rather than being derived here because the refusals differ on
- * what position they can honestly claim: a read the provider advances by page number reports the requested page,
- * while a cursor-only account-wide read has no addressable position and reports 1 (see
- * {@link ProviderPageInfo.currentPage}).
- */
-export function refusedPage<T>(
-	currentPage: number,
-	warnings: ProviderWarning[],
-	fetchFailed: boolean,
-): ProviderPagedResult<T> {
-	return {
-		items: [],
-		warnings: warnings,
-		page: { currentPage: currentPage, itemsPerPage: 0 },
-		hasMore: false,
-		fetchFailed: fetchFailed || undefined,
-	};
-}
-
-/** A provider page normalized off the SDK's `paging`, before continuation and position are reconciled. */
-export interface NormalizedPage {
-	page: ProviderPageInfo;
-	hasMore: boolean;
-	cursor?: string;
-	truncated: boolean;
-}
-
-/**
- * Normalizes a `PagedResult.paging` into the page-oriented shape Kepler consumes: `page`, `hasMore`, and an
- * opaque `cursor` retained only for cursor-only hosts (where jumping straight to page N isn't possible, so the
- * caller threads the cursor back instead).
- */
-export function toProviderPageInfo(
-	itemsPerPage: number,
-	paging: { more?: boolean; cursor?: string; page?: number; pageSize?: number; truncated?: boolean } | undefined,
-): NormalizedPage {
-	let cursor: string | undefined;
-	let cursorPage: number | undefined;
-	const raw = paging?.cursor;
-	if (raw != null && raw !== '{}') {
-		try {
-			const parsed = JSON.parse(raw) as { type?: string; cursors?: unknown; page?: number };
-			// Retain opaque cursor strings for cursor-only hosts, per-repo/project cursor bundles for
-			// PagingMode.Repo/Project reads, AND page/offset cursors. The latter matters for reads with no
-			// caller-visible page param to increment — e.g. Bitbucket Server's account-wide PR read threads
-			// its next `start` offset as a `type:'page'` cursor; dropping it left the caller with
-			// `hasMore:true` and nothing to continue with. A page cursor is a valid opaque continuation, so
-			// threading it back is always safe even where a page number is also reported.
-			if (parsed.type === 'cursor' || parsed.type === 'page' || Array.isArray(parsed.cursors)) {
-				cursor = raw;
-				// Per-repo/project cursor bundles also carry the current page number so the facade can report the
-				// real page when the consumer continues using only the cursor.
-				if (Array.isArray(parsed.cursors) && parsed.page != null) {
-					cursorPage = parsed.page;
-				}
-			}
-		} catch {}
-	}
-	// Only echo a page number the provider actually honored. Numbered-page hosts report their own
-	// `paging.page`; cursor-only hosts (e.g. GitHub PR search) report none and ignore a synthesized
-	// page-number cursor — returning their first page — so echoing the requested `page` would mislabel
-	// page 1 as page N. Report page 1 in that case rather than the unapplied request. Per-repo/project
-	// bundles may carry the page explicitly in the cursor.
-	const currentPage = paging?.page ?? cursorPage ?? 1;
-	return {
-		page: {
-			currentPage: Math.max(1, currentPage),
-			itemsPerPage: paging?.pageSize ?? itemsPerPage,
-		},
-		hasMore: paging?.more ?? false,
-		cursor: cursor,
-		// A single-page provider read that couldn't confirm completeness carries `paging.truncated`;
-		// surface it so callers can flag `page.truncated` instead of publishing a partial read as complete.
-		truncated: paging?.truncated ?? false,
-	};
-}
-
-/**
- * Reconciles a provider's `hasMore` with the continuation it actually handed back. A provider can report
- * `hasNextPage: true` while omitting the `endCursor` (the paging layer surfaces that as the sentinel `'{}'`,
- * which {@link toProviderPageInfo} drops), leaving `hasMore: true` with no cursor — a consumer that pages while
- * `hasMore` would then re-request the same page forever. Prefer the real provider cursor; else synthesize the
- * next page number, but only for a read the provider actually advances by page number (`nextPage` omitted for
- * cursor-only reads, which ignore it and answer with their first page again); else report the read as
- * terminal-but-incomplete (`hasMore: false` + `truncated`).
- */
-export function resolveContinuation(
-	paged: { hasMore: boolean; cursor?: string; truncated: boolean },
-	nextPage: number | undefined,
-): { hasMore: boolean; cursor?: string; truncated: boolean } {
-	if (!paged.hasMore) return { hasMore: false, cursor: undefined, truncated: paged.truncated };
-	if (paged.cursor != null) return { hasMore: true, cursor: paged.cursor, truncated: paged.truncated };
-
-	const synthesized = nextPage != null ? pageToCursor(nextPage) : undefined;
-	if (synthesized != null) return { hasMore: true, cursor: synthesized, truncated: paged.truncated };
-
-	return { hasMore: false, cursor: undefined, truncated: true };
-}
-
-/**
- * Resolves the position a paged read reports as `page.currentPage`, per the single convention documented on
- * {@link ProviderPageInfo.currentPage}. Every paged read routes through here so the field means the same thing
- * on all of them: positional, never constant-1 for one read and positional for another.
- *
- * `providerPage` is what the provider (or the internal drain) established — 1 means "nothing reported".
- * `pageAdvanceable` is whether the read honors a page number at all: false for a cursor-only read, which
- * answers with its first page when handed a synthesized page-number cursor, so the requested `page` must NOT be
- * echoed there.
- */
-export function resolveCurrentPage(options: {
-	providerPage: number;
-	requestedPage: number;
-	suppliedCursor: string | undefined;
-	pageAdvanceable: boolean;
-}): number {
-	if (options.providerPage > 1) return options.providerPage;
-	// A caller-threaded cursor DID advance the provider, so the caller's own position is authoritative for a
-	// provider that reports none: prefer the page the cursor encodes, else the `page` supplied alongside it.
-	if (options.suppliedCursor != null) return parsePageCursor(options.suppliedCursor) ?? options.requestedPage;
-
-	return options.pageAdvanceable ? options.requestedPage : 1;
-}
-
-/**
- * One page of a FLAT cursor-paged provider read — the shape both issue reads' providers return directly.
- *
- * `hasMore` is optional because one of the two providers leaves it off on a terminal page; absent reads as "no
- * more", which is what a provider that reported no continuation means.
- */
-export interface FlatPage {
-	cursor?: string;
-	hasMore?: boolean;
-	page?: number;
-	truncated: boolean;
-}
-
-/** What {@link drainFlatPagesToRequestedPage} established by walking. */
-export interface FlatDrainResult<T extends FlatPage> {
-	/** The requested page, or `undefined` when a page failed mid-walk (which also sets `fetchFailed`). */
-	value: T | undefined;
-	/** The position actually reached — 1 when nothing advanced. */
-	currentPage: number;
-	truncated: boolean;
-	/** True when the requested page is past the provider's last one, so it is genuinely EMPTY. */
-	requestedPageMissing: boolean;
-	fetchFailed: boolean;
-}
-
-/**
- * Walks a FLAT cursor-paged read forward to a requested page number, for a provider whose pages are
- * `{cursor, hasMore, page, truncated}` rather than the SDK's `PagedResult` wrapper.
- *
- * The sibling of {@link drainToRequestedPage}, not a replacement for it: that one normalizes SDK `paging` through
- * {@link toProviderPageInfo}, which retains only cursors whose JSON declares `type`/`cursors` — and the composite
- * cursor of a GitHub aliased search declares neither, so it would be dropped and the walk would never advance.
- * Two shapes genuinely need two entry points; what they must NOT have is two copies of the rules below, which is
- * what this exists to prevent (see this module's own note on drift).
- *
- * The rules, all of them subtle enough to be worth stating once:
- * - The empty-cursor sentinel is filtered, so a provider claiming another page without handing back a usable
- *   continuation is not re-read with `'{}'` (which it would answer with page 1 again).
- * - A provider that returns the SAME cursor isn't advancing: stop, and report the read as truncated rather than
- *   looping forever or publishing a short page as complete.
- * - A page requested past the provider's terminal cursor is that EMPTY page N — never the last available page
- *   relabeled, per {@link ProviderPageInfo.currentPage}.
- * - A page that fails mid-walk latches `fetchFailed` and leaves no value, so the caller can tell "the read broke"
- *   from "the page was genuinely empty".
- *
- * `fold` is the one thing the callers differ on — one merges SDK collection metadata across the walked pages, the
- * other keeps the largest reported match count — so it is a callback rather than a branch.
- */
-export async function drainFlatPagesToRequestedPage<T extends FlatPage>(
-	first: { value?: T; warning?: ProviderWarning },
-	options: {
-		requestedPage: number;
-		suppliedCursor: string | undefined;
-		warnings: ProviderWarning[];
-		readPage: (cursor: string) => Promise<{ value?: T; warning?: ProviderWarning }>;
-		fold?: (page: T) => void;
-	},
-): Promise<FlatDrainResult<T>> {
-	let value = first.value;
-	let fetchFailed = first.warning != null && value == null;
-	let currentPage = value?.page ?? 1;
-	let truncated = value?.truncated ?? false;
-
-	const walkable = options.suppliedCursor == null && options.requestedPage > 1;
-	if (walkable && value != null) {
-		for (
-			let nextCursor = usableCursor(value.cursor);
-			currentPage < options.requestedPage && value.hasMore === true && nextCursor != null;
-			nextCursor = usableCursor(value.cursor)
-		) {
-			const next = await options.readPage(nextCursor);
-			if (next.warning != null) {
-				appendDedupedWarning(options.warnings, next.warning);
-			}
-			if (next.value == null) {
-				fetchFailed = next.warning != null;
-				value = undefined;
-				break;
-			}
-
-			value = next.value;
-			truncated = truncated || value.truncated;
-			options.fold?.(value);
-			currentPage = value.page ?? currentPage + 1;
-			if (usableCursor(value.cursor) === nextCursor) {
-				truncated = true;
-				break;
-			}
-		}
-	}
-
-	return {
-		value: value,
-		currentPage: currentPage,
-		truncated: truncated,
-		// Every case reduces to "a page was asked for by number and the walk fell short", including the one where
-		// no walk ran at all because the first page was already terminal.
-		requestedPageMissing: walkable && currentPage < options.requestedPage,
-		fetchFailed: fetchFailed,
-	};
-}
-
-/** The mutable state {@link drainToRequestedPage} carries across the pages it walks. */
-export interface DrainState<T> {
-	items: T[];
-	paged: NormalizedPage;
-	metadata: CollectionMetadata | undefined;
-	fetchFailed: boolean;
-}
-
-/**
- * Advances a cursor-only read to the requested page when the caller supplied only `page`.
- *
- * A cursor-only read accepts a synthesized page-number cursor and ignores it, answering with its first page, so
- * the requested page has to be reached by walking the provider's own opaque continuations. Only the last
- * successfully-read page's items are kept — returning pages 1..N as "page N" would duplicate items for a normal
- * paged consumer — while warnings and metadata are merged across the whole drained prefix.
- *
- * Shared by the repo-scoped and account-wide PR reads and the repo-scoped issue read so the three cannot drift:
- * a requested page past the provider's terminal cursor is an EMPTY page N on all of them, per
- * {@link ProviderPageInfo.currentPage}, never the last available page relabeled.
- */
-export async function drainToRequestedPage<T>(
-	state: DrainState<T>,
-	options: {
-		requestedPage: number;
-		itemsPerPage: number | undefined;
-		warnings: ProviderWarning[];
-		readPage: (cursor: string) => Promise<{
-			value?: (PagedResult<T> & { metadata?: CollectionMetadata }) | undefined;
-			warning?: ProviderWarning;
-		}>;
-	},
-): Promise<DrainState<T>> {
-	let { items, metadata, fetchFailed } = state;
-	let currentPage = 1;
-	let currentCursor = usableCursor(state.paged.hasMore ? state.paged.cursor : undefined);
-	let currentHasMore = state.paged.hasMore && currentCursor != null;
-	let currentTruncated = state.paged.truncated;
-	// A page requested past the terminal cursor is that empty page N. Distinguished from the first read
-	// having failed outright, which is already reported as page N with no items by the caller's own state.
-	const missRequestedPage = () => {
-		items = [];
-		currentPage = options.requestedPage;
-		currentCursor = undefined;
-		currentHasMore = false;
-	};
-
-	if (fetchFailed) {
-		missRequestedPage();
-	}
-
-	while (currentPage < options.requestedPage && currentHasMore && currentCursor != null) {
-		const { value, warning } = await options.readPage(currentCursor);
-		if (warning != null) {
-			appendDedupedWarning(options.warnings, warning);
-		}
-		if (value == null) {
-			fetchFailed = true;
-			missRequestedPage();
-			break;
-		}
-
-		items = value.values;
-		metadata = mergeCollectionMetadata(metadata, value.metadata);
-		const next = toProviderPageInfo(options.itemsPerPage ?? value.values.length, value.paging);
-		currentPage++;
-		currentTruncated ||= next.truncated;
-		const nextCursor = usableCursor(next.cursor);
-		// A provider that hands back the same cursor (or none) isn't advancing; stop rather than refetch
-		// the same page forever.
-		if (nextCursor == null || nextCursor === currentCursor) {
-			currentCursor = undefined;
-			currentHasMore = false;
-			break;
-		}
-
-		currentCursor = nextCursor;
-		currentHasMore = next.hasMore;
-	}
-
-	if (currentPage < options.requestedPage) {
-		missRequestedPage();
-	}
-
-	return {
-		items: items,
-		paged: {
-			page: { currentPage: currentPage, itemsPerPage: options.itemsPerPage ?? items.length },
-			hasMore: currentHasMore,
-			cursor: currentCursor,
-			truncated: currentTruncated,
-		},
-		metadata: metadata,
-		fetchFailed: fetchFailed,
-	};
-}