pi-lens 4.3.0 → 4.4.0

package/dist/clients/lsp-mutation.js

@@ -1,396 +0,0 @@
-import * as path from "node:path";
-import { BoundedSet } from "./bounded-cache.js";
-import { createReadGuardEditBatchSummary, getReadGuardCorrelationId, logReadGuardEvent, } from "./read-guard-logger.js";
-import { normalizeMapKey } from "./path-utils.js";
-import { getMutationBridge } from "./mutation-bridge.js";
-import { noteMutationHandled } from "./observed-mutation.js";
-import { recordDegradationOnce } from "./degradation-ledger.js";
-// #2450 fix round 3 (minor): gates the "bridge unavailable" dbg line to fire
-// once per session, not once per FILE — a rename that touches many files in
-// a bridge-less process (e.g. the MCP server) would otherwise spam the same
-// line once per touched file. `recordDegradationOnce` itself has no
-// first-occurrence return value (unlike `incrementDegradationCount`), so this
-// tiny local latch is the minimal way to mirror that "once" intent onto the
-// dbg line without changing the ledger's own once-per-(kind,subject) record.
-//
-// Catalog shape 17 (a process-lifetime latch must re-arm at session_start):
-// wired into `handleSessionStart` (`clients/runtime-session.ts`) directly,
-// NOT into `resetDegradationLedger()` — that lives in `degradation-ledger.ts`,
-// which this module already imports, and reaching back from there to here
-// would be a cycle.
-let noBridgeDbgLogged = false;
-export function resetLspMutationNoBridgeDbgLatch() {
-    noBridgeDbgLogged = false;
-}
-/**
- * Test-only peek at the latch's current value (#2450 fix round 5, F2) — lets
- * the session-state registry probe the actual internal state instead of
- * inferring it from captured dbg messages, the same pattern every other
- * process-private-scalar probe in `tests/support/session-state-registry.ts`
- * uses (e.g. `_observedMutationStateForTests`).
- */
-export function _lspMutationNoBridgeDbgLoggedForTests() {
-    return noBridgeDbgLogged;
-}
-const MAX_SAMPLES = 100;
-const MAX_SUMMARIES_PER_CONTEXT = 100;
-export function newLspMutationCorrelationId(toolCallId) {
-    return getReadGuardCorrelationId(toolCallId ? { toolCallId } : {});
-}
-function allResults(options) {
-    return options.results ?? [];
-}
-function combineResults(results) {
-    const files = new Set();
-    const fileDetails = [];
-    const appliedIndexes = [];
-    const paths = [];
-    let requestedTotal = 0;
-    let appliedTotal = 0;
-    let textEdits = 0;
-    let create = 0;
-    let rename = 0;
-    let deleteCount = 0;
-    for (const result of results) {
-        requestedTotal += result.operationTotal;
-        appliedTotal += result.appliedOperationTotal;
-        textEdits += result.operationCounts.textEdits;
-        create += result.operationCounts.create;
-        rename += result.operationCounts.rename;
-        deleteCount += result.operationCounts.delete;
-        for (const index of result.appliedOperationIndexes)
-            appliedIndexes.push(index);
-        for (const file of result.files) {
-            if (!files.has(file)) {
-                files.add(file);
-                paths.push(file);
-            }
-        }
-        fileDetails.push(...result.fileDetails);
-    }
-    return {
-        requestedTotal,
-        appliedTotal,
-        appliedIndexes,
-        files: [...files],
-        fileDetails,
-        textEdits,
-        create,
-        rename,
-        delete: deleteCount,
-        paths,
-    };
-}
-function uniqueDetails(files, fileDetails) {
-    const byPath = new Map();
-    // #2016: `files` and `fileDetails` name the same paths, so without this the
-    // map-build loop and the lookup below each pay `realpathSync.native` for the
-    // same path (~200 microseconds per call on Windows; since #3098 POSIX pays it
-    // too, measured at 1.8 microseconds for an existing path — still under any CI
-    // timing gate, so the waste stays invisible either way). The memo lives for
-    // one call, so it has no staleness window at all and needs no freshness
-    // design.
-    const keyMemo = new Map();
-    const keyFor = (filePath) => {
-        let key = keyMemo.get(filePath);
-        if (key === undefined) {
-            key = normalizeMapKey(path.resolve(filePath));
-            keyMemo.set(filePath, key);
-        }
-        return key;
-    };
-    for (const detail of fileDetails) {
-        const key = keyFor(detail.filePath);
-        const previous = byPath.get(key);
-        if (!previous) {
-            byPath.set(key, detail);
-            continue;
-        }
-        byPath.set(key, {
-            filePath: previous.filePath,
-            range: previous.range && detail.range
-                ? {
-                    start: Math.min(previous.range.start, detail.range.start),
-                    end: Math.max(previous.range.end, detail.range.end),
-                }
-                : (previous.range ?? detail.range),
-            importsChanged: previous.importsChanged || detail.importsChanged,
-        });
-    }
-    return files.map((filePath) => byPath.get(keyFor(filePath)) ?? {
-        filePath,
-        // Resource operations have no already-computed text range. A small
-        // range is still enough to invalidate the touched-file turn state;
-        // never synchronously re-read the whole file here.
-        range: { start: 1, end: 1 },
-        importsChanged: true,
-    });
-}
-function bookkeepLspMutation(context, files, fileDetails) {
-    const details = uniqueDetails(files, fileDetails);
-    // The direct path below needs `context.runtime` (the receipt) and
-    // `context.cacheManager` (the turn-state entry) threaded in by the caller.
-    // One caller structurally cannot: `workspace/applyEdit`'s server-initiated-edit
-    // handler (clients/lsp/client.ts) builds its own fallback `LspMutationContext`
-    // with neither, because there is no live reference to the runtime/cache-manager
-    // singletons at that call site. Rather than silently drop bookkeeping for that
-    // write, fall back to the mutation bridge (#2423) — mounted once at extension
-    // activation with live getters closing over the SAME singletons — which is
-    // exactly the seam built for "a producer that cannot reach the bookkeeping
-    // surfaces directly". This keeps `recordLspMutation` the ONE call site every
-    // LSP-applied edit bookkeeps through; the bridge is this function's OWN
-    // fallback, not a second seam other code reaches for on its behalf (#2450).
-    // Only the receipt (recordProjectMutation) and the turn-state entry
-    // (cacheManager.addModifiedRange) need the bridge fallback: the read-guard
-    // stamp is independent of them and already tolerates either being absent.
-    const useBridgeFallback = !context.runtime || !context.cacheManager;
-    for (const detail of details) {
-        const filePath = path.resolve(detail.filePath);
-        // #2450 review round 2 (F4): the SAME recordability gate the bridge
-        // fallback below already applies internally (`isRecordable`, mounted in
-        // index.ts). Applied here too so the direct (deps-threaded) branch
-        // cannot record a write the fallback branch would have silently
-        // dropped for the exact same reason (`no-read-guard` / ignored /
-        // vendor) — the two branches must agree on which files count as
-        // project source, not just on how they bookkeep the ones that do.
-        if (context.isRecordable && !context.isRecordable(filePath)) {
-            context.dbg?.(`lsp mutation not recordable, skipping ${filePath}`);
-            continue;
-        }
-        if (context.readGuard) {
-            try {
-                context.readGuard.recordWritten(filePath);
-            }
-            catch (err) {
-                context.dbg?.(`lsp mutation read-guard stamp failed for ${filePath}: ${err}`);
-            }
-        }
-        if (useBridgeFallback) {
-            const bridge = getMutationBridge();
-            if (!bridge) {
-                // #2450 review round 2 (F4)/round 3 (minor): the MCP server process
-                // (`mcp/server.ts`) builds `lsp_navigation` with no
-                // `mutationDeps` AND never mounts the bridge (that only happens
-                // inside pi's own in-process extension activation, `index.ts`) —
-                // every LSP-applied edit in that process is unrecorded. Once per
-                // session (not once per file — a rename can touch many files in
-                // one call, and a session can issue many renames), not once per
-                // process lifetime silently. `subject` is a fixed, one-word
-                // degradation-ledger key (round 3: was `context.tool`, which
-                // varies per LSP operation — "lsp_navigation:rename" vs
-                // "...executeCommand" — so the SAME session could log this
-                // degradation once per operation kind instead of truly once).
-                recordDegradationOnce({
-                    kind: "lsp-mutation-bridge-unmounted",
-                    subject: "lsp-mutation:no-bridge",
-                    reason: "workspace/applyEdit bookkeeping fell back to the mutation " +
-                        "bridge, but no bridge is mounted in this process (e.g. the " +
-                        "MCP server, which never runs pi's extension activation) — " +
-                        "the write is unrecorded (no read-guard stamp, no turn-state " +
-                        "entry, no change-log receipt).",
-                });
-                if (!noBridgeDbgLogged) {
-                    noBridgeDbgLogged = true;
-                    context.dbg?.(`lsp mutation bridge unavailable, dropping bookkeeping for ${filePath}`);
-                }
-            }
-            else {
-                try {
-                    const recorded = bridge.recordMutation({
-                        filePath,
-                        kind: "edit",
-                        editRanges: detail.range
-                            ? [[detail.range.start, detail.range.end]]
-                            : undefined,
-                        consumer: context.tool,
-                        // Real value threaded through, not the bridge's own
-                        // historical `false` default (#2450 review round 2, F1) —
-                        // the tsserver organize-imports/add-import case is exactly
-                        // the import-changing one.
-                        importsChanged: detail.importsChanged ?? true,
-                        // Equivalence with the direct branch below, which never
-                        // enqueues a deferred autofix/format pass for an LSP-applied
-                        // edit (#2450 review round 2, F3).
-                        deferAutofix: false,
-                    });
-                    if (!recorded) {
-                        context.dbg?.(`lsp mutation bridge dropped ${filePath} (recordMutation() ` +
-                            "returned false — see the mutation_bridge debug output " +
-                            "for the reason: malformed entry, out-of-scope path, or " +
-                            "a bookkeeping error)");
-                    }
-                }
-                catch (err) {
-                    context.dbg?.(`lsp mutation bridge fallback failed for ${filePath}: ${err}`);
-                }
-            }
-        }
-        else {
-            const runtime = context.runtime;
-            // One mutation seam (#2000 phase 1): bump + receipt + change-log live in
-            // RuntimeCoordinator.recordProjectMutation.
-            runtime?.recordProjectMutation?.({
-                filePath,
-                source: context.source,
-                cwd: context.cwd,
-                changedRange: detail.range,
-                onAppendError: (err) => context.dbg?.(`lsp mutation project change append failed for ${filePath}: ${err}`),
-            });
-            if (context.cacheManager) {
-                try {
-                    context.cacheManager.addModifiedRange(filePath, detail.range ?? { start: 1, end: 1 }, detail.importsChanged ?? true, context.cwd, runtime?.telemetrySessionId, undefined, 
-                    // #2504 review round 2 (F1). #2504 added a containment
-                    // filter to `addModifiedRange` and gave it `cwd` as the
-                    // project root. On THIS path `cwd` is the calling
-                    // directory, which `tools/lsp-navigation.ts` scopes to a
-                    // sub-package while the LSP client's root spans the whole
-                    // monorepo — so a server-initiated edit on a sibling
-                    // package (#2450) and every edit whose context carries a
-                    // separate bookkeeping cwd (#2479) were silently dropped
-                    // from the worklist. Same root the `isRecordable` gate a
-                    // few frames up already uses: `runtime.projectRoot`,
-                    // falling back to `cwd` when no runtime is threaded.
-                    runtime?.projectRoot ?? context.cwd);
-                }
-                catch (err) {
-                    context.dbg?.(`lsp mutation modified-range tracking failed for ${filePath}: ${err}`);
-                }
-            }
-            // #2465: the bridge-fallback branch above gets this for free —
-            // `recordMutationThroughSeam` calls `noteMutationHandled` for every
-            // entry it takes. The DIRECT branch bypasses the bridge entirely, so
-            // it has to say so itself, or the #2430 `agent_settled` sweep finds
-            // bytes on disk it has no baseline for and reports every
-            // `lsp_navigation` rename as drift no tool call explains. Placed
-            // after `addModifiedRange` and OUTSIDE its `try`: the mark is about
-            // "pi-lens wrote this file", which is true whether or not the
-            // turn-state insert threw.
-            noteMutationHandled(filePath);
-        }
-        // #2450 review round 2 (Partial-deps): its own surface, independent of
-        // whether the receipt/turn-state used the direct path or the bridge
-        // fallback above — a half-threaded caller (one that provides
-        // `recordAutofix` but not `runtime`/`cacheManager`) must not lose its
-        // turn-summary publisher just because SOME other surface fell back.
-        if (context.recordAutofix && context.source === "autofix") {
-            const key = normalizeMapKey(filePath);
-            const seen = context.autofixRecordedPaths ?? new BoundedSet(MAX_SAMPLES);
-            context.autofixRecordedPaths = seen;
-            if (!seen.has(key)) {
-                seen.add(key);
-                context.recordAutofix(filePath);
-            }
-        }
-    }
-    if (context.publishFilesTouched && files.length > 0) {
-        context.publishFilesTouched(files);
-    }
-}
-function telemetryFor(context, options) {
-    const combined = combineResults(allResults(options));
-    const requestedTotal = options.requestedTotal ?? combined.requestedTotal;
-    const appliedTotal = combined.appliedTotal;
-    const requestedIndexes = Array.from({ length: Math.min(requestedTotal, MAX_SAMPLES) }, (_, index) => index);
-    const rejectedTotal = Math.max(0, requestedTotal - appliedTotal);
-    const rejectedReasons = Array.from({ length: Math.min(rejectedTotal, MAX_SAMPLES) }, (_, index) => ({ index, code: "write_failed" }));
-    const status = options.status ?? (appliedTotal > 0 ? "success" : "skipped");
-    const editBatchSummary = createReadGuardEditBatchSummary({
-        requestedIndexes,
-        requestedTotal,
-        resolvedIndexes: requestedIndexes,
-        resolvedTotal: requestedTotal,
-        rejectedReasons,
-        rejectedTotal,
-        appliedIndexes: combined.appliedIndexes,
-        appliedTotal,
-        participantIds: [context.correlationId],
-        participantTotal: 1,
-        commitStatus: status === "failed"
-            ? "failed"
-            : appliedTotal > 0
-                ? "committed"
-                : "no_changes",
-        terminalStatus: status,
-    });
-    return {
-        editBatchSummary,
-        operationCounts: {
-            requested: requestedTotal,
-            applied: appliedTotal,
-            textEdits: combined.textEdits,
-            create: combined.create,
-            rename: combined.rename,
-            delete: combined.delete,
-        },
-        sampledPaths: combined.paths.slice(0, MAX_SAMPLES),
-        sampledPathsTotal: combined.paths.length,
-        sampledPathsTruncated: combined.paths.length > MAX_SAMPLES,
-        considered: options.considered,
-        completed: options.completed,
-        failedCount: options.failedCount,
-    };
-}
-export function recordLspMutation(context, options = {}) {
-    const results = allResults(options);
-    const combined = combineResults(results);
-    if (options.bookkeep !== false && combined.files.length > 0) {
-        bookkeepLspMutation(context, combined.files, combined.fileDetails);
-    }
-    const telemetry = telemetryFor(context, options);
-    if (context.emitSummary !== false) {
-        const summaryCount = context.summaryCount ?? 0;
-        context.summaryEmitted = true;
-        if (summaryCount < MAX_SUMMARIES_PER_CONTEXT) {
-            context.summaryCount = summaryCount + 1;
-            logReadGuardEvent({
-                event: "edit_batch_summary",
-                correlationId: context.correlationId,
-                filePath: combined.files[0] ?? context.cwd,
-                metadata: {
-                    tool: context.tool,
-                    source: context.source,
-                    outcome: telemetry.editBatchSummary.terminalStatus,
-                    // The outer correlation identifies the soliciting tool call; the
-                    // bounded sequence distinguishes multiple applyEdit requests within it.
-                    summaryIndex: summaryCount,
-                    editBatchSummary: telemetry.editBatchSummary,
-                    operationCounts: telemetry.operationCounts,
-                    sampledPaths: telemetry.sampledPaths,
-                    sampledPathsTotal: telemetry.sampledPathsTotal,
-                    sampledPathsTruncated: telemetry.sampledPathsTruncated,
-                    considered: telemetry.considered,
-                    completed: telemetry.completed,
-                    failedCount: telemetry.failedCount,
-                },
-            });
-        }
-        else if (!context.summaryOverflowed) {
-            context.summaryOverflowed = true;
-            logReadGuardEvent({
-                event: "edit_batch_summary_overflow",
-                correlationId: context.correlationId,
-                filePath: combined.files[0] ?? context.cwd,
-                metadata: {
-                    tool: context.tool,
-                    source: context.source,
-                    summaryLimit: MAX_SUMMARIES_PER_CONTEXT,
-                    suppressedSummaries: "one or more",
-                },
-            });
-        }
-    }
-    return telemetry;
-}
-export function recordLspMutationBatch(context, options) {
-    const previous = context.emitSummary;
-    context.emitSummary = true;
-    try {
-        return recordLspMutation(context, options);
-    }
-    finally {
-        context.emitSummary = previous;
-    }
-}
-export function recordLspMutationOutcome(context, status) {
-    return recordLspMutation(context, { status, requestedTotal: 0 });
-}

package/dist/clients/lsp-status.js

@@ -1,59 +0,0 @@
-/**
- * Footer LSP-status selection (#170).
- *
- * Turns three raw inputs — the alive server ids (#267), the raw failed-spawn
- * server ids (widget-state), and the file-kinds in use this session — into the
- * two id lists the footer renders: `LSP Active: …` (green) and `LSP Failed: …`
- * (red). Kept as a pure function so the policy is unit-testable without spawning
- * a server or touching module state.
- *
- * Failure is a per-LANGUAGE property, not per-server:
- *   - A failed language server is surfaced ONLY when no live language-server
- *     sibling covers its extensions (so pyright-fails-but-jedi-succeeds reads as
- *     Active, never Failed), AND its kind is still in use this session (a failure
- *     for a language you're no longer editing is stale and dropped).
- *   - Auxiliary servers (opengrep/ast-grep) are cross-cutting scanners, not a
- *     language's LSP — they neither provide language coverage nor surface as a
- *     language failure. They still appear in the Active list via the alive set.
- */
-import { getFileKindsForExtension } from "./file-kinds.js";
-import { LSP_SERVERS } from "./lsp/server.js";
-function serverById(id) {
-    return LSP_SERVERS.find((s) => s.id === id);
-}
-export function selectLspStatus(aliveServerIds, failedServerIds, sessionKinds) {
-    // Extensions covered by an ALIVE language server (auxiliaries excluded — an
-    // alive opengrep must not make a failed python server look "covered").
-    const aliveSet = new Set(aliveServerIds);
-    const aliveLangExts = new Set();
-    for (const s of LSP_SERVERS) {
-        if (s.role === "auxiliary" || !aliveSet.has(s.id))
-            continue;
-        for (const ext of s.extensions)
-            aliveLangExts.add(ext.toLowerCase());
-    }
-    const sessionKindSet = new Set(sessionKinds);
-    const failedIds = [];
-    const seen = new Set();
-    for (const id of failedServerIds) {
-        if (seen.has(id))
-            continue;
-        seen.add(id);
-        const server = serverById(id);
-        if (!server || server.role === "auxiliary")
-            continue; // language servers only
-        const exts = server.extensions.map((e) => e.toLowerCase());
-        // (a) a live language sibling already covers this language → not a failure.
-        if (exts.some((e) => aliveLangExts.has(e)))
-            continue;
-        // (b) still in use this session? else the failure is stale.
-        const kinds = new Set();
-        for (const e of exts)
-            for (const k of getFileKindsForExtension(e))
-                kinds.add(k);
-        if (![...kinds].some((k) => sessionKindSet.has(k)))
-            continue;
-        failedIds.push(id);
-    }
-    return { activeIds: [...aliveServerIds], failedIds };
-}

package/dist/clients/map-with-concurrency.js

@@ -1,40 +0,0 @@
-/**
- * Bounded worker pool — the repo's one shared `mapWithConcurrency`.
- *
- * Lived in `dependency-checker.ts` (its first caller) and was exported from
- * there for `dispatch/runners/biome-check.ts`. #2504 added a third caller,
- * `runtime-turn.ts`'s bounded test-runner batch — and runtime-turn must not
- * drag an ANALYZER client into its eager import graph for a 15-line helper
- * (that is exactly the eager-bootstrap cost #2467 removed). So the value moved
- * to this zero-dependency leaf and `dependency-checker.ts` re-exports it: every
- * existing importer keeps its specifier, and a consumer that only needs the
- * pool no longer loads madge.
- *
- * Same shape as `ledger-bounds.ts` (#2426) for the degradation ledger's bound.
- *
- * Callers that need results use the result-returning overload; callers that
- * only need completion use the void overload. Keeping both forms here prevents
- * per-caller worker-pool copies from drifting.
- */
-export async function mapWithConcurrency(items, concurrency, mapper, signal) {
-    if (items.length === 0)
-        return [];
-    let nextIndex = 0;
-    const workerCount = Math.max(1, Math.min(concurrency, items.length));
-    const results = new Map();
-    const worker = async () => {
-        while (true) {
-            if (signal?.aborted)
-                return;
-            const index = nextIndex++;
-            if (index >= items.length)
-                return;
-            results.set(index, (await mapper(items[index])));
-        }
-    };
-    const workers = Array.from({ length: workerCount }, () => worker());
-    await Promise.all(workers);
-    return [...results.entries()]
-        .sort(([left], [right]) => left - right)
-        .map(([, result]) => result);
-}