@herbertgao/pi-extensions 2026.10.0 → 2026.10.2

package/node_modules/pi-lens/dist/clients/diagnostic-line-freshness.js

@@ -1,305 +0,0 @@
-/**
- * Past-EOF validity gate for cited diagnostic line numbers (#1641).
- *
- * A live session served diagnostics at lines 407-410 of a file that was 402
- * lines on disk and UNMODIFIED — the LSP's in-memory document had drifted
- * longer than disk without any file write happening (no didChange lost, just
- * an in-memory buffer that grew past what the file actually contains). That
- * makes this a DIFFERENT invariant than #1622/#1627's freshness gate: those
- * compare a cited file's MTIME against a scan timestamp, which says nothing
- * when the file was never touched on disk at all. The cheap, orthogonal check
- * available here is structural: a diagnostic whose start line exceeds the
- * file's CURRENT line count cannot possibly be describing current content,
- * regardless of what any timestamp says.
- *
- * CONTRACT:
- *   - A diagnostic citing a line beyond the file's current line count is
- *     DEMOTED (`stale: true`), never dropped — the underlying issue may well
- *     still be real, just at a coordinate this gate can no longer vouch for.
- *   - RE-ARMS, never latches: the verdict is recomputed from the gate's own
- *     inputs (current mtime, current line count) on every call, so a
- *     TRANSIENT shrink (truncate-then-write, a formatter pass, a checkout)
- *     that later restores the line un-demotes on its own next read. A
- *     persisted `stale: true` a caller stores between calls is a cache of the
- *     last-derived verdict, never an independent source of truth that
- *     overrides the next derivation — the #1633-V1 lesson: derive, don't latch.
- *   - Fail open: a file that cannot be stat'ed/read, or exceeds the size gate,
- *     yields no verdict at all (every diagnostic passes through unchanged).
- *     An unreadable/oversized file is not evidence of anything; demoting on
- *     it would manufacture false staleness.
- *   - The line count matches the LSP's own addressing, not `wc -l`: a
- *     document with N newline characters has N+1 ADDRESSABLE lines (the
- *     line after the final `\n`, or the single line of an empty document,
- *     is a real position a server can anchor a diagnostic to — e.g. TS1005
- *     "'}' expected" at EOF). `WidgetDiagnostic.line` is always
- *     `range.start.line + 1` from that same addressing, so the gate must
- *     use it too or every trailing-newline file demotes its own valid EOF
- *     diagnostics.
- *   - Cost discipline: one `statSync` per served file; the file's newlines
- *     are counted in bounded chunks via a raw fd read (never a full
- *     `readFileSync` — no whole-file string materialization, no per-render
- *     re-read of unchanged content), gated by a byte-size cap matching
- *     `captureReadContentBinding`'s precedent (`read-guard.ts`). The count is
- *     memoized in a small SHARED cache (module-scoped, bounded, evicted FIFO
- *     past a cap), keyed on BOTH mtime AND size (review round V1) — mtime
- *     ALONE is not a reliable cache key on hosts where its resolution is
- *     coarser than back-to-back writes: a truncate-then-write, a formatter
- *     write-back, or pi-lens's own auto-format immediately followed by the
- *     agent's write can land two DIFFERENT contents on the identical
- *     `mtimeMs`, and a mtime-only cache serves the first content's count for
- *     the second — measured live at 207/300 shrink/restore cycles. Requiring
- *     both to match narrows a same-tick collision to same-mtime-AND-same-size
- *     (which a shrink/restore cycle never produces, since the whole point is
- *     the size changed). A changed file is always recounted. Tests can still
- *     inject an isolated {@link LineCountCache} via {@link createLineCountCache}
- *     to avoid cross-test bleed.
- *   - On a demotion EDGE (a diagnostic that was not already flagged past-EOF
- *     and now is), best-effort trigger a document resync (didClose/didOpen or
- *     a full-sync didChange) so the desync HEALS instead of re-serving the
- *     same stale verdict next time (#1631's resync-before-revalidate rule).
- *     Resync is fire-and-forget: a failure here must never block or fail the
- *     render path that discovered the drift. Already-demoted entries and the
- *     healing direction never re-trigger a resync — only the rising edge does.
- */
-import * as fs from "node:fs";
-import { BoundedFifoMap } from "./bounded-cache.js";
-import { logLatency } from "./latency-logger.js";
-import { toProjectRelativePath } from "./path-utils.js";
-import { STALE_LINE_MARKER } from "./stale-marker.js";
-/** Marker rendered in place of a demoted diagnostic's now-untrustworthy line.
- * Shares the base `STALE_LINE_MARKER` vocabulary with sibling freshness
- * gates (#1622/#1633) and adds this gate's own specific reason as a suffix. */
-export const PAST_EOF_STALE_MARKER = `${STALE_LINE_MARKER} — line past EOF`;
-/** How many demoted cited lines a single record names before it stops. */
-const MAX_LOGGED_PAST_EOF_LINES = 3;
-/** Byte-size gate matching `captureReadContentBinding`'s precedent
- * (`clients/read-guard.ts`'s `READ_BINDING_MAX_BYTES`) — a hot render/serve
- * path must never pay to scan an unbounded file. */
-const MAX_GATE_BYTES = 4 * 1024 * 1024;
-/** Chunk size for the bounded newline scan — small enough to keep one read's
- * memory footprint negligible, large enough that a multi-MB file finishes in
- * single-digit reads rather than thousands. */
-const NEWLINE_SCAN_CHUNK_BYTES = 64 * 1024;
-const NEWLINE_BYTE = 0x0a;
-/** Test-only isolation seam. Production code uses the shared default cache. */
-export function createLineCountCache() {
-    return new Map();
-}
-// Shared across calls/render passes so N files touched once per session cost
-// N recounts total, not N per render — the actual fix for the cost this gate
-// exists to control. Still cannot go stale: every read re-stats first and a
-// mismatched mtime always recomputes (see the module doc comment). Bounded
-// via FIFO eviction (mirrors `READ_GUARD_MAX_FILES`'s precedent) so a long
-// session's file count cannot grow this without limit.
-export const MAX_SHARED_CACHE_ENTRIES = 512;
-const sharedLineCountCache = new BoundedFifoMap(MAX_SHARED_CACHE_ENTRIES);
-function rememberInSharedCache(cache, filePath, entry) {
-    // `cache` is either the shared BoundedFifoMap (bounds itself on `set`) or a
-    // test-isolated plain `Map` (unbounded by design) — either way, this is
-    // just a write.
-    cache.set(filePath, entry);
-}
-/** Test-only: drop the shared memo between test files/cases. */
-export function _resetSharedLineCountCacheForTests() {
-    sharedLineCountCache.clear();
-}
-/**
- * Test-only: seed the shared memo with a fabricated entry, deterministically
- * proving a cache HIT (and that the default cache used by callers with no
- * explicit `lineCountCache` is genuinely shared across calls) without relying
- * on a real write's mtime round-tripping through `fs.utimesSync` at full
- * precision — a platform-specific hazard on its own, unrelated to what such a
- * test is actually verifying.
- */
-export function _seedSharedLineCountCacheForTests(filePath, entry) {
-    sharedLineCountCache.set(filePath, entry);
-}
-/** #2442 test-only: membership check that bypasses the real-stat requirement
- *  `getCachedLineCount` imposes on every call — proves capacity eviction
- *  directly, without needing 513 real files on disk. */
-export function _sharedLineCountCacheHasForTests(filePath) {
-    return sharedLineCountCache.has(filePath);
-}
-/**
- * Count newline bytes in `filePath` via bounded raw reads — never a full
- * `readFileSync`/UTF-8 decode of the whole file. `0x0A` (LF) never appears as
- * a continuation byte in UTF-8, so counting it at the byte level is exact
- * regardless of encoding, and multi-byte characters spanning a chunk boundary
- * cannot produce a false match (a continuation byte's high bits can never
- * equal `0x0A`).
- */
-function countNewlinesChunked(filePath, sizeBytes) {
-    const fd = fs.openSync(filePath, "r");
-    try {
-        const buffer = Buffer.allocUnsafe(Math.min(NEWLINE_SCAN_CHUNK_BYTES, sizeBytes) || 1);
-        let newlineCount = 0;
-        let position = 0;
-        while (position < sizeBytes) {
-            const bytesRead = fs.readSync(fd, buffer, 0, buffer.length, position);
-            if (bytesRead <= 0)
-                break;
-            for (let i = 0; i < bytesRead; i++) {
-                if (buffer[i] === NEWLINE_BYTE)
-                    newlineCount++;
-            }
-            position += bytesRead;
-        }
-        return newlineCount;
-    }
-    finally {
-        fs.closeSync(fd);
-    }
-}
-/**
- * LSP-addressable line count for `filePath`: newline count + 1, matching
- * `range.start.line + 1` (a document with zero newlines still has one
- * addressable line — position 0,0 — and a trailing `\n` adds one more,
- * empty, addressable line after it). Memoized in `cache` (the shared default
- * unless a caller injects its own) and invalidated by mtime AND size (V1 —
- * mtime alone is not a reliable cache key on hosts where its resolution is
- * coarser than back-to-back writes). Returns `undefined` when the file
- * cannot be stat'ed, is zero bytes, exceeds the byte-size gate, or
- * cannot be read (deleted, unreadable, permissions, oversized) — callers MUST
- * treat `undefined` as "no verdict", not as "zero lines" (the empty-result-
- * must-distinguish-clean-from-errored screen).
- */
-export function getCachedLineCount(filePath, cache = sharedLineCountCache) {
-    let stat;
-    try {
-        stat = fs.statSync(filePath);
-    }
-    catch {
-        return undefined;
-    }
-    // A zero-byte stat can be a truncate-then-write window. It is not evidence
-    // that the file has one addressable line, so fail open like an unreadable
-    // file rather than manufacturing a verdict from the empty read.
-    if (stat.size === 0)
-        return undefined;
-    const cached = cache.get(filePath);
-    if (cached && cached.mtimeMs === stat.mtimeMs && cached.size === stat.size) {
-        return cached.lineCount;
-    }
-    if (stat.size > MAX_GATE_BYTES)
-        return undefined;
-    let newlineCount;
-    try {
-        newlineCount = countNewlinesChunked(filePath, stat.size);
-    }
-    catch {
-        return undefined;
-    }
-    const lineCount = newlineCount + 1;
-    rememberInSharedCache(cache, filePath, {
-        mtimeMs: stat.mtimeMs,
-        size: stat.size,
-        lineCount,
-    });
-    return lineCount;
-}
-/**
- * Recompute the past-EOF verdict for every diagnostic in `diagnostics`
- * against `filePath`'s CURRENT line count. Sets `stale: true` on a match and
- * `stale: false` when a previously-past-EOF line is back in bounds (a
- * transient shrink healed) — the verdict is always derived fresh, never
- * latched. Pure apart from the cached stat/read in {@link getCachedLineCount}
- * and the bounded `logLatency`/`resync` calls, which fire only on a RISING
- * edge (a diagnostic that was not already flagged and now is) so an
- * unchanged persistent demotion doesn't re-log or re-resync on every render.
- * Never throws, never mutates the input array.
- */
-export function demotePastEofDiagnostics(args) {
-    const { diagnostics } = args;
-    if (diagnostics.length === 0) {
-        return { diagnostics: [], demotedCount: 0 };
-    }
-    const lineCount = getCachedLineCount(args.filePath, args.lineCountCache ?? sharedLineCountCache);
-    if (lineCount === undefined) {
-        // Fail open: no verdict available for this file right now. Leave
-        // whatever `stale` value each entry already carries untouched — this
-        // gate has nothing to say either direction.
-        return { diagnostics: [...diagnostics], demotedCount: 0 };
-    }
-    const risingEdgeLines = [];
-    const out = diagnostics.map((d) => {
-        // Another gate's demotion is not this gate's to heal (#1631 drift
-        // entries have in-bounds lines; re-deriving here would clear them).
-        if (d.stale && d.staleReason !== undefined && d.staleReason !== "past-eof")
-            return d;
-        const isPastEof = typeof d.line === "number" && d.line > lineCount;
-        if (isPastEof === !!d.stale)
-            return d; // no transition either direction
-        if (isPastEof)
-            risingEdgeLines.push(d.line);
-        return {
-            ...d,
-            stale: isPastEof,
-            staleReason: isPastEof ? "past-eof" : undefined,
-        };
-    });
-    if (risingEdgeLines.length === 0) {
-        return { diagnostics: out, demotedCount: 0 };
-    }
-    let sizeBytes;
-    try {
-        sizeBytes = fs.statSync(args.filePath).size;
-    }
-    catch {
-        sizeBytes = 0;
-    }
-    logLatency({
-        type: "phase",
-        phase: "diagnostic_past_eof",
-        filePath: args.cwd,
-        durationMs: 0,
-        metadata: {
-            store: args.store,
-            file: toProjectRelativePath(args.filePath, args.cwd),
-            sizeBytes,
-            actualLineCount: lineCount,
-            demotedCount: risingEdgeLines.length,
-            sampleCitedLines: risingEdgeLines.slice(0, MAX_LOGGED_PAST_EOF_LINES),
-        },
-    });
-    if (args.resync) {
-        try {
-            args.resync(args.filePath);
-        }
-        catch {
-            // Best-effort — a resync failure must never fail the render path.
-        }
-    }
-    return { diagnostics: out, demotedCount: risingEdgeLines.length };
-}
-/**
- * Default resync trigger (#1641 criterion 2): re-open the file with its
- * CURRENT disk content so the LSP's in-memory document is replaced rather
- * than left to re-publish the same divergent diagnostics next time
- * (`LSPService.openFile` sends a full-sync didChange, or a didClose/didOpen
- * pair for servers that only re-scan on a fresh open — see
- * `handleNotifyOpen`'s `reopenOnResync` branch). Dynamic import (mirrors
- * `clients/module-report-lsp.ts`) so this module — used from widget-state.ts
- * and tools/lens-diagnostics.ts — never statically depends on `clients/lsp`,
- * which itself imports widget-state.ts.
- *
- * Fire-and-forget by design: the caller is a synchronous render path and must
- * not block on (or fail because of) a resync. No LSP service, no client for
- * this file's language, or a transient read/spawn failure all degrade to a
- * silent no-op — the next successful touch/dispatch resyncs anyway.
- */
-export function resyncDocumentOnPastEof(filePath) {
-    void (async () => {
-        try {
-            const [{ getLSPService }, content] = await Promise.all([
-                import("./lsp/index.js"),
-                fs.promises.readFile(filePath, "utf-8"),
-            ]);
-            await getLSPService().openFile(filePath, content, {
-                preserveDiagnostics: false,
-            });
-        }
-        catch {
-            // Best-effort heal only.
-        }
-    })();
-}

package/node_modules/pi-lens/dist/clients/diagnostic-logger.js

@@ -1,73 +0,0 @@
-/**
- * Diagnostic Logger — append-only JSONL log for cross-session analytics
- *
- * Log file: ~/.pi-lens/logs/{date}.jsonl
- */
-import * as path from "node:path";
-import { isTestMode } from "./env-utils.js";
-import { getGlobalPiLensLogDir } from "./probe-home-state.js";
-import { getMaxLogSizeMB } from "./log-cleanup.js";
-import { createNdjsonLogger } from "./ndjson-logger.js";
-function getLogDir() {
-    return path.join(getGlobalPiLensLogDir(), "logs");
-}
-function getLogFile() {
-    const date = new Date().toISOString().split("T")[0];
-    return path.join(getLogDir(), `${date}.jsonl`);
-}
-// Module-level singleton — persists across all writes
-let _logger = null;
-export function getDiagnosticLogger() {
-    if (!_logger) {
-        _logger = createDiagnosticLogger();
-    }
-    return _logger;
-}
-function createDiagnosticLogger() {
-    // Lazy filePath: the log file is keyed on the current date, resolved per
-    // drain so a long-lived logger rolls over at midnight. A daily rollover
-    // is NOT a size bound (#2505): one very busy day, or a long-lived
-    // process that never re-runs the session-start sweep, grows a single
-    // day file without limit. The backup keeps the .jsonl extension so the
-    // retention sweep in log-cleanup.ts (which matches /\.jsonl$/ under
-    // logs/) reaps it like any other daily file.
-    const writer = createNdjsonLogger({
-        filePath: () => getLogFile(),
-        maxBytes: getMaxLogSizeMB() * 1024 * 1024,
-        backupPath: () => getLogFile().replace(/\.jsonl$/, ".rotated.jsonl"),
-    });
-    return {
-        log(entry) {
-            if (isTestMode()) {
-                return;
-            }
-            writer.log(entry); // async, non-blocking
-        },
-        logCaught(d, context, shownInline = false) {
-            this.log({
-                timestamp: new Date().toISOString(),
-                tool: d.tool || "unknown",
-                ruleId: d.rule || d.id || "unknown",
-                severity: d.severity || "warning",
-                language: d.language || "unknown",
-                filePath: d.filePath,
-                line: d.line || 1,
-                column: d.column || 1,
-                message: d.message || "",
-                caughtByPipeline: true,
-                shownInline,
-                autoFixed: false,
-                shownToAgent: shownInline,
-                agentFixed: false,
-                unresolved: true,
-                model: context.model,
-                sessionId: context.sessionId,
-                turnIndex: context.turnIndex,
-                writeIndex: context.writeIndex,
-            });
-        },
-        async flush() {
-            await writer.flush();
-        },
-    };
-}

package/node_modules/pi-lens/dist/clients/diagnostic-tracker.js

@@ -1,105 +0,0 @@
-/**
- * Diagnostic Tracker — in-memory tracking for session-level feedback
- *
- * Links diagnostics to resolutions, tracks violation patterns.
- */
-// Module-level singleton — persists across all writes
-let _tracker = null;
-export function getDiagnosticTracker() {
-    if (!_tracker) {
-        _tracker = createDiagnosticTracker();
-    }
-    return _tracker;
-}
-export function createDiagnosticTracker() {
-    const shown = new Map();
-    const occurrenceCounts = new Map();
-    let totalShown = 0;
-    let totalAutoFixed = 0;
-    let totalAgentFixed = 0;
-    const key = (filePath, ruleId, line) => `${filePath}:${ruleId}:${line}`;
-    return {
-        trackShown(diagnostics) {
-            for (const d of diagnostics) {
-                const ruleId = d.rule || d.id || "unknown";
-                const line = d.line || 1;
-                const k = key(d.filePath, ruleId, line);
-                occurrenceCounts.set(k, (occurrenceCounts.get(k) ?? 0) + 1);
-                // Don't double-count if already tracked
-                if (!shown.has(k)) {
-                    shown.set(k, {
-                        ruleId,
-                        filePath: d.filePath,
-                        line,
-                        shownAt: new Date(),
-                        autoFixed: false,
-                        agentFixed: false,
-                    });
-                    totalShown++;
-                }
-            }
-        },
-        trackAutoFixed(count) {
-            if (count > 0) {
-                totalAutoFixed += count;
-            }
-        },
-        trackAgentFixed(count) {
-            if (count > 0) {
-                totalAgentFixed += count;
-            }
-        },
-        getStats() {
-            const ruleCounts = new Map();
-            const rulePaths = new Map();
-            for (const entry of shown.values()) {
-                ruleCounts.set(entry.ruleId, (ruleCounts.get(entry.ruleId) || 0) + 1);
-                if (!rulePaths.has(entry.ruleId))
-                    rulePaths.set(entry.ruleId, new Set());
-                rulePaths.get(entry.ruleId)?.add(entry.filePath);
-            }
-            const topViolations = [...ruleCounts.entries()]
-                .sort((a, b) => b[1] - a[1])
-                .slice(0, 10)
-                .map(([ruleId, count]) => ({
-                ruleId,
-                count,
-                samplePaths: [...(rulePaths.get(ruleId) ?? new Set())]
-                    .sort((a, b) => a.localeCompare(b))
-                    .slice(0, 3),
-            }));
-            const repeatOffenders = [...occurrenceCounts.entries()]
-                .filter(([, count]) => count >= 2)
-                .sort((a, b) => b[1] - a[1])
-                .slice(0, 10)
-                .map(([k, count]) => {
-                const parts = k.split(":");
-                const lineStr = parts.pop() ?? "1";
-                const ruleId = parts.pop() ?? "unknown";
-                const filePath = parts.join(":");
-                return {
-                    key: k,
-                    ruleId,
-                    filePath,
-                    line: Number.parseInt(lineStr, 10) || 1,
-                    count,
-                };
-            });
-            return {
-                totalShown,
-                totalAutoFixed,
-                totalAgentFixed,
-                totalUnresolved: totalShown - totalAutoFixed - totalAgentFixed,
-                topViolations,
-                repeatOffenders,
-            };
-        },
-        reset() {
-            shown.clear();
-            occurrenceCounts.clear();
-            totalShown = 0;
-            totalAutoFixed = 0;
-            totalAgentFixed = 0;
-        },
-    };
-}