import { tool, type Tool } from "@lmstudio/sdk";
import { existsSync } from "fs";
import { readdir, readFile, stat, writeFile } from "fs/promises";
import { extname, join } from "path";
import { z } from "zod";
import { snapshot } from "../backup";
import { validateSyntax } from "../validate";
import { clamp, formatBytes, IGNORED_DIRS, type Workspace } from "../workspace";

/**
 * Tools that trade many round-trips for one. Every tool call costs a full
 * prediction on a local model, so reading six files one at a time is six
 * generations of latency; doing it in one call is one.
 */

const MAX_FILES_READ = 12;
const MAX_TOTAL_CHARS = 40000;
const MAX_FILES_CHANGED = 60;
const MAX_RECENT_LISTED = 40;

const SKIP_EXTENSIONS = new Set([
  ".png", ".jpg", ".jpeg", ".gif", ".webp", ".ico", ".pdf", ".zip", ".gz", ".exe",
  ".dll", ".so", ".dylib", ".mp3", ".wav", ".mp4", ".bin", ".3dsx", ".cia", ".woff", ".woff2",
]);

export function bulkTools(ws: Workspace): Tool[] {
  const tools: Tool[] = [
    tool({
      name: "read_many_files",
      description:
        "Read several files at once and get them back together. Use this instead of calling " +
        "read_file repeatedly -- it is much faster. Good for reading a module and its tests, or " +
        "every file you are about to change. Each file is truncated if long.",
      parameters: {
        paths: z.array(z.string()).min(1).describe("File paths relative to the workspace root."),
        max_lines_each: z
          .number()
          .int()
          .min(1)
          .default(200)
          .describe("Lines to show per file before truncating."),
      },
      implementation: async ({ paths, max_lines_each }, ctx) => {
        const wanted = paths.slice(0, MAX_FILES_READ);
        ctx.status(`Reading ${wanted.length} file(s)`);
        const sections: string[] = [];
        let budget = MAX_TOTAL_CHARS;

        for (const candidate of wanted) {
          let absPath: string;
          try {
            absPath = ws.resolveInRoot(candidate);
          } catch (error) {
            sections.push(`===== ${candidate} =====\n${(error as Error).message}`);
            continue;
          }
          if (!existsSync(absPath)) {
            sections.push(`===== ${candidate} =====\nDoes not exist.`);
            continue;
          }
          const info = await stat(absPath);
          if (info.isDirectory()) {
            sections.push(`===== ${candidate} =====\nThis is a directory, not a file.`);
            continue;
          }
          if (budget <= 0) {
            sections.push(`===== ${candidate} =====\nSkipped: combined output limit reached.`);
            continue;
          }
          try {
            const lines = (await readFile(absPath, "utf-8")).split(/\r?\n/);
            const shown = lines.slice(0, max_lines_each);
            const width = String(shown.length).length;
            const body = shown
              .map((line, i) => `${String(i + 1).padStart(width, " ")}\t${line}`)
              .join("\n");
            const more =
              lines.length > shown.length
                ? `\n[${lines.length - shown.length} more line(s); use read_file with start_line=${shown.length + 1}]`
                : "";
            const section = `===== ${ws.rel(absPath)} (${lines.length} lines, ${formatBytes(info.size)}) =====\n${body}${more}`;
            budget -= section.length;
            sections.push(section);
          } catch {
            sections.push(`===== ${candidate} =====\nCould not be read as text.`);
          }
        }

        const skipped =
          paths.length > wanted.length
            ? `\n\n[${paths.length - wanted.length} path(s) not read: at most ${MAX_FILES_READ} per call]`
            : "";
        return clamp(sections.join("\n\n") + skipped, MAX_TOTAL_CHARS + 2000, "combined files");
      },
    }),
  ];

  tools.push(
    tool({
      name: "recent_changes",
      description:
        "List files in the workspace modified recently, whoever changed them. Use this when " +
        "resuming work, or when the user says they edited something: changed_files only knows " +
        "about edits YOU made, so this is the only way to notice theirs.",
      parameters: {
        minutes: z.number().int().min(1).default(30).describe("How far back to look, in minutes."),
      },
      implementation: async ({ minutes }, ctx) => {
        ctx.status(`Looking for files changed in the last ${minutes} min`);
        const now = Date.now();
        const cutoff = now - minutes * 60_000;
        const found: { display: string; ageMin: number; size: number }[] = [];
        await walk(ws, ws.root, /^.*$/, async (absPath) => {
          try {
            const info = await stat(absPath);
            if (info.mtimeMs < cutoff) return;
            found.push({
              display: ws.rel(absPath),
              ageMin: Math.max(0, Math.round((now - info.mtimeMs) / 60_000)),
              size: info.size,
            });
          } catch {
            // Vanished between listing and stat; ignore.
          }
        });
        if (found.length === 0) {
          return `No files in the workspace were modified in the last ${minutes} minute(s).`;
        }
        found.sort((a, b) => a.ageMin - b.ageMin);
        const shown = found.slice(0, MAX_RECENT_LISTED);
        const lines = shown.map(
          (item) =>
            `  ${item.display} -- ${item.ageMin === 0 ? "just now" : `${item.ageMin} min ago`} (${formatBytes(item.size)})`,
        );
        const more =
          found.length > shown.length ? `\n  ...and ${found.length - shown.length} more` : "";
        return (
          `${found.length} file(s) modified in the last ${minutes} minute(s), newest first:\n` +
          `${lines.join("\n")}${more}\n\n` +
          `Read any you did not write yourself before assuming you know what they contain.`
        );
      },
    }),
  );

  if (ws.allowWrite) {
    tools.push(
      tool({
        name: "replace_in_files",
        description:
          "Replace an exact piece of text everywhere it appears across many files. Use this for " +
          "renames and other sweeping changes, so no call site is missed. ALWAYS run it once with " +
          "dry_run true to see what would change, then again with dry_run false to apply it.",
        parameters: {
          find: z.string().default("").describe("Exact text to find. Not a regular expression."),
          replace: z.string().default("").describe("Text to put in its place."),
          file_glob: z
            .string()
            .default("**/*")
            .describe("Which files to touch, e.g. '**/*.ts'. Defaults to every text file."),
          dry_run: z
            .boolean()
            .default(true)
            .describe("True reports what would change without writing. Run this way first."),
        },
        implementation: async ({ find, replace, file_glob, dry_run }, ctx) => {
          if (find.trim() === "") return "Error: find is empty. Pass the exact text to replace.";
          if (find === replace) {
            return "Error: find and replace are identical, so nothing would change.";
          }
          ctx.status(dry_run ? `Previewing replacement of ${find}` : `Replacing ${find}`);

          const matcher = globToRegExp(file_glob);
          const hits: { absPath: string; display: string; count: number; content: string }[] = [];
          await walk(ws, ws.root, matcher, async (absPath, content) => {
            if (hits.length >= MAX_FILES_CHANGED) return;
            const count = countOccurrences(content, find);
            if (count > 0) hits.push({ absPath, display: ws.rel(absPath), count, content });
          });

          if (hits.length === 0) {
            return `No file matching "${file_glob}" contains "${find}". Check the spelling, or widen file_glob.`;
          }

          const total = hits.reduce((sum, hit) => sum + hit.count, 0);
          const listing = hits
            .map((hit) => `  ${hit.display}: ${hit.count} occurrence${hit.count === 1 ? "" : "s"}`)
            .join("\n");

          if (dry_run) {
            return (
              `Dry run -- nothing was changed.\n\n${total} occurrence(s) of "${find}" in ` +
              `${hits.length} file(s):\n${listing}\n\nRun again with dry_run false to apply.`
            );
          }

          const complaints: string[] = [];
          for (const hit of hits) {
            await snapshot(ws, hit.absPath);
            await writeFile(hit.absPath, hit.content.split(find).join(replace), "utf-8");
            const complaint = await validateSyntax(ws, hit.absPath);
            if (complaint !== "") complaints.push(complaint.trim());
          }
          return (
            `Replaced ${total} occurrence(s) of "${find}" with "${replace}" across ${hits.length} file(s):\n` +
            `${listing}\n\nEvery file was backed up first. Run verify now.` +
            `${complaints.length > 0 ? `\n\n${complaints.join("\n")}` : ""}`
          );
        },
      }),
    );
  }

  return tools;
}

function countOccurrences(haystack: string, needle: string): number {
  if (needle === "") return 0;
  return haystack.split(needle).length - 1;
}

/** Same translation the search tools use: **, *, ? and {a,b}. */
function globToRegExp(pattern: string): RegExp {
  let out = "";
  for (let i = 0; i < pattern.length; i++) {
    const char = pattern[i];
    if (char === "*") {
      if (pattern[i + 1] === "*") {
        out += ".*";
        i++;
        if (pattern[i + 1] === "/") i++;
      } else {
        out += "[^/]*";
      }
    } else if (char === "?") {
      out += "[^/]";
    } else if (char === "{") {
      out += "(";
    } else if (char === "}") {
      out += ")";
    } else if (char === ",") {
      out += "|";
    } else if ("\\^$+.()|[]".includes(char)) {
      out += `\\${char}`;
    } else {
      out += char;
    }
  }
  return new RegExp(`^${out}$`, "i");
}

async function walk(
  ws: Workspace,
  dir: string,
  matcher: RegExp,
  onFile: (absPath: string, content: string) => Promise<void>,
  depth = 0,
): Promise<void> {
  if (depth > 12) return;
  let entries;
  try {
    entries = await readdir(dir, { withFileTypes: true });
  } catch {
    return;
  }
  for (const entry of entries) {
    const full = join(dir, entry.name);
    if (entry.isDirectory()) {
      if (IGNORED_DIRS.has(entry.name) || entry.name.startsWith(".")) continue;
      await walk(ws, full, matcher, onFile, depth + 1);
      continue;
    }
    if (SKIP_EXTENSIONS.has(extname(entry.name).toLowerCase())) continue;
    const rel = ws.rel(full).replace(/\\/g, "/");
    if (!matcher.test(rel) && !matcher.test(entry.name)) continue;
    let info;
    try {
      info = await stat(full);
    } catch {
      continue;
    }
    if (info.size > ws.maxBytes) continue;
    try {
      await onFile(full, await readFile(full, "utf-8"));
    } catch {
      // Binary or unreadable; skip.
    }
  }
}

import { tool, type Tool } from "@lmstudio/sdk";
import { existsSync } from "fs";
import { readdir, readFile, stat, writeFile } from "fs/promises";
import { extname, join } from "path";
import { z } from "zod";
import { snapshot } from "../backup";
import { validateSyntax } from "../validate";
import { clamp, formatBytes, IGNORED_DIRS, type Workspace } from "../workspace";

/**
 * Tools that trade many round-trips for one. Every tool call costs a full
 * prediction on a local model, so reading six files one at a time is six
 * generations of latency; doing it in one call is one.
 */

const MAX_FILES_READ = 12;
const MAX_TOTAL_CHARS = 40000;
const MAX_FILES_CHANGED = 60;
const MAX_RECENT_LISTED = 40;

const SKIP_EXTENSIONS = new Set([
  ".png", ".jpg", ".jpeg", ".gif", ".webp", ".ico", ".pdf", ".zip", ".gz", ".exe",
  ".dll", ".so", ".dylib", ".mp3", ".wav", ".mp4", ".bin", ".3dsx", ".cia", ".woff", ".woff2",
]);

export function bulkTools(ws: Workspace): Tool[] {
  const tools: Tool[] = [
    tool({
      name: "read_many_files",
      description:
        "Read several files at once and get them back together. Use this instead of calling " +
        "read_file repeatedly -- it is much faster. Good for reading a module and its tests, or " +
        "every file you are about to change. Each file is truncated if long.",
      parameters: {
        paths: z.array(z.string()).min(1).describe("File paths relative to the workspace root."),
        max_lines_each: z
          .number()
          .int()
          .min(1)
          .default(200)
          .describe("Lines to show per file before truncating."),
      },
      implementation: async ({ paths, max_lines_each }, ctx) => {
        const wanted = paths.slice(0, MAX_FILES_READ);
        ctx.status(`Reading ${wanted.length} file(s)`);
        const sections: string[] = [];
        let budget = MAX_TOTAL_CHARS;

        for (const candidate of wanted) {
          let absPath: string;
          try {
            absPath = ws.resolveInRoot(candidate);
          } catch (error) {
            sections.push(`===== ${candidate} =====\n${(error as Error).message}`);
            continue;
          }
          if (!existsSync(absPath)) {
            sections.push(`===== ${candidate} =====\nDoes not exist.`);
            continue;
          }
          const info = await stat(absPath);
          if (info.isDirectory()) {
            sections.push(`===== ${candidate} =====\nThis is a directory, not a file.`);
            continue;
          }
          if (budget <= 0) {
            sections.push(`===== ${candidate} =====\nSkipped: combined output limit reached.`);
            continue;
          }
          try {
            const lines = (await readFile(absPath, "utf-8")).split(/\r?\n/);
            const shown = lines.slice(0, max_lines_each);
            const width = String(shown.length).length;
            const body = shown
              .map((line, i) => `${String(i + 1).padStart(width, " ")}\t${line}`)
              .join("\n");
            const more =
              lines.length > shown.length
                ? `\n[${lines.length - shown.length} more line(s); use read_file with start_line=${shown.length + 1}]`
                : "";
            const section = `===== ${ws.rel(absPath)} (${lines.length} lines, ${formatBytes(info.size)}) =====\n${body}${more}`;
            budget -= section.length;
            sections.push(section);
          } catch {
            sections.push(`===== ${candidate} =====\nCould not be read as text.`);
          }
        }

        const skipped =
          paths.length > wanted.length
            ? `\n\n[${paths.length - wanted.length} path(s) not read: at most ${MAX_FILES_READ} per call]`
            : "";
        return clamp(sections.join("\n\n") + skipped, MAX_TOTAL_CHARS + 2000, "combined files");
      },
    }),
  ];

  tools.push(
    tool({
      name: "recent_changes",
      description:
        "List files in the workspace modified recently, whoever changed them. Use this when " +
        "resuming work, or when the user says they edited something: changed_files only knows " +
        "about edits YOU made, so this is the only way to notice theirs.",
      parameters: {
        minutes: z.number().int().min(1).default(30).describe("How far back to look, in minutes."),
      },
      implementation: async ({ minutes }, ctx) => {
        ctx.status(`Looking for files changed in the last ${minutes} min`);
        const now = Date.now();
        const cutoff = now - minutes * 60_000;
        const found: { display: string; ageMin: number; size: number }[] = [];
        await walk(ws, ws.root, /^.*$/, async (absPath) => {
          try {
            const info = await stat(absPath);
            if (info.mtimeMs < cutoff) return;
            found.push({
              display: ws.rel(absPath),
              ageMin: Math.max(0, Math.round((now - info.mtimeMs) / 60_000)),
              size: info.size,
            });
          } catch {
            // Vanished between listing and stat; ignore.
          }
        });
        if (found.length === 0) {
          return `No files in the workspace were modified in the last ${minutes} minute(s).`;
        }
        found.sort((a, b) => a.ageMin - b.ageMin);
        const shown = found.slice(0, MAX_RECENT_LISTED);
        const lines = shown.map(
          (item) =>
            `  ${item.display} -- ${item.ageMin === 0 ? "just now" : `${item.ageMin} min ago`} (${formatBytes(item.size)})`,
        );
        const more =
          found.length > shown.length ? `\n  ...and ${found.length - shown.length} more` : "";
        return (
          `${found.length} file(s) modified in the last ${minutes} minute(s), newest first:\n` +
          `${lines.join("\n")}${more}\n\n` +
          `Read any you did not write yourself before assuming you know what they contain.`
        );
      },
    }),
  );

  if (ws.allowWrite) {
    tools.push(
      tool({
        name: "replace_in_files",
        description:
          "Replace an exact piece of text everywhere it appears across many files. Use this for " +
          "renames and other sweeping changes, so no call site is missed. ALWAYS run it once with " +
          "dry_run true to see what would change, then again with dry_run false to apply it.",
        parameters: {
          find: z.string().default("").describe("Exact text to find. Not a regular expression."),
          replace: z.string().default("").describe("Text to put in its place."),
          file_glob: z
            .string()
            .default("**/*")
            .describe("Which files to touch, e.g. '**/*.ts'. Defaults to every text file."),
          dry_run: z
            .boolean()
            .default(true)
            .describe("True reports what would change without writing. Run this way first."),
        },
        implementation: async ({ find, replace, file_glob, dry_run }, ctx) => {
          if (find.trim() === "") return "Error: find is empty. Pass the exact text to replace.";
          if (find === replace) {
            return "Error: find and replace are identical, so nothing would change.";
          }
          ctx.status(dry_run ? `Previewing replacement of ${find}` : `Replacing ${find}`);

          const matcher = globToRegExp(file_glob);
          const hits: { absPath: string; display: string; count: number; content: string }[] = [];
          await walk(ws, ws.root, matcher, async (absPath, content) => {
            if (hits.length >= MAX_FILES_CHANGED) return;
            const count = countOccurrences(content, find);
            if (count > 0) hits.push({ absPath, display: ws.rel(absPath), count, content });
          });

          if (hits.length === 0) {
            return `No file matching "${file_glob}" contains "${find}". Check the spelling, or widen file_glob.`;
          }

          const total = hits.reduce((sum, hit) => sum + hit.count, 0);
          const listing = hits
            .map((hit) => `  ${hit.display}: ${hit.count} occurrence${hit.count === 1 ? "" : "s"}`)
            .join("\n");

          if (dry_run) {
            return (
              `Dry run -- nothing was changed.\n\n${total} occurrence(s) of "${find}" in ` +
              `${hits.length} file(s):\n${listing}\n\nRun again with dry_run false to apply.`
            );
          }

          const complaints: string[] = [];
          for (const hit of hits) {
            await snapshot(ws, hit.absPath);
            await writeFile(hit.absPath, hit.content.split(find).join(replace), "utf-8");
            const complaint = await validateSyntax(ws, hit.absPath);
            if (complaint !== "") complaints.push(complaint.trim());
          }
          return (
            `Replaced ${total} occurrence(s) of "${find}" with "${replace}" across ${hits.length} file(s):\n` +
            `${listing}\n\nEvery file was backed up first. Run verify now.` +
            `${complaints.length > 0 ? `\n\n${complaints.join("\n")}` : ""}`
          );
        },
      }),
    );
  }

  return tools;
}

function countOccurrences(haystack: string, needle: string): number {
  if (needle === "") return 0;
  return haystack.split(needle).length - 1;
}

/** Same translation the search tools use: **, *, ? and {a,b}. */
function globToRegExp(pattern: string): RegExp {
  let out = "";
  for (let i = 0; i < pattern.length; i++) {
    const char = pattern[i];
    if (char === "*") {
      if (pattern[i + 1] === "*") {
        out += ".*";
        i++;
        if (pattern[i + 1] === "/") i++;
      } else {
        out += "[^/]*";
      }
    } else if (char === "?") {
      out += "[^/]";
    } else if (char === "{") {
      out += "(";
    } else if (char === "}") {
      out += ")";
    } else if (char === ",") {
      out += "|";
    } else if ("\\^$+.()|[]".includes(char)) {
      out += `\\${char}`;
    } else {
      out += char;
    }
  }
  return new RegExp(`^${out}$`, "i");
}

async function walk(
  ws: Workspace,
  dir: string,
  matcher: RegExp,
  onFile: (absPath: string, content: string) => Promise<void>,
  depth = 0,
): Promise<void> {
  if (depth > 12) return;
  let entries;
  try {
    entries = await readdir(dir, { withFileTypes: true });
  } catch {
    return;
  }
  for (const entry of entries) {
    const full = join(dir, entry.name);
    if (entry.isDirectory()) {
      if (IGNORED_DIRS.has(entry.name) || entry.name.startsWith(".")) continue;
      await walk(ws, full, matcher, onFile, depth + 1);
      continue;
    }
    if (SKIP_EXTENSIONS.has(extname(entry.name).toLowerCase())) continue;
    const rel = ws.rel(full).replace(/\\/g, "/");
    if (!matcher.test(rel) && !matcher.test(entry.name)) continue;
    let info;
    try {
      info = await stat(full);
    } catch {
      continue;
    }
    if (info.size > ws.maxBytes) continue;
    try {
      await onFile(full, await readFile(full, "utf-8"));
    } catch {
      // Binary or unreadable; skip.
    }
  }
}