import { readFile, stat } from "node:fs/promises";
import { AgenticError } from "../core/errors";
import { sha256Text } from "../core/hash";

export interface TextFileSnapshot {
  content: string;
  bytes: number;
  sha256: string;
  lineCount: number;
  newline: "lf" | "crlf" | "mixed" | "none";
  hasFinalNewline: boolean;
}

export function appearsBinary(buffer: Buffer): boolean {
  const sample = buffer.subarray(0, Math.min(buffer.length, 8192));
  if (sample.includes(0)) return true;
  let suspicious = 0;
  for (const byte of sample) {
    if (byte < 7 || (byte > 13 && byte < 32)) suspicious++;
  }
  return sample.length > 0 && suspicious / sample.length > 0.12;
}

/**
 * String.split(/\r?\n/) produces a trailing "" element whenever content ends
 * in a newline (e.g. "a\nb\n" -> ["a","b",""]), which would otherwise inflate
 * the reported line count by one. Drop that phantom trailing empty element so
 * "a\nb\n" and "a\nb" both count as 2 lines; "" still counts as 0 (handled by
 * callers before this function is reached).
 */
export function countLines(lines: string[]): number {
  if (lines.length > 1 && lines[lines.length - 1] === "") {
    return lines.length - 1;
  }
  return lines.length;
}

export function detectNewline(content: string): TextFileSnapshot["newline"] {
  const crlf = (content.match(/\r\n/g) ?? []).length;
  const lf = (content.match(/(?<!\r)\n/g) ?? []).length;
  if (crlf === 0 && lf === 0) return "none";
  if (crlf > 0 && lf > 0) return "mixed";
  return crlf > 0 ? "crlf" : "lf";
}

export async function readTextSnapshot(
  path: string,
  maxBytes: number,
): Promise<TextFileSnapshot> {
  const info = await stat(path);
  if (!info.isFile()) {
    throw new AgenticError("INVALID_INPUT", "Expected a regular file.", { path });
  }
  if (info.size > maxBytes) {
    throw new AgenticError(
      "FILE_TOO_LARGE",
      `File is ${info.size.toLocaleString()} bytes; limit is ${maxBytes.toLocaleString()} bytes.`,
      { path, bytes: info.size, maxBytes },
    );
  }
  const buffer = await readFile(path);
  if (buffer.length > maxBytes) {
    throw new AgenticError(
      "FILE_TOO_LARGE",
      `File grew to ${buffer.length.toLocaleString()} bytes while being read; limit is ${maxBytes.toLocaleString()} bytes.`,
      { path, bytes: buffer.length, maxBytes },
    );
  }
  if (appearsBinary(buffer)) {
    throw new AgenticError("BINARY_FILE", "Binary files are not editable as text.", {
      path,
    });
  }
  const content = buffer.toString("utf8");
  if (!Buffer.from(content, "utf8").equals(buffer)) {
    throw new AgenticError(
      "BINARY_FILE",
      "File is not valid UTF-8 text and cannot be edited safely.",
      { path },
    );
  }
  return snapshotFromContent(content);
}

export function snapshotFromContent(content: string): TextFileSnapshot {
  return {
    content,
    bytes: Buffer.byteLength(content),
    sha256: sha256Text(content),
    lineCount: content.length === 0 ? 0 : countLines(content.split(/\r?\n/)),
    newline: detectNewline(content),
    hasFinalNewline: /\r?\n$/.test(content),
  };
}

/**
 * Matches one line of `numberedRange` output: optional leading padding, the
 * captured line number, " | ", then the source line (which may be empty, so the
 * separator is allowed to end the line).
 */
const NUMBERED_LINE = /^[ \t]*(\d+)[ \t]*\|([ \t]|$)/;

/**
 * True when `content` looks like it was pasted back out of `numberedRange`
 * (i.e. out of a `workspace_inspect read` result) instead of being real file
 * text. Small models routinely copy the `12 | ` gutter into edit content, which
 * silently corrupts the file or makes an exact replace miss.
 *
 * Three conditions, all required. The decisive one is the **run**: read output
 * is by construction `n, n+1, n+2`, so the matched prefixes must be adjacent
 * lines whose numbers ascend by exactly one, at least twice in a row. Data that
 * happens to use pipes almost never does that — measured over 36 inputs, adding
 * the run requirement (and only then dropping the count threshold from three
 * matched lines to two) moved false positives from 12 to 8 and missed
 * detections from 6 to 3.
 *
 * What it still does *not* catch, deliberately: a genuinely consecutive numeric
 * first column — `1 | a`, `2 | b`, `3 | c`, or consecutive years — is
 * indistinguishable from read output, and `allow_line_numbers: true` is the
 * escape hatch for it. What it reliably leaves alone is a leading-pipe markdown
 * table (no line starts with a bare number) and any non-monotonic numeric
 * column: descending, sparse, or stepped. A single numbered line is below any
 * threshold a majority rule can safely use and is never flagged; the zero-match
 * hint on `replace` covers that case after the fact instead.
 */
export function looksLineNumbered(content: string): boolean {
  const lines = content.split(/\r?\n/).filter((line) => line.trim() !== "");
  if (lines.length < 2) return false;
  let numbered = 0;
  let longestRun = 0;
  let run = 0;
  let previous: number | undefined;
  for (const line of lines) {
    const match = NUMBERED_LINE.exec(line);
    if (!match) {
      run = 0;
      previous = undefined;
      continue;
    }
    numbered++;
    const value = Number(match[1]);
    run = previous !== undefined && value === previous + 1 ? run + 1 : 1;
    if (run > longestRun) longestRun = run;
    previous = value;
  }
  return numbered >= 2 && longestRun >= 2 && numbered * 2 > lines.length;
}

/**
 * Removes the `12 | ` gutter `numberedRange` adds, from every line. Used only
 * to explain a failure after the fact (see `workspace/transactions.ts`), never
 * to rewrite what a caller asked for: silently repairing an edit would let a
 * model commit text it never actually looked at.
 *
 * The trailing separator class matches `NUMBERED_LINE`'s own tolerance: with
 * ` ?` alone, a tab-separated `1\t|\ta` was recognised as numbered but stripped
 * to `\ta`, so the sentence that quotes the stripped text was wrong by a tab.
 */
export function stripLineNumberGutter(content: string): string {
  return content.replace(/^[ \t]*\d+[ \t]*\|[ \t]?/gm, "");
}

export function numberedRange(
  content: string,
  startLine: number,
  endLine: number,
): { text: string; totalLines: number; startLine: number; endLine: number } {
  const lines = content.split(/\r?\n/);
  const totalLines = content.length === 0 ? 0 : countLines(lines);
  if (totalLines === 0) {
    return { text: "", totalLines: 0, startLine: 0, endLine: 0 };
  }
  const start = Math.max(1, Math.min(startLine, totalLines));
  const end = Math.max(start, Math.min(endLine, totalLines));
  const width = String(end).length;
  const selected = lines
    .slice(start - 1, end)
    .map((line, index) => `${String(start + index).padStart(width, " ")} | ${line}`)
    .join("\n");
  return { text: selected, totalLines, startLine: start, endLine: end };
}

import { readFile, stat } from "node:fs/promises";
import { AgenticError } from "../core/errors";
import { sha256Text } from "../core/hash";

export interface TextFileSnapshot {
  content: string;
  bytes: number;
  sha256: string;
  lineCount: number;
  newline: "lf" | "crlf" | "mixed" | "none";
  hasFinalNewline: boolean;
}

export function appearsBinary(buffer: Buffer): boolean {
  const sample = buffer.subarray(0, Math.min(buffer.length, 8192));
  if (sample.includes(0)) return true;
  let suspicious = 0;
  for (const byte of sample) {
    if (byte < 7 || (byte > 13 && byte < 32)) suspicious++;
  }
  return sample.length > 0 && suspicious / sample.length > 0.12;
}

/**
 * String.split(/\r?\n/) produces a trailing "" element whenever content ends
 * in a newline (e.g. "a\nb\n" -> ["a","b",""]), which would otherwise inflate
 * the reported line count by one. Drop that phantom trailing empty element so
 * "a\nb\n" and "a\nb" both count as 2 lines; "" still counts as 0 (handled by
 * callers before this function is reached).
 */
export function countLines(lines: string[]): number {
  if (lines.length > 1 && lines[lines.length - 1] === "") {
    return lines.length - 1;
  }
  return lines.length;
}

export function detectNewline(content: string): TextFileSnapshot["newline"] {
  const crlf = (content.match(/\r\n/g) ?? []).length;
  const lf = (content.match(/(?<!\r)\n/g) ?? []).length;
  if (crlf === 0 && lf === 0) return "none";
  if (crlf > 0 && lf > 0) return "mixed";
  return crlf > 0 ? "crlf" : "lf";
}

export async function readTextSnapshot(
  path: string,
  maxBytes: number,
): Promise<TextFileSnapshot> {
  const info = await stat(path);
  if (!info.isFile()) {
    throw new AgenticError("INVALID_INPUT", "Expected a regular file.", { path });
  }
  if (info.size > maxBytes) {
    throw new AgenticError(
      "FILE_TOO_LARGE",
      `File is ${info.size.toLocaleString()} bytes; limit is ${maxBytes.toLocaleString()} bytes.`,
      { path, bytes: info.size, maxBytes },
    );
  }
  const buffer = await readFile(path);
  if (buffer.length > maxBytes) {
    throw new AgenticError(
      "FILE_TOO_LARGE",
      `File grew to ${buffer.length.toLocaleString()} bytes while being read; limit is ${maxBytes.toLocaleString()} bytes.`,
      { path, bytes: buffer.length, maxBytes },
    );
  }
  if (appearsBinary(buffer)) {
    throw new AgenticError("BINARY_FILE", "Binary files are not editable as text.", {
      path,
    });
  }
  const content = buffer.toString("utf8");
  if (!Buffer.from(content, "utf8").equals(buffer)) {
    throw new AgenticError(
      "BINARY_FILE",
      "File is not valid UTF-8 text and cannot be edited safely.",
      { path },
    );
  }
  return snapshotFromContent(content);
}

export function snapshotFromContent(content: string): TextFileSnapshot {
  return {
    content,
    bytes: Buffer.byteLength(content),
    sha256: sha256Text(content),
    lineCount: content.length === 0 ? 0 : countLines(content.split(/\r?\n/)),
    newline: detectNewline(content),
    hasFinalNewline: /\r?\n$/.test(content),
  };
}

/**
 * Matches one line of `numberedRange` output: optional leading padding, the
 * captured line number, " | ", then the source line (which may be empty, so the
 * separator is allowed to end the line).
 */
const NUMBERED_LINE = /^[ \t]*(\d+)[ \t]*\|([ \t]|$)/;

/**
 * True when `content` looks like it was pasted back out of `numberedRange`
 * (i.e. out of a `workspace_inspect read` result) instead of being real file
 * text. Small models routinely copy the `12 | ` gutter into edit content, which
 * silently corrupts the file or makes an exact replace miss.
 *
 * Three conditions, all required. The decisive one is the **run**: read output
 * is by construction `n, n+1, n+2`, so the matched prefixes must be adjacent
 * lines whose numbers ascend by exactly one, at least twice in a row. Data that
 * happens to use pipes almost never does that — measured over 36 inputs, adding
 * the run requirement (and only then dropping the count threshold from three
 * matched lines to two) moved false positives from 12 to 8 and missed
 * detections from 6 to 3.
 *
 * What it still does *not* catch, deliberately: a genuinely consecutive numeric
 * first column — `1 | a`, `2 | b`, `3 | c`, or consecutive years — is
 * indistinguishable from read output, and `allow_line_numbers: true` is the
 * escape hatch for it. What it reliably leaves alone is a leading-pipe markdown
 * table (no line starts with a bare number) and any non-monotonic numeric
 * column: descending, sparse, or stepped. A single numbered line is below any
 * threshold a majority rule can safely use and is never flagged; the zero-match
 * hint on `replace` covers that case after the fact instead.
 */
export function looksLineNumbered(content: string): boolean {
  const lines = content.split(/\r?\n/).filter((line) => line.trim() !== "");
  if (lines.length < 2) return false;
  let numbered = 0;
  let longestRun = 0;
  let run = 0;
  let previous: number | undefined;
  for (const line of lines) {
    const match = NUMBERED_LINE.exec(line);
    if (!match) {
      run = 0;
      previous = undefined;
      continue;
    }
    numbered++;
    const value = Number(match[1]);
    run = previous !== undefined && value === previous + 1 ? run + 1 : 1;
    if (run > longestRun) longestRun = run;
    previous = value;
  }
  return numbered >= 2 && longestRun >= 2 && numbered * 2 > lines.length;
}

/**
 * Removes the `12 | ` gutter `numberedRange` adds, from every line. Used only
 * to explain a failure after the fact (see `workspace/transactions.ts`), never
 * to rewrite what a caller asked for: silently repairing an edit would let a
 * model commit text it never actually looked at.
 *
 * The trailing separator class matches `NUMBERED_LINE`'s own tolerance: with
 * ` ?` alone, a tab-separated `1\t|\ta` was recognised as numbered but stripped
 * to `\ta`, so the sentence that quotes the stripped text was wrong by a tab.
 */
export function stripLineNumberGutter(content: string): string {
  return content.replace(/^[ \t]*\d+[ \t]*\|[ \t]?/gm, "");
}

export function numberedRange(
  content: string,
  startLine: number,
  endLine: number,
): { text: string; totalLines: number; startLine: number; endLine: number } {
  const lines = content.split(/\r?\n/);
  const totalLines = content.length === 0 ? 0 : countLines(lines);
  if (totalLines === 0) {
    return { text: "", totalLines: 0, startLine: 0, endLine: 0 };
  }
  const start = Math.max(1, Math.min(startLine, totalLines));
  const end = Math.max(start, Math.min(endLine, totalLines));
  const width = String(end).length;
  const selected = lines
    .slice(start - 1, end)
    .map((line, index) => `${String(start + index).padStart(width, " ")} | ${line}`)
    .join("\n");
  return { text: selected, totalLines, startLine: start, endLine: end };
}