char-slop/berry

git clone https://git.t4t.associates/char-slop/berry

Charlotte SomConfigure code editing as naive, hashline, or both per conversation8744a752026-10-11

import { Buffer, isUtf8 } from "node:buffer";
import { createHash } from "node:crypto";
import { structuredPatch } from "diff";
import anchorData from "./anchor-pieces.json" with { type: "json" };

export const MAX_LINES = 2000;
export const MAX_BYTES = 50 * 1024;

export function hash(line: string): string {
  const value = createHash("sha256").update(line).digest().readUInt32BE(0);
  const prefixCount = anchorData.prefixes.length / 2;
  const prefix = (value % prefixCount) * 2;
  const suffix = (Math.floor(value / prefixCount) % (anchorData.suffixes.length / 2)) * 2;
  return anchorData.prefixes.slice(prefix, prefix + 2) +
    anchorData.suffixes.slice(suffix, suffix + 2);
}

export type TextFile = {
  readonly raw: string;
  readonly bom: string;
  readonly text: string;
  /** Lines with their line endings. */
  readonly rows: readonly string[];
  readonly lines: readonly string[];
};

/** The most text read or edit takes in. */
export const MAX_TEXT_BYTES = 50 * 1024 * 1024;

export function decodeText(content: Uint8Array): TextFile {
  const bytes = Buffer.from(content.buffer, content.byteOffset, content.byteLength);
  if (!isUtf8(bytes) || bytes.includes(0)) {
    throw new Error("Read and edit require UTF-8 text without NUL bytes.");
  }
  const raw = bytes.toString("utf8");
  const bom = raw.startsWith("\uFEFF") ? "\uFEFF" : "";
  const text = raw.slice(bom.length);
  const rows = text.match(/[^\n]*\n|[^\n]+$/g) ?? [];
  return { raw, bom, text, rows, lines: rows.map((row) => row.replace(/\r?\n$/, "")) };
}

/** How many leading rows fit in `maxBytes`, newlines between them included. */
export function fitting(rows: readonly string[], maxBytes = MAX_BYTES, separatorBytes = 1): number {
  let bytes = -separatorBytes;
  for (const [i, row] of rows.entries()) {
    bytes += Buffer.byteLength(row) + separatorBytes;
    if (bytes > maxBytes) return i;
  }
  return rows.length;
}

export function renderRead(
  file: TextFile,
  offset = 1,
  limit = MAX_LINES,
  hashes = true,
): string {
  const { lines } = file;
  const start = offset - 1;
  if (start > 0 && start >= lines.length) throw new Error("Offset is beyond end of file.");
  if (!lines.length) {
    return hashes
      ? "[Empty file; edit with old: [] to insert.]"
      : '[Empty file; edit with oldText: "" to insert.]';
  }
  const end = start + Math.min(limit, MAX_LINES);
  const rows = hashes
    ? lines.slice(start, end).map((line) => `${hash(line)}│${line}`)
    : file.rows.slice(start, end);
  const count = fitting(rows, MAX_BYTES, hashes ? 1 : 0);
  if (count === 0) throw new Error(`Line ${start + 1} exceeds 50 KiB; use bash to inspect it.`);
  const next = start + count;
  const continuation = next < lines.length
    ? `\n\n[Showing lines ${start + 1}-${next} of ${lines.length}. Use offset=${
      next + 1
    } to continue.]`
    : "";
  return rows.slice(0, count).join(hashes ? "\n" : "") + continuation;
}

export type HashlineEdit = {
  readonly before?: readonly string[];
  readonly old: readonly string[];
  readonly after?: readonly string[];
  readonly new: readonly string[];
};

/** The edited file contents and a diff; `content === file.raw` when nothing changes. */
export function planEdit(
  file: TextFile,
  edits: readonly (HashlineEdit | { readonly oldText: string; readonly newText: string })[],
  hashes = true,
): { content: string; diff: string[] } {
  const { bom, text, rows, lines } = file;
  const anchors = edits.some((edit) => "old" in edit) ? lines.map(hash) : [];
  const eol = text.match(/\r?\n/)?.[0] ?? "\n";
  const offsets = [0];
  for (const row of rows) offsets.push(offsets.at(-1)! + row.length);

  const changes = edits.map((edit, index) => {
    if ("oldText" in edit) {
      if (/[\0]/.test(edit.oldText + edit.newText)) {
        throw new Error(`Edit ${index + 1}: text must not contain NUL.`);
      }
      if (!edit.oldText && text) {
        throw new Error(`Edit ${index + 1}: empty oldText is allowed only for an empty file.`);
      }
      const start = text.indexOf(edit.oldText);
      if (start === -1) {
        throw new Error(
          `Edit ${index + 1}: oldText not found (stale reference); read the file again.`,
        );
      }
      if (edit.oldText && text.indexOf(edit.oldText, start + 1) !== -1) {
        throw new Error(`Edit ${index + 1}: ambiguous oldText; include more surrounding text.`);
      }
      return {
        start,
        end: start + edit.oldText.length,
        replacement: edit.newText,
        hashline: false,
        separate: false,
      };
    }

    const before = edit.before ?? [];
    const sequence = [...before, ...edit.old, ...(edit.after ?? [])];
    if (!sequence.length && lines.length) {
      throw new Error(`Edit ${index + 1}: supply old or adjacent context hashes.`);
    }
    if (edit.new.some((line) => /[\r\n\0]/.test(line))) {
      throw new Error(`Edit ${index + 1}: new must contain individual lines, without CR/LF/NUL.`);
    }
    let match = -1;
    for (let i = 0; i <= anchors.length - sequence.length; i++) {
      if (!sequence.every((anchor, j) => anchor === anchors[i + j])) continue;
      if (match !== -1) {
        throw new Error(`Edit ${index + 1}: ambiguous hashes; add adjacent before/after context.`);
      }
      match = i;
    }
    if (match === -1) {
      throw new Error(
        `Edit ${index + 1}: hashes not found (stale reference); read the file again.`,
      );
    }
    const start = match + before.length;
    const end = start + edit.old.length;
    const replacement = edit.new.map((line, i) => {
      const row = start + i < end && line === lines[start + i] ? rows[start + i] : line + eol;
      return i < edit.new.length - 1 && !row.endsWith("\n") ? row + eol : row;
    }).join("");
    return {
      start: offsets[start],
      end: offsets[end],
      replacement,
      hashline: true,
      separate: start === rows.length,
    };
  }).sort((a, b) => a.start - b.start);

  for (let i = 1; i < changes.length; i++) {
    if (changes[i].start < changes[i - 1].end || changes[i].start === changes[i - 1].start) {
      throw new Error("Overlapping edits (or insertions at the same position); merge them.");
    }
  }

  const updated: string[] = [];
  let cursor = 0;
  for (const { start, end, replacement, separate } of changes) {
    updated.push(text.slice(cursor, start));
    // Inserting after an unterminated last line needs a separator, not a joined line.
    if (
      separate && replacement && updated.findLast((part) => part.length)?.endsWith("\n") === false
    ) {
      updated.push(eol);
    }
    updated.push(replacement);
    cursor = end;
  }
  updated.push(text.slice(cursor));
  let result = updated.join("");
  const last = changes.at(-1);
  if (
    last?.hashline && last.end === text.length && text && !text.endsWith("\n") &&
    result !== eol && !result.endsWith("\n" + eol)
  ) {
    result = result.replace(/\r?\n$/, "");
  }

  const diff = structuredPatch("", "", text, result, "", "", { context: 3 }).hunks.flatMap((
    hunk,
  ) => [
    `@@ original line ${hunk.oldStart} @@`,
    ...hunk.lines.flatMap((row, i) => {
      if (row.startsWith("\\")) return [];
      const line = hunk.lines[i + 1]?.startsWith("\\")
        ? row.slice(1)
        : row.slice(1).replace(/\r$/, "");
      return [
        row.startsWith("+") || row.startsWith("-")
          ? `${row[0]}${hashes ? `${hash(line)}│` : ""}${line}`
          : ` ${line}`,
      ];
    }),
  ]);
  return { content: bom + result, diff };
}