@cplieger/deadset-ts 8.0.0-dev.2 → 8.0.0-dev.21

package/src/text/glob.ts

@@ -0,0 +1,210 @@
+/**
+ * Glob patterns over paths in two forms. The configured form (the configuration, convention
+ * rows, bundler entries, workspace member patterns) has `*` and `?` within a segment, `**`
+ * across segments and brace lists. A `.gitignore` line has git's form: bracket expressions
+ * and backslash escapes, no brace lists (https://git-scm.com/docs/gitignore#_pattern_format).
+ * A configured root is not a glob: it names a symbol reference, and is read where it is used.
+ */
+
+type GlobForm = "configured" | "gitignore";
+
+/**
+ * One character of a pattern as the expression syntax reads it literally. The set escaped
+ * is closed: under the unicode flag an escape of any other character is refused outright.
+ */
+function quoted(character: string): string {
+  return /[.*+?^${}()|[\]\\/]/u.test(character) ? `\\${character}` : character;
+}
+
+/**
+ * The ASCII ranges each POSIX class a bracket expression may name holds, as git's own
+ * character table reads them, each range written as its first and last character.
+ */
+const CLASSES: ReadonlyMap<string, string> = new Map([
+  ["alnum", "09AZaz"],
+  ["alpha", "AZaz"],
+  ["blank", "\t\t  "],
+  ["cntrl", "\0\x1f\x7f\x7f"],
+  ["digit", "09"],
+  ["graph", "!~"],
+  ["lower", "az"],
+  ["print", " ~"],
+  ["punct", "!/:@[`{~"],
+  ["space", "\t\n\r\r  "],
+  ["upper", "AZ"],
+  ["xdigit", "09AFaf"],
+]);
+
+/** What a pattern git cannot read matches: nothing, as git aborts the whole match. */
+const NOTHING = "(?!)";
+
+/** One code point as a member of an expression class. */
+function member(point: number): string {
+  return `\\u{${point.toString(16)}}`;
+}
+
+/**
+ * The class a bracket expression opening at `open` denotes, and the index past its `]`,
+ * read as git's wildmatch reads it: a leading `!` or `^` negates it, a `]` first is a
+ * member, `\` escapes one character, `x-y` is a range (one whose `y` is below `x` holds
+ * `x` alone) and `[:name:]` a POSIX class. An unclosed bracket or an unknown class name
+ * matches nothing. Neither form matches the solidus.
+ */
+function bracket(pattern: string, open: number): { source: string; end: number } {
+  let at = open + 1;
+  const negated = pattern[at] === "!" || pattern[at] === "^";
+  if (negated) {
+    at += 1;
+  }
+  const next = (): number | undefined => {
+    const point = pattern.codePointAt(at);
+    at += point !== undefined && point > 0xffff ? 2 : 1;
+    return point;
+  };
+  let members = "";
+  let low: number | undefined;
+  for (let first = true; ; first = false) {
+    let point = next();
+    if (point === undefined) {
+      return { source: NOTHING, end: pattern.length };
+    }
+    if (point === 0x5d && !first) {
+      return { source: negated ? `[^/${members}]` : `(?!/)[${members}]`, end: at };
+    }
+    if (point === 0x2d && low !== undefined && at < pattern.length && pattern[at] !== "]") {
+      let high = next();
+      if (high === 0x5c) {
+        high = next();
+      }
+      if (high === undefined) {
+        return { source: NOTHING, end: pattern.length };
+      }
+      members += high >= low ? `${member(low)}-${member(high)}` : "";
+      low = undefined;
+      continue;
+    }
+    if (point === 0x5b && pattern[at] === ":") {
+      const close = pattern.indexOf("]", at + 1);
+      if (close < 0) {
+        return { source: NOTHING, end: pattern.length };
+      }
+      if (close > at + 1 && pattern[close - 1] === ":") {
+        const ranges = CLASSES.get(pattern.slice(at + 1, close - 1));
+        if (ranges === undefined) {
+          return { source: NOTHING, end: pattern.length };
+        }
+        for (let one = 0; one < ranges.length; one += 2) {
+          members += `${member(ranges.charCodeAt(one))}-${member(ranges.charCodeAt(one + 1))}`;
+        }
+        low = undefined;
+        at = close + 1;
+        continue;
+      }
+    }
+    if (point === 0x5c) {
+      point = next();
+      if (point === undefined) {
+        return { source: NOTHING, end: pattern.length };
+      }
+    }
+    members += member(point);
+    low = point;
+  }
+}
+
+/** The expression source of one pattern with no brace list. */
+function source(pattern: string, form: GlobForm): string {
+  let text = "";
+  let at = 0;
+  while (at < pattern.length) {
+    if (pattern.startsWith("**/", at)) {
+      text += "(?:[^/]*/)*";
+      at += 3;
+      continue;
+    }
+    if (pattern.startsWith("**", at)) {
+      text += ".*";
+      at += 2;
+      continue;
+    }
+    const character = pattern[at] ?? "";
+    if (form === "gitignore" && character === "[") {
+      const read = bracket(pattern, at);
+      text += read.source;
+      at = read.end;
+      continue;
+    }
+    if (form === "gitignore" && character === "\\" && at + 1 < pattern.length) {
+      text += quoted(pattern[at + 1] ?? "");
+      at += 2;
+      continue;
+    }
+    at += 1;
+    text += character === "*" ? "[^/]*" : character === "?" ? "[^/]" : quoted(character);
+  }
+  return text;
+}
+
+/** A pattern with every brace list expanded into its alternatives; a lone `{` stands for itself. */
+export function braceExpanded(pattern: string): readonly string[] {
+  const open = pattern.indexOf("{");
+  const close = open < 0 ? -1 : pattern.indexOf("}", open);
+  if (close < 0) {
+    return [pattern];
+  }
+  const before = pattern.slice(0, open);
+  const after = pattern.slice(close + 1);
+  return pattern
+    .slice(open + 1, close)
+    .split(",")
+    .flatMap((member) => braceExpanded(before + member + after));
+}
+
+/** The expression one whole-path pattern of the configured form denotes. */
+export function globExpression(pattern: string): RegExp {
+  const alternatives = braceExpanded(pattern).map((one) => source(one, "configured"));
+  return new RegExp(`^(?:${alternatives.join("|")})$`, "u");
+}
+
+/** The expression one `.gitignore` pattern denotes over a path below its file's directory. */
+export function gitignoreExpression(pattern: string): RegExp {
+  return new RegExp(`^${source(pattern, "gitignore")}$`, "u");
+}
+
+const SEGMENTS = new Map<string, RegExp>();
+
+/**
+ * Whether one directory name matches one segment of a workspace's member pattern, its brace
+ * lists already expanded. Neither `*` nor `?` matches a leading dot, as npm's workspaces read
+ * them: https://docs.npmjs.com/cli/using-npm/workspaces
+ */
+export function segmentMatches(segment: string, name: string): boolean {
+  if (name.startsWith(".") && !segment.startsWith(".")) {
+    return false;
+  }
+  let expression = SEGMENTS.get(segment);
+  if (expression === undefined) {
+    expression = globExpression(segment);
+    SEGMENTS.set(segment, expression);
+  }
+  return expression.test(name);
+}
+
+/**
+ * Whether one path's segments match a member pattern's, `**` standing for any run of segments
+ * that starts with no dot.
+ */
+export function pathMatches(pattern: readonly string[], path: readonly string[]): boolean {
+  const [head, ...rest] = pattern;
+  if (head === undefined) {
+    return path.length === 0;
+  }
+  if (head === "**") {
+    return (
+      pathMatches(rest, path) ||
+      (path.length > 0 && !(path[0] ?? "").startsWith(".") && pathMatches(pattern, path.slice(1)))
+    );
+  }
+  const [name, ...below] = path;
+  return name !== undefined && segmentMatches(head, name) && pathMatches(rest, below);
+}

package/src/text/named-files.ts

@@ -0,0 +1,388 @@
+/** Layer 3 of the command reader: the files its words name, and the package names a command's text names. */
+
+import type { HeldWord } from "../commands/held.ts";
+import { PART } from "../commands/shell.ts";
+import { APPENDED_EXTENSIONS, extensionOf } from "../extensions.ts";
+import { dirnamePath, normalizePath, relativePath } from "../paths.ts";
+
+/** The extensions a name of a file drops: a source's, an emitted file's, and a declaration's (roots.13). */
+const NAMED_EXTENSIONS = [...APPENDED_EXTENSIONS, ".d.ts", ".d.mts", ".d.cts"];
+
+/** A character a package's or a command's name may hold, so one beside a name makes it part of a longer one. */
+const NAME_CHARACTER = /[\w.@-]/u;
+
+const SEPARATOR = /[/\\]/u;
+
+/**
+ * What makes one name of a file a path where it appears: a `stem`, a last segment without its
+ * extension, may have the extension after it; a `segment`, a whole last segment, is one where
+ * it holds an extension of its own; a `directory`, an `index` file's, may have a separator
+ * after it; and `root` is `.`, the package's own directory.
+ */
+type NameKind = "stem" | "segment" | "directory" | "root";
+
+interface FileName {
+  readonly text: string;
+  readonly kind: NameKind;
+}
+
+/** One file, and every path it is spelled by: its own and each one the project writes it to. */
+interface Spelled<File> {
+  readonly file: File;
+  readonly paths: readonly string[];
+}
+
+/**
+ * The names whose appearance as a path in a text holds one path below `dir`: its last segment
+ * without the extension, or whole where nothing else is left or it has none, and for an
+ * `index` file its directory's last segment, and `.` too where that is `dir`. Every spelling
+ * of the path that writes it as a path holds one of them (roots.13). None is empty, because
+ * an empty name is in every text.
+ */
+function namesOf(path: string, dir: string): readonly FileName[] {
+  const last = path.slice(path.lastIndexOf("/") + 1);
+  const bare = last.slice(0, last.length - (extensionOf(last, NAMED_EXTENSIONS)?.length ?? 0));
+  const own: FileName =
+    bare === "" || bare === last ? { text: last, kind: "segment" } : { text: bare, kind: "stem" };
+  if (own.text !== "index") {
+    return [own];
+  }
+  const parent = dirnamePath(path);
+  const directory = parent.slice(parent.lastIndexOf("/") + 1);
+  return [
+    own,
+    ...(directory === "" ? [] : [{ text: directory, kind: "directory" as const }]),
+    ...(parent === normalizePath(dir) ? [{ text: ".", kind: "root" as const }] : []),
+  ];
+}
+
+/**
+ * Whether a name that ends at `end` ends there as a path: no name character follows, nor a
+ * separator, which goes on to a path below it, nor a part, whose text goes on with the name,
+ * nor a `.` that a word character or a part follows, which goes on to another extension.
+ */
+function endsAt(text: string, end: number): boolean {
+  const next = text.charAt(end);
+  const after = text.charAt(end + 1);
+  return (
+    !/[\w@/\\-]/u.test(next) &&
+    next !== PART &&
+    !(next === "." && (/\w/u.test(after) || after === PART))
+  );
+}
+
+/** A character a word that writes a path may hold, a part only its run knows among them. */
+const PATH_CHARACTER = new RegExp(`[\\w.@+/\\\\${PART}-]`, "u");
+
+/** A run of {@link PATH_CHARACTER}s: a word as {@link noFileEnd} reads one. */
+const PATH_WORD = new RegExp(`${PATH_CHARACTER.source}+`, "gu");
+
+const SCOPED_PACKAGE = /^@[\w-][\w.-]*\/[\w-][\w.-]*(?:@[\w.-]*)?$/u;
+
+/**
+ * The end of what `text` writes from `start`, where `word` starts, that names an image or a
+ * package and never a file, or `start` where it names none (roots.13): a value of a `docker:`
+ * that starts no longer name, and a scoped package's name, which no part goes on with, since
+ * a part beside it may make it a path.
+ */
+function noFileEnd(text: string, start: number, word: string): number {
+  const end = start + word.length;
+  const image =
+    word.startsWith("//") &&
+    start >= 7 &&
+    text.startsWith("docker:", start - 7) &&
+    !NAME_CHARACTER.test(text.charAt(start - 8));
+  return image || (SCOPED_PACKAGE.test(word) && !holdsExtension(word)) ? end : start;
+}
+
+/** Each place of `text` that {@link noFileEnd} rules out, as a 1. */
+function placesNamingNoFile(text: string): Uint8Array {
+  const ruledOut = new Uint8Array(text.length);
+  for (const { 0: word, index: start } of text.matchAll(PATH_WORD)) {
+    ruledOut.fill(1, start, noFileEnd(text, start, word));
+  }
+  return ruledOut;
+}
+
+/** The place after the separators `text` holds from `at`. */
+function pastSeparators(text: string, at: number): number {
+  let after = at;
+  while (SEPARATOR.test(text.charAt(after))) {
+    after += 1;
+  }
+  return after;
+}
+
+/** Whether `name`, found in `text` at `at`, appears there as a path (roots.13). */
+function appearsAsPath(text: string, at: number, name: FileName): boolean {
+  const before = text.charAt(at - 1);
+  const end = at + name.text.length;
+  if (before === PART) {
+    return false;
+  }
+  if (name.kind === "root") {
+    const after = pastSeparators(text, end);
+    return (
+      [before, text.charAt(after)].every(
+        (one) => !NAME_CHARACTER.test(one) && !SEPARATOR.test(one),
+      ) && text.charAt(after) !== PART
+    );
+  }
+  if (NAME_CHARACTER.test(before)) {
+    return false;
+  }
+  const after = pastSeparators(text, end);
+  if (name.kind === "directory" && after > end && endsAt(text, after)) {
+    return true;
+  }
+  if (
+    (SEPARATOR.test(before) || (name.kind === "segment" && name.text.indexOf(".", 1) > 0)) &&
+    endsAt(text, end)
+  ) {
+    return true;
+  }
+  return (
+    name.kind === "stem" &&
+    NAMED_EXTENSIONS.some(
+      (extension) => text.startsWith(extension, end) && endsAt(text, end + extension.length),
+    )
+  );
+}
+
+/** Whether some text holds one of the extensions as a whole. */
+function holdsExtension(written: string): boolean {
+  return NAMED_EXTENSIONS.some((extension) =>
+    new RegExp(`${extension.replaceAll(".", "\\.")}(?!\\w)`, "u").test(written),
+  );
+}
+
+/**
+ * A word with a part only its run knows: the text a path it names ends in after a `/`, split
+ * at each part, and whether it names a directory.
+ */
+interface DynamicWord {
+  readonly glob: readonly string[];
+  readonly directory: boolean;
+}
+
+/** Whether a word's segment writes path text: text beside its parts other than `.` and `..`. */
+function writesPath(segment: string): boolean {
+  return segment.split(PART).some((piece) => piece !== "" && piece !== "." && piece !== "..");
+}
+
+/**
+ * Each run of path characters of `word`, at `start`, whose last segment holds a part, read as
+ * roots.13 reads one from past its last segment before the last that writes no path: one that
+ * writes none names nothing and is `unwritten` where the word may write the file a command runs,
+ * where a bare one names a file too. The word stands `depth` segments below the target root,
+ * whose own directories no checkout names alike: a leading `..` past it writes none, and an
+ * absolute run reads from its last segment that writes a path, `unwritten` where more is left.
+ */
+function dynamicWords(
+  word: HeldWord,
+  start: number,
+  ruledOut: Uint8Array,
+  depth: number,
+): { readonly found: readonly DynamicWord[]; readonly unwritten: boolean } {
+  const { text, runs } = word;
+  const found: DynamicWord[] = [];
+  let unwritten = false;
+  const read = (from: number, next: number): void => {
+    const entry = next < 0;
+    // A `.` the word ends with ends a sentence, as endsAt reads one after a name.
+    const run = entry ? text : text.slice(from, next).replace(/\.+$/u, "");
+    const all = run.replaceAll("\\", "/").split("/");
+    const directory = all.length > 1 && all.at(-1) === "";
+    const written = all.filter((segment) => segment !== "" && segment !== ".");
+    const last = (written.at(-1) ?? "").split(PART);
+    // A runtime's entry is a path whatever it looks like, a scoped package's name among them.
+    if (!entry && (last.length < 2 || ruledOut[start + from] === 1)) {
+      return;
+    }
+    const climbs = written.findIndex((segment) => segment !== "..");
+    const above = Math.max(0, (climbs < 0 ? written.length : climbs) - depth);
+    const segments = written.filter((_, at) => at < climbs || at >= climbs + above);
+    if (!segments.some(writesPath)) {
+      unwritten ||= runs && (!entry || above > 0);
+    } else if (all.length > 1 || holdsExtension(last.join("")) || runs) {
+      const kept = directory ? segments.length : segments.length - 1;
+      const anyDirectory = segments.slice(0, kept).findLastIndex((one) => !writesPath(one));
+      const absolute = all[0] === "" && all.length > 1 && anyDirectory < 0;
+      const from = absolute ? segments.findLastIndex(writesPath) : anyDirectory + 1;
+      unwritten ||= runs && absolute && from < segments.length - 1;
+      const glob = segments.slice(from).join("/");
+      found.push({ glob: glob === "" ? ["", ""] : glob.split(PART), directory });
+    }
+  };
+  if (runs && !text.includes(PART)) {
+    read(0, -1);
+  }
+  let at = text.indexOf(PART);
+  while (at >= 0) {
+    let from = at;
+    while (from > 0 && PATH_CHARACTER.test(text.charAt(from - 1))) {
+      from -= 1;
+    }
+    let next = at;
+    while (next < text.length && PATH_CHARACTER.test(text.charAt(next))) {
+      next += 1;
+    }
+    read(from, next);
+    at = text.indexOf(PART, Math.max(next, at + 1));
+  }
+  return { found, unwritten };
+}
+
+/** Whether `segment` is the text `glob` writes, any text standing between two of its parts. */
+function standsFor(glob: readonly string[], segment: string): boolean {
+  const first = glob[0] ?? "";
+  const last = glob.at(-1) ?? "";
+  if (glob.length === 1) {
+    return segment === first;
+  }
+  if (
+    !segment.startsWith(first) ||
+    !segment.endsWith(last) ||
+    first.length + last.length > segment.length
+  ) {
+    return false;
+  }
+  let at = first.length;
+  const end = segment.length - last.length;
+  for (const middle of glob.slice(1, -1)) {
+    const found = segment.indexOf(middle, at);
+    if (found < 0 || found + middle.length > end) {
+      return false;
+    }
+    at = found + middle.length;
+  }
+  return true;
+}
+
+/** Whether a word's glob may write `path` or the end of it after a `/`. */
+function mayStandFor(glob: readonly string[], path: string): boolean {
+  let at = 0;
+  while (!standsFor(glob, path.slice(at))) {
+    const slash = path.indexOf("/", at);
+    if (slash < 0) {
+      return false;
+    }
+    at = slash + 1;
+  }
+  return true;
+}
+
+/**
+ * Each spelling of one path below `dir` a word may write the end of: the whole path from `root`,
+ * so a word may write the directories between them too, but no checkout's own, without its
+ * extension, and without its `/index` part, which alone is a directory's spelling.
+ */
+function spellingsBelow(
+  path: string,
+  dir: string,
+  root: string,
+): { readonly file: readonly string[]; readonly directory: readonly string[] } {
+  const fromRoot = relativePath(root, path);
+  if (relativePath(dir, path) === undefined || fromRoot === undefined) {
+    return { file: [], directory: [] };
+  }
+  const whole = `/${fromRoot}`;
+  const extension = extensionOf(whole, NAMED_EXTENSIONS);
+  const bare = extension === undefined ? whole : whole.slice(0, -extension.length);
+  const directory = bare.endsWith("/index") ? [bare.slice(0, -"/index".length)] : [];
+  return { file: [whole, bare, ...directory], directory };
+}
+
+/**
+ * Every file of `files` below `dir` that `words` name: one of whose {@link namesOf} a path of it
+ * has appears in a word as a path, or a spelling of one of whose paths a word that writes a
+ * path with a part only its run knows may stand for, and whether a word that may write the
+ * file a command runs writes no path text. A file the words do not name may be named too, and
+ * no file a word writes as a path outside an image's or a package's name is missed.
+ */
+export function spelledFiles<File>(
+  words: readonly HeldWord[],
+  dir: string,
+  root: string,
+  files: readonly Spelled<File>[],
+): { readonly files: readonly File[]; readonly unwritten: boolean } {
+  const byName = new Map<string, { readonly file: File; readonly name: FileName }[]>();
+  for (const { file, paths } of files) {
+    const [own] = paths;
+    if (own === undefined || relativePath(dir, own) === undefined) {
+      continue;
+    }
+    const seen = new Set<string>();
+    for (const name of paths.flatMap((path) => namesOf(path, dir))) {
+      const key = `${name.kind}\u0000${name.text}`;
+      if (!seen.has(key)) {
+        seen.add(key);
+        byName.set(name.text, [...(byName.get(name.text) ?? []), { file, name }]);
+      }
+    }
+  }
+  const found = new Set<File>();
+  const text = words.map((word) => word.text).join("\n");
+  const ruledOut = placesNamingNoFile(text);
+  // One pass per name length: each place of the text is looked up once for every length.
+  for (const length of new Set([...byName.keys()].map((name) => name.length))) {
+    for (let at = 0; at + length <= text.length; at += 1) {
+      for (const { file, name } of byName.get(text.slice(at, at + length)) ?? []) {
+        if (!found.has(file) && ruledOut[at] !== 1 && appearsAsPath(text, at, name)) {
+          found.add(file);
+        }
+      }
+    }
+  }
+  const dynamic: DynamicWord[] = [];
+  let unwritten = false;
+  let start = 0;
+  const depth =
+    normalizePath(dir) === normalizePath(root)
+      ? 0
+      : (relativePath(root, dir)?.split("/").length ?? 0);
+  for (const word of words) {
+    const read = dynamicWords(word, start, ruledOut, depth);
+    dynamic.push(...read.found);
+    unwritten ||= read.unwritten;
+    start += word.text.length + 1;
+  }
+  if (dynamic.length > 0) {
+    for (const { file, paths } of files) {
+      const spellings = paths.map((path) => spellingsBelow(path, dir, root));
+      if (
+        !found.has(file) &&
+        dynamic.some((word) =>
+          spellings.some((spelled) =>
+            (word.directory ? spelled.directory : spelled.file).some((one) =>
+              mayStandFor(word.glob, one),
+            ),
+          ),
+        )
+      ) {
+        found.add(file);
+      }
+    }
+  }
+  return { files: [...found], unwritten };
+}
+
+/**
+ * Each of `names`, a package's or a command's name, that `text` holds with no name character
+ * before it and no character after it but one a path or a version may go on with, `/`, `.` or
+ * `@` among them: `module-alias/register` holds `module-alias`, and `tsxx` holds no `tsx`.
+ */
+export function heldNames(text: string, names: readonly string[]): ReadonlySet<string> {
+  const found = new Set<string>();
+  for (const name of names) {
+    for (let at = text.indexOf(name); at >= 0 && name !== ""; at = text.indexOf(name, at + 1)) {
+      const before = text.charAt(at - 1);
+      const after = text.charAt(at + name.length);
+      if (!NAME_CHARACTER.test(before) && !/[\w-]/u.test(after)) {
+        found.add(name);
+        break;
+      }
+    }
+  }
+  return found;
+}