@dittowords/text-extract

Finds strings in a codebase that are potentially user-facing.

Usage

A whole directory:

import { runExtract } from "@dittowords/text-extract";

const { candidates, summary } = await runExtract({ inputPath: "/path/to/repo" });

Skips vendored, build and test trees, and honors .gitignore.

One file:

import { extractFile } from "@dittowords/text-extract";

const { candidates, skipped } = await extractFile({
  relPath: "src/Button.tsx", // used verbatim as `location.file`
  source: fileContents,
  framework: ["react", "next"], // optional
});

Scans whatever path it's given — no exclusions are applied. framework is the same token list runExtract derives from package.json files and native project markers; omitting it only costs a classification hint.

skipped is set when the file wasn't scanned at all (unsupported_language, unconfirmed_i18n_file, minified) — not the same as scanning it and finding nothing.

Both entry points reach the same verdict on the same file.

Notes

• Candidate order may not be stable between runs, because globby's file iteration order varies. The set is stable; compare accordingly.
• summary.filesFailed counts files an extractor threw on. They're dropped from the results, so a non-zero count means strings are missing.

Development

yarn test
yarn typecheck
yarn build

Tests
lang/extractors/*.test.ts	Per-construct classification. Exact line and column.
golden.test.ts	Realistic fixtures from testing/golden/, with hand-written expectations. Fixtures are tested under a relPath that need not match their location on disk, since path shape drives i18n admission and locale detection.
extract-file.test.ts	That the two entry points agree per file.
extract-failures.test.ts	That a throwing extractor is counted, not swallowed.
extract-performance.test.ts	That the regex fallback doesn't backtrack catastrophically on dense input.