editorconfig Cyrillic triage

Profile keys consumed

• make.format
• make.lint_eslint

Overview

editorconfig-checker measures a line in UTF-8 bytes; Prettier's printWidth and ESLint's max-len measure characters. Cyrillic encodes as two bytes per character, so a line that is comfortably inside the column limit can be reported as over it. Decide which count applies before touching the line — a hand-wrap that Prettier disagrees with is reverted by the next format run and turns a false positive into a real formatting failure.

When to use

• A finding cites a line length that does not match what the editor shows.
• The flagged line holds a Ukrainian or Russian string literal, label, or story argument.
• You are about to wrap a JSX line to satisfy a length complaint.
• Not for: lines that are genuinely over the column limit, and not for non-text overruns such as long imports or URLs, which are wrapped normally.

Applicability by repository shape

• React SPA shape (feature modules under the source root, a bootable app, an aggregate CI target): partial — .editorconfig sets max_line_length = 100 and .prettierrc sets printWidth: 100, but .qlty/qlty.toml registers no editorconfig-checker; the enforced character-count gates are the repository's check-only Prettier target and ESLint max-len at 100, run through the target mapped by make.lint_eslint.
• Next.js app shape (routed pages, no aggregate duplication gate): yes — .qlty/qlty.toml runs editorconfig-checker with mode = "comment"; .editorconfig is 100 and .prettierrc printWidth is 100. The target mapped by make.format is the Prettier pass.
• Component-library shape (Storybook-first, no bootable app, published package): yes — same qlty mode = "comment" setup, 100/100 limits, plus ESLint max-len at 100; its check-only formatter target runs prettier . --check.

Procedure

1. Measure both counts for the flagged line.

   sed -n '<line>p' <file> | python3 -c "import sys
   raw = sys.stdin.buffer.read().rstrip(b'\n')
   print('chars', len(raw.decode()), 'bytes', len(raw))"
   

   Do not reach for awk '{print length}': its length returns characters under a UTF-8 locale and bytes under a C locale, so it silently answers whichever question the environment happens to ask. Read the byte count from the raw bytes and the character count from the decoded string.

2. Read the two limits from the repo, never from memory.

   grep max_line_length .editorconfig
   grep printWidth .prettierrc
   

3. Confirm the difference is the encoding. If bytes - chars equals the number of Cyrillic characters on the line, the overrun is byte inflation, not width.

4. Ask Prettier who owns the line. Run the repository's check-only formatter target. Where no check-only target exists, run the target mapped by make.format (skip with a recorded note when it maps to null) and confirm it leaves the line untouched. If Prettier reports the file as formatted, it keeps the line single-line by choice and any manual wrap is undone on the next format run.

   make lint-prettier    # a React SPA whose check-only Prettier target is `lint-prettier`
   make format-check     # a component library whose check-only Prettier target is `format-check`
   make format           # no check-only target: run the formatter and confirm no diff
   

5. Classify and act. Characters within printWidth while bytes exceed max_line_length is a byte-count artefact: reply with both numbers and leave the line. Characters over printWidth is a real overrun: restructure the line and re-run the gate.

6. Check the mode before skipping. mode = "comment" in .qlty/qlty.toml means the checker is advisory and the character-count gates decide. If the checker is ever wired as a blocking gate, the fix is to shorten the line for real — extract the literal to a named constant or an i18n key — not to relax the limit.

Reply shape

State both numbers, name the encoding, name the owner: the line is N characters (within the configured printWidth) and M UTF-8 bytes because each Cyrillic character encodes as two; Prettier counts columns, reports the file formatted, and would revert a manual wrap.

Common mistakes

• Wrapping the line to silence the finding, then failing the formatting gate on the next run.
• Quoting only the byte number in a reply, which reads as an admission that the line is too long.
• Assuming the checker is advisory — read the mode from the qlty config in the repo at hand.
• Extending the diagnosis to a line whose character count really is over the limit; extract the long literal instead.