streetsidesoftware/cspell v10.5.0

on GitHub

feat: Skip nested git worktrees by default (#9604)

Summary

cspell now skips git worktrees inside the root folder, such as .worktrees/feature or .claude/worktrees/feature. A worktree is another checkout of the same repository, often of another branch, so its files are not the files being checked.

A new preset in ignorePathPresets, nestedWorktrees, can disable this.

Fixes #9515

Feature

A repository with a worktree of its feature branch, made with git worktree add worktrees/feature feature:

project/
  README.md
  worktrees/
    feature/
      README.md

Before, running cspell in project/ checked both copies of README.md:

$ cspell .
README.md:3:21 - Unknown word (finshed) fix: (finished)
worktrees/feature/README.md:3:21 - Unknown word (finshed) fix: (finished)
CSpell: Files checked: 2, Issues found: 2 in 2 files.

After, it checks only the repository's own README.md:

$ cspell .
README.md:3:21 - Unknown word (finshed) fix: (finished)
CSpell: Files checked: 1, Issues found: 1 in 1 file.

Running cspell in worktrees/feature/ still checks that worktree.

To check nested worktrees again:

cspell.json

{
  "ignorePathPresets": { "nestedWorktrees": false }
}

What the nestedWorktrees preset does:

• It is enabled by default.
• It skips the worktrees whether or not .gitignore excludes them, and with or without --gitignore.
• It finds the worktrees in the repository's own records, in .git/worktrees. Finding them does not search the worktrees' folders, so it stays fast however many worktrees there are.
• It skips only the worktrees of the repository that holds the root folder. Submodules and other repositories inside the root folder are still checked.
• A run started inside a worktree checks that worktree's files.
• When the path to the root folder goes through a symbolic link, it doesn't match the paths git recorded, so the worktrees are checked.
• A file in a nested worktree named on the command line is skipped too, unless --force-check is given, as with ignorePaths.

feat: Add ignorePathPresets and skip git's own files by default (#9600)

Summary

cspell now skips git's own files inside the root folder:

• the .git file that git puts in a worktree or a submodule
• the files inside .git folders

Git's message files, such as .git/COMMIT_EDITMSG, are still checked, so a commit-msg hook that runs cspell .git/COMMIT_EDITMSG keeps working.

A new setting, ignorePathPresets, can disable this.

Fixes #9590

Feature

cspell only reaches these files when dot files are checked (--dot or enableGlobDot), or when they are named on the command line.

Before, checking a git worktree with --dot reported the words in its .git file:

$ cspell --dot .
.git:1:1 - Unknown word (gitdir)
.git:1:31 - Unknown word (worktrees)
CSpell: Files checked: 2, Issues found: 2 in 1 file.

After:

$ cspell --dot .
CSpell: Files checked: 1, skipped: 1, Issues found: 0 in 0 files.

No ignorePaths glob could do this before: one that ignores the .git file also ignores .git/COMMIT_EDITMSG. Rules you already have for .git keep working.

To check git's own files again:

cspell.json

{
  "ignorePathPresets": { "git": false }
}

What the git preset does:

• It is enabled by default.
• It keeps git's message files: the files whose names end in MSG, such as COMMIT_EDITMSG, MERGE_MSG, and TAG_EDITMSG, and EDIT_DESCRIPTION. They are kept directly inside a .git folder, and inside a linked worktree's folder in .git/worktrees.
• It skips only files inside the root folder. A hook in a worktree or a submodule that passes the message file with --file keeps working: git keeps that file in the main repository's .git folder, outside the root.
• With the root inside a .git folder, the files there are checked.
• A config that sets ignorePathPresets only changes the presets it names. Presets from imported configs are combined.

feat: Add features.overrideGlobsMatchFilesOnly so override globs match only the path they name (#9598)

Summary

New feature option features.overrideGlobsMatchFilesOnly. When it is enabled, a filename glob in overrides matches only the path it names, not the files inside a folder with that name. It is disabled by default, so existing configs keep working.

See #9590.

Feature

Today an override glob also matches the files inside a folder with the same name. For example, examples.ts also matches every file inside a folder named examples.ts, and src/** also matches a file named src.

With the option enabled, each glob matches only what it names. To apply an override to the files in a folder, end the glob with /.

cspell.config.yaml

features:
  overrideGlobsMatchFilesOnly: true
overrides:
  # The files in every folder named `fixtures`.
  - filename: 'fixtures/'
    enabled: false

What a glob matches:

• The option can be set in an imported config. It applies to all overrides, including the ones from imported configs.
• The default is planned to become true in v11.

fix: Add a Vim file type for .vim files and .vimrc (#9579)

Adds a vim file type with the .vim extension and the filenames .vimrc, _vimrc, vimrc, .gvimrc and _gvimrc, as Jason3S confirmed on the issue. filetypes.test.ts checks each of them.

Closes #9489

fix: Check files in a folder whose name starts with a space (#9608)

Summary

Files in a folder whose name starts with a space can now be checked. Before, cspell found the file but dropped the space from its path, then reported it as not found.

This affects:

• globs, such as cspell "**/readme.txt" or cspell "./ notes/readme.txt"
• --file and --file-list entries, such as notes/readme.txt

Globs, such as those in ignorePaths and overrides, now match a file whose name ends with a space correctly. For a file named b (with a trailing space):

• b no longer matches it.
• b[ ] now matches it.

Before, with notes/readme.txt (the folder name starts with a space):

$ cspell --no-progress "**/readme.txt"
Linter: File not found: "/projects/demo/notes/readme.txt"
CSpell: Files checked: 0, skipped: 1, Issues found: 0 in 1 file with 1 error.

After:

$ cspell --no-progress "**/readme.txt"
 notes/readme.txt:1:12 - Unknown word (mispeling)
CSpell: Files checked: 1, Issues found: 1 in 1 file.

Entries in --file and --file-list keep their leading spaces. Trailing spaces and Windows line endings are still removed.

fix: Check folders named on the command line with # or a leading space (#9610)

Summary

A folder named on the command line is now checked when its name contains a # after a space, starts with #, or starts with a space. Before, cspell found no files in it, or checked a different folder.

cspell-glob also exports globInFolder, to make a glob that matches inside a folder. The folder's glob symbols are escaped.

Fix

A folder notes #2 with todo.md inside.

Before:

$ cspell "notes #2"
CSpell: Files checked: 0, Issues found: 0 in 0 files.
No files found to check.

After:

$ cspell "notes #2"
notes #2/todo.md:1:25 - Unknown word (finshed) fix: (finished)
CSpell: Files checked: 1, Issues found: 1 in 1 file.

Before, a folder named notes (with a leading space) checked the folder notes instead.

fix: Glob symbols in a config folder's name no longer match other folders (#9605)

Summary

When a config file sits in a folder whose name has glob symbols, such as docs (old), its files and ignorePaths globs could match files in other folders. The folder name was read as part of the glob, so docs (old) also matched docs old.

Folder names are now escaped before they are added to a glob.

cspell-glob also exports escapeGlobPath, to add a folder to a glob so that it matches only that folder.

Fix

Files:

• docs (old)/cspell.json: { "ignorePaths": ["*.md"] }
• docs (old)/a.md
• docs old/b.md

Before, docs old/b.md was skipped:

$ cspell --no-progress --config "docs (old)/cspell.json" "**/*.md"
CSpell: Files checked: 0, Issues found: 0 in 0 files.
No files found to check.

After:

$ cspell --no-progress --config "docs (old)/cspell.json" "**/*.md"
docs old/b.md:1:1 - Unknown word (helo) fix: (hello)
docs old/b.md:1:6 - Unknown word (wrold) fix: (world)
CSpell: Files checked: 1, Issues found: 2 in 1 file.

With { "files": ["*.md"] } in the same config, cspell used to check docs old/b.md instead of docs (old)/a.md. It now checks docs (old)/a.md.

fix: correct the ignorePaths default and its .git/COMMIT_EDITMSG advice (#9599)

Summary

The documentation for ignorePaths (hover text, JSON schema, website) described a default and a .git workaround that don't match what cspell does. This corrects it. Nothing about how cspell works has changed.

• The default is [], not ["node_modules/"].
• The cspell command line skips node_modules/** on its own, and only when no --exclude is given.
• The warning about .git/COMMIT_EDITMSG now suggests [".git/!(COMMIT_EDITMSG)"]. It ignores everything in .git except the commit message. The old suggestion still let files such as .git/config and .git/hooks/* be checked.

fix: allow globs with --file again (#9601)

Summary

Since v10.2.0, cspell lint refused globs combined with --file or --files, with error: mixing globs and --file is not supported. That broke command lines that worked before, including two examples in cspell lint --help. They work again: the globs filter the files given with --file, as they did before v10.2.0.

cspell "/*.md" --no-must-find-files --files $FILES

This checks only the Markdown files at the root among $FILES, and exits without an error when none match.

Globs combined with --force-check are still an error.

fix: Fix word splitter pruning for words following compound prefix joined by punctuation (#9573)

Summary

When segmenting identifiers with internal punctuation (., -, _), the word splitter previously pruned search branches against parentCost stored in knownPathsByIndex. Because knownPathsByIndex holds the cost of the first path that reached the index rather than the optimal suffix cost, a later branch exploring a compound or sub-segmented dictionary word (such as max_redirections after HTTPRedirectHandler.) was prematurely skipped.

Only prune punctuation-separated breaks against parentCost when parentCost === 0 (which cannot be improved upon), allowing valid dictionary words following punctuation separators to be matched.

Fixes #9469

fix: Stop skipping files like console.log.ts as log files (#9589)

Summary

File type patterns that contain * matched anywhere in a file name instead of the whole name, so some files got the wrong file type and were skipped or misread:

• console.log.ts was treated as a log file, so cspell lint skipped it, even with --force.
• cspell.env.config.js was treated as a shell script.
• README.ignore.md was treated as an ignore file.

Each of them now gets the file type of its extension. Patterns now have to match the whole file name, as in VS Code.

fix: Add a Vim file type for .vim files and .vimrc (#9579)

Adds a vim file type with the .vim extension and the filenames .vimrc, _vimrc, vimrc, .gvimrc and _gvimrc, as Jason3S confirmed on the issue. filetypes.test.ts checks each of them.

Closes #9489

fix: Accept long words with many alternate characters (#9574)

Summary

Fixes long words being reported as unknown even though the dictionary accepts their spelling. This could happen when a dictionary's repMap allowed alternate characters in many places in a word, because cspell only tried some of the combinations.

fix: Stop cspell trace crashing on an empty word with compound words (#9576)

Summary

cspell trace no longer crashes when it is given an empty word and compound words are allowed. It now reports the word as not found in every dictionary, the same as any other unknown word.

cspell trace "" --allow-compound-words

Before:

TypeError: Cannot read properties of undefined (reading 'length')

After:

Word F Dictionary           Dictionary Location
     - [flagWords]*         From Settings `flagWords`
     - [ignoreWords]*       From Settings `ignoreWords`
     ...

Code that uses cspell-trie-lib directly gets the same fix: looking up an empty word with legacy compounds now returns a not-found result instead of throwing.

fix: Avoid slow checks on non-ASCII words with many repMap matches (#9564)

Summary

Checking a non-ASCII word no longer takes seconds when a dictionary's repMap has many matches for that word.

• With the nn-no-alt dictionary, é followed by 12 as took about 3 seconds. It now takes a few milliseconds.
• For non-ASCII words, the result of a replacement is no longer replaced again. With a → b and b → c, éa no longer matches éc. ASCII words already work this way.

fix: Avoid running out of memory on words with many repMap matches (#9543)

Summary

Checking a word with many letters that match a dictionary's repMap no longer runs out of memory or fails with a RangeError.

• Words with long runs of characters matching single-letter or multi-letter repMap entries (such as in @cspell/dict-da-dk, @cspell/dict-de-de, or custom repMap definitions) no longer crash cspell.
• When a word has more than 256 replacement forms, only a sample of them is looked up, so a correct word can be reported as unknown. With nn-no-alt, this affected 65 of 34,077 words in a document with every æ, ø, and å written as ae, oe, and aa.

Related to #9535