// ZSH-MORE-COMPLETIONS — THE LARGEST CURATED ZSH COMPLETION CORPUS IN EXISTENCE // 47,393 FILES

Install

# Zinit (recommended) -- nocompletions ice keeps fpath order correct
zinit ice lucid nocompile wait'0e' nocompletions
zinit load MenkeTechnologies/zsh-more-completions

# Oh My Zsh -- needs explicit nested-dir fpath wiring before compinit
git clone https://github.com/MenkeTechnologies/zsh-more-completions \
    "$HOME/.oh-my-zsh/custom/plugins/zsh-more-completions"
# plugins+=(zsh-more-completions)
# plus the fpath snippet from the README -- OMZ does not add nested *src dirs

# Manual
git clone https://github.com/MenkeTechnologies/zsh-more-completions
# copy _* files from src/, more_src/ .. more_src8/, man_src/, override_src/, architecture_src/
# to somewhere on your fpath, then: autoload -Uz compinit && compinit

Why nocompletions matters under Zinit: by default Zinit symlinks every _* file it finds into one flat completions/ directory. That collapses the per-source-dir grouping and breaks override_src/ precedence (override files no longer win against the rest). The nocompletions ice tells Zinit to leave the files where they are; the plugin's own entrypoint splices the 12 directories onto $fpath in the correct order at source time.

Why Oh My Zsh needs extra wiring: OMZ adds a plugin's top-level directory to $fpath, but it does not descend into nested *src/ subdirectories. Without the fpath snippet from the README, compinit never sees the 47,316 files. OMZ also requires compinit to run twice — once in OMZ's own startup and once after this plugin sources — for the appended directories to take effect (documented inline in zsh-more-completions.plugin.zsh).

Boot sequence

# initialize compsys
autoload -Uz compinit && compinit

# faster boot: skip the security check and load the cached dump
compinit -C

The plugin itself never calls compinit — it only puts directories on $fpath. Choosing when (and how) to initialize completion is the user's job, so the corpus works with any init strategy: lazy, eager, or a cached ~/.zcompdump.

Directory layout

12 fpath-prefixed source directories. override_src/ is prepended so its 10 hand-maintained files win against everything else; the remaining 11 are appended in load order (first match wins for any given _NAME). The more_src split is purely a filesystem-performance decision — some tools degrade past ~10k entries per directory (find walks, git index renames, zsh's own dir scan), so the corpus is spread across siblings to keep each comfortably under the ceiling.

Per-directory totals come from scripts/print-repo-stats.zsh; they shift on every harvest run. The Engineering Report carries the full breakdown plus the fpath-order rationale.

Harvested ecosystems (sample)

Stats

Counts change constantly — do not edit totals by hand. From the repo root:

zsh scripts/print-repo-stats.zsh

Prints _-prefixed completion file counts per directory (src, more_src..more_src8, man_src, override_src, architecture_src), total_completions, and zunit_tests (number of @test entries under tests/). At the current HEAD that is 47,316 completion files and 23,112 tests.

Completion file shape

Every _NAME file follows the same shape: a #compdef NAME first line, a comment header carrying the upstream provenance URL + harvest timestamp, then an _arguments / _values body whose specs are derived from the upstream argument parser. There is deliberately no trailing compdef _NAME NAME call — that would force the file to evaluate at compinit time instead of being lazily autoloaded on first tab.

How the corpus is built (harvest pipelines)

Most files are auto-generated by 6 Python harvest scripts in scripts/ (3,361 LOC total). Each targets a different upstream registry, validates every capture, header-stamps provenance, records the stem in blacklist.txt, and appends an existence pin to the matching tests/t-*-existence*.zsh. These scripts are maintainer-only — the runtime is pure zsh and needs no Python.

Every capture passes the same gate before landing: dedup against src/ / more_src*/ / blacklist.txt; reject cobra runtime-dispatch stubs; strip any trailing compdef line; parse with zsh -nf; stamp the provenance header; record the stem in the blacklist; append the existence test. The blacklist.txt (46,937 entries) is the corpus's negative space — every name a harvester rejected, so re-runs converge instead of re-probing.

Override completions

override_src/ is prepended to $fpath so its 10 files win against the upstream zsh-completions corpus, OMZ-shipped completions, and the 11 sibling directories in this same plugin. These are hand-maintained, not harvested.

Testing & CI

The corpus is validated by a zunit suite of 23,112 @test cases across 21 tests/*.zsh files, plus 21 tests/*.sh doc / repo-hygiene gates. CI runs zunit tests/*.zsh on Ubuntu via .github/workflows/ci.yml.

Troubleshooting

Sibling plugins

Part of the MenkeTechnologies zsh plugin family — the MenkeTechnologiesMeta umbrella:

• zsh-better-npm-completion — cache-aware npm install completion
• zsh-cargo-completion — live cargo add / cargo install completion
• zsh-cpan-completion — live MetaCPAN cpan / cpanm completion
• zsh-dotnet-completion — dotnet CLI completion
• zsh-expand — spacebar expansion (11,683 tests)
• zsh-gem-completion — ruby gem install remote completion
• zsh-git-acp — 159 git aliases + add/commit/push keybindings
• zsh-git-repo-cache — index every git repo on disk
• zsh-learn — MySQL-backed learning collection / quiz
• zsh-nginx — nginx commands + service-wrapper aliases
• zsh-pip-description-completion — pip remote completion with descriptions
• zsh-sed-sub — in-place sed substitution on the command line
• zsh-sudo — ESC ESC to prepend sudo
• zsh-xcode-completions — xcodebuild / xcrun / swift completion