Repository navigation
Testing and Development
Run all commands from the repository root.
Missing dependencies are downloaded into gitignored .tests/:
nvim --headless -u tests/minimal_init.lua \
-c "PlenaryBustedDirectory tests/ {minimal_init = 'tests/minimal_init.lua', sequential = true}"Set NVIM_TEST_DEPS to a directory containing existing plenary.nvim and
nvim-web-devicons checkouts. This mode does not download missing dependencies —
they fail instead:
export NVIM_TEST_DEPS="$HOME/.local/share/nvim/lazy"
nvim --headless -i NONE -n -u tests/minimal_init.lua \
-c "PlenaryBustedDirectory tests/ {minimal_init='tests/minimal_init.lua', sequential=true}"nvim --headless -i NONE -n -u tests/minimal_init.lua \
-c "PlenaryBustedDirectory tests/multiline_layout_spec.lua {minimal_init='tests/minimal_init.lua', sequential=true}"Use Neovim 0.12+ when testing multiline.
The documentation pipeline has its own stdlib-only tests:
python3 scripts/gen_help.py --check # generated help is in sync
python3 scripts/test_docs.py # drift, link and image parity
python3 scripts/test_wiki_sync.py # wiki publishing safetyImages in wiki/assets come from real Neovim sessions, never edited by hand. Each scene in
scripts/preview/scene.lua runs in a private tmux server; ansi2html.py frames the pane
as HTML:
NVIM_TEST_DEPS="$HOME/.local/share/nvim/lazy" scripts/preview/render-all.sh /tmp/previewsScreenshot each page's #shot element at device scale factor 2 (for example with
Playwright) and save it in the wiki/assets directory, named after the scene. Fonts:
FiraCode Nerd Font Mono and Symbols Nerd Font Mono. scripts/test_docs.py requires every
capture to be at least 2000 px wide and under 600 KiB.
- A plain
nvim +lua ... +qacan exit zero despite a Lua error. Use an explicitcquit 1failure path. - Run the complete suite after focused tests, and inspect errors as well as exit status.
- Never load the live user config for automated tests; test state is redirected into
.tests/. - Check formatting with
stylua --check --config-path=stylua.toml lua/when StyLua is available, and rungit diff --check.
Contributors and AI coding agents should read AGENTS.md for architecture, compatibility boundaries, ownership safety and PR workflow.
Start with the guide's read-only preflight and task map. It points from setup, layout,
rendering and input tasks to the matching modules and regression suites. CLAUDE.md
loads that shared guide; keep agent rules there rather than copying them into several
provider-specific instruction files.
For prose-only changes, run the three documentation checks above and git diff --check.
For Lua changes, add a failing regression, run the focused suite, then the full suite.
UI and input changes also need isolated interactive testing, not just headless tests.
Keep machine-specific paths, downloaded dependencies and test output out of commits.
README files are bilingual feature introductions. Put detailed settings and runnable
examples in the paired wiki pages, and reference behavior in wiki/help/. Every push
to main mirrors the Markdown pages from wiki/ to GitHub's wiki after validation;
local edits alone do not update it. See Documentation Pipeline.
English
Getting started
Reference
Development