Node.js .gitignore template

This is the template people reach for when they want coverage rather than precision. It carries node_modules twice, which is harmless, and adds jspm_packages, bower_components and web_modules for repositories that predate the current tooling. The reasoning is the same as always: the dependency tree is restored from the lockfile, and the lockfile is the file you commit.

The log rules are unusually thorough because each package manager names its crash log differently: npm-debug.log, yarn-debug.log, yarn-error.log, lerna-debug.log and pnpm-debug.log, all with a trailing wildcard to catch the numbered variants. The report.*.json pattern catches the diagnostic report the runtime writes when it hits an unrecoverable error, which includes the full environment and command line.

The environment rules cover .env, .env.test and .env.production. Notably they do not cover .env.local or .env.development, so a project using those names needs to add them. This is the most common gap people hit with this template, and it is easy to miss because the file that gets committed is usually the one that was named slightly differently.

The cache rules span bundlers and linters: .cache, .parcel-cache, .npm, .eslintcache, .stylelintcache, .nyc_output and coverage. The .serverless, .fusebox and .dynamodb entries cover deployment and local emulation tooling. .tern-port holds the port number of a running editor language server, which is a good illustration of how much local runtime state ends up in a project root.

The Yarn block is the newest addition and the one that trips people up. .yarn/cache, .yarn/unplugged, .yarn/build-state.yml and .yarn/install-state.gz are ignored, along with .pnp.*. Whether that is correct depends on which install mode you use, because the zero install workflow deliberately commits the cache so a clone needs no network at all.