This project was a crutch to work around a limitation in old versions of NPM. Since the introduction of the prepare hook, it should not be needed anymore.
Clones and (re)installs packages from remote git repos. It is meant as a temporary solution until npm/npm#3055 is resolved.
npm install --save npm-git-installIn your package.json add:
{
"scripts": {
"install": "npm-git install"
}
"gitDependencies": {
"private-package-name": "[email protected]:user/repo.git#revision",
"public-package-name": "https://github.com/user/repo.git#revision"
}
}Obviously replace *-package-name and git URLs with values relevant to your project. URLs has to be in canonical form (i.e. one that you would provide to git clone on command line) - no fancy NPM shortcuts like user/repobitbucket:user/repo
Then install your dependencies as usual:
npm installIf you want to lock versions of git dependencies, use:
./node_modules/.bin/npm-git install --saveThis will reinstall all git dependencies, but also write last matching commit's sha to package.json, effectively locking the versions.
You can also add a dependency and lock it's sha in one go:
./node_modules/.bin/npm-git install --save [email protected]:me/my-awesome-thing.gitThis is probably the safest option, as it guarantees the same revision to be installed every time.
Now it should be easy to deploy, as long as the git executable is available in the environment.
IMO there is a serious defect in current versions of NPM regarding installation process of dependencies from git repositories. It basically prevents us from installing anything that needs a build step directly from git repos. Because of that some authors are keeping build artifacts in the repos, which I would consider a hurdle at best. Here is relevant issue with ongoing discussion.
If you npm install ../some-local-directory/my-package then npm will run prepublish script of the my-package and then install it in current project. This is fine.
One would expect that running npm install git@remote-git-server:me/my-package.git would also run prepublish before installing. Unfortunately it won't. Further more, it will apply .npmignore, which will most likely remove all your source files and make it hard to recover. Boo...
npm-git installThis simple script will do the following for every <url> of gitDependencies section of package.json:
-
Clone the repo it into temporary directory
using
git clone <url>. -
Run
npm installin this directorywhich will trigger
prepublishhook of the package being installed. -
then run
npm install <temporary directory>in your project path.
In effect you will get your dependency properly installed.
You can optionally specify different paths for package.json:
npm-git install -c git-dependencies.jsonYou may want to do this if you find it offensive to put non-standard section in your package.json.
Also try --help for more options.
Just like with plain NPM, on the command line you can specify a space separated list of packages to be installed:
npm-git install https://github.com/someone/awesome.git [email protected]/me/is-also-awesome.git#experimental-branchAfter hash you can specify a branch name, tag or a specific commit's sha. By default master branch is used.
With --save option it will write the sha of tha HEAD (i.e. last matching commit) to the package.json, effectively locking the version of the dependency.
You can also use it programmatically. Just require npm-git-install. It exposes four methods:
-
discover (path)Reads list of packages from file at given path (e.g. a package.json) and returns array of
{url, revision}objects. You can supply this toreinstall_allmethod. -
reinstall_all (options, packages)Executes
reinstallin series for each package inpackages. Options are also passed to eachreinstallcall.This function is curried, so if you provide just
optionsargument you will get a new function that takes only one argument -packagesarray.Options are the same as for
reinstall.Returns a
Promisethat resolves toreport, i.e. an array ofmetadataobjects that you can pass tosave. See below. -
reinstall (options, package)Most of the heavy lifting happens here:
-
Clones the repo at
package.url, -
Checks out
package.revision -
runs
npm installat cloned repos directory -
installs the package from there.
Options are:
silent: Suppress child processes standard output. Boolean. Default isfalse.verbose: Print debug messages. Boolean. Default isfalse.
Returns a
Promisethat will resolve to ametadataobject:{ name: "my-awesome-thing" sha: "ef88c40" url: "[email protected]/me/my-awesome-thing.git" }You probably don't want to use it directly. Just call
reinstall_allwith relevant options. -
-
save (file, report)Takes a path to a package.json and an array of
metadata(e.g. areportpromised byreinstall_all). Updates the contents of the package.json file according to the report.Returns
undefined.
If you are a Gulp user, then it should be easy enough to integrate it with your gulpfile. See [./src/cli.coffee][] for example use of the API.
I tried and it's hard, because NPM supports fancy things as Git URLs. See messy-auto-discovery branch. You are welcome to take it from where I left.
There is also another reason. User may not want to reinstall all Git dependencies this way. For example I use gulp version 4, which is only available from GitHub and it is perfectly fine to install it with standard NPM. I don't want to rebuild it on my machine every time I install it. Now I can leave it in devDependencies and only use npm-git-install for stuff that needs it.
React
Typescript
Django
Laravel
Facebook
Microsoft
Google
Alibaba
D3
Tencent