Why does, for example, a Grunt plugin define its dependency on grunt as "peer dependencies"?
Why can't the plugin just have Grunt as its own dependency in grunt-plug/node_modules?
Peer dependencies are described here: https://nodejs.org/en/blog/npm/peer-dependencies/
But I don't really get it.
Example
I'm working with AppGyver Steroids at the moment which uses Grunt tasks to build my source files into a /dist/ folder to be served on a local device. I'm quite new at npm and grunt so I want to fully comprehend what is going on.
So far I get this:
[rootfolder]/package.json tells npm it depends on the grunt-steroids
npm package for development:
"devDependencies": {
"grunt-steroids": "0.x"
},
Okay. Running npm install in [rootfolder] detects the dependency and installs grunt-steroids in [rootfolder]/node_modules/grunt-steroids.
Npm then reads [rootfolder]/node_modules/grunt-steroids/package.json so it can install grunt-steroids
own dependencies.:
"devDependencies": {
"grunt-contrib-nodeunit": "0.3.0",
"grunt": "0.4.4"
},
"dependencies": {
"wrench": "1.5.4",
"chalk": "0.3.0",
"xml2js": "0.4.1",
"lodash": "2.4.1"
},
"peerDependencies": {
"grunt": "0.4.4",
"grunt-contrib-copy": "0.5.0",
"grunt-contrib-clean": "0.5.0",
"grunt-contrib-concat": "0.4.0",
"grunt-contrib-coffee": "0.10.1",
"grunt-contrib-sass": "0.7.3",
"grunt-extend-config": "0.9.2"
},
The "dependencies" packages are installed into [rootfolder]/node_modules/grunt-steroids/node_modules which is logical for me.
The "devDependencies" aren't installed, which I'm sure is controlled by npm detecting I'm just trying to use grunt-steroids
, and not develop on it.
But then we have the "peerDependencies".
These are installed in [rootfolder]/node_modules, and I don't understand why there and not in [rootfolder]/node_modules/grunt-steroids/node_modules so that conflicts with other grunt plugins (or whatever) are avoided?
TL;DR: peerDependencies
are for dependencies that are exposed to (and expected to be used by) the consuming code, as opposed to "private" dependencies that are not exposed, and are only an implementation detail.
The problem peer dependencies solve
NPM's module system is hierarchical. One big advantage for simpler scenarios is that when you install an npm package, that package brings its own dependencies with it so it will work out of the box.
But problems arise when:
Both your project and some module you are using depend on another module.
The three modules have to talk to each other.
In Example
Let's say you are building YourCoolProject
and you're using both JacksModule 1.0
and JillsModule 2.0
. And let's suppose that JacksModule
also depends on JillsModule
, but on a different version, say 1.0
. As long as those 2 versions don't meet, there is no problem. The fact that JacksModule
is using JillsModule
below the surface is just an implementation detail. We are bundling JillsModule
twice, but that's a small price to pay when we get stable software out of the box.
But now what if JacksModule
exposes its dependency on JillsModule
in some way. It accepts an instance of JillsClass
for example... What happens when we create a new JillsClass
using version 2.0
of the library and pass it along to jacksFunction
? All hell will break loose! Simple things like jillsObject instanceof JillsClass
will suddenly return false
because jillsObject
is actually an instance of another JillsClass
, the 2.0
version.
How peer dependencies solve this
They tell npm
I need this package, but I need the version that is part of the project, not some version private to my module.
When npm sees that your package is being installed into a project that does not have that dependency, or that has an incompatible version of it, it will warn the user during the installation process.
When should you use peer dependencies?
When you are building a library to be used by other projects, and
This library is using some other library, and
You expect/need the user to work with that other library as well
Common scenarios are plugins for larger frameworks. Think of things like Gulp, Grunt, Babel, Mocha, etc. If you write a Gulp plugin, you want that plugin to work with the same Gulp that the user's project is using, not with your own private version of Gulp.
I would recommend you to read the article again first. It's a bit confusing but the example with winston-mail shows you the answer why:
For example, let's pretend that winston-mail@0.2.3 specified "winston": "0.5.x" in its "dependencies" object because that's the latest version it was tested against. As an app developer, you want the latest and greatest stuff, so you look up the latest versions of winston and of winston-mail and put them in your package.json as { "dependencies": { "winston": "0.6.2", "winston-mail": "0.2.3" } } But now, running npm install results in the unexpected dependency graph of ├── winston@0.6.2 └─┬ winston-mail@0.2.3 └── winston@0.5.11
In this case, it is possible to have multiple versions of a package which would cause some issues. Peer dependencies allow npm developers to make sure that the user has the specific module (in the root folder). But you're correct with the point that describing one specific version of a package would lead to issues with other packages using other versions. This issue has to do with npm developers, as the articles states
One piece of advice: peer dependency requirements, unlike those for regular dependencies, should be lenient. You should not lock your peer dependencies down to specific patch versions.
Therefore developers should follow semver for defining peerDependencies. You should open an issue for the grunt-steroids package on GitHub...
multiple versions of a package which would cause some issues
but isn't that the whole point of a package manager? They even discuss this further up in the same article where there are 2 versions of the same package in the project: one provided by the developer and one supplied by a 3rd party library.
winston
example am I now just unable to use the winston-mail
library because my version does not match the peer dependency? I would much rather have that temporary downgrade from latest and greatest for the 1 library than to not be able to use it at all.
peerDependencies
explained with the simplest example possible:
{
"name": "myPackage",
"dependencies": {
"foo": "^4.0.0",
"react": "^15.0.0"
}
}
{
"name": "foo"
"peerDependencies": {
"react": "^16.0.0"
}
}
running npm install in myPackage will throw an error because it is trying to install React version ^15.0.0
AND foo
which is only compatible with React ^16.0.0
.
peerDependencies are NOT installed.
foo
works with both React 15 and React 16 then it could list its peerDependency as >=15 < 17
.
Success story sharing
"grunt": "0.4.4"
is both in devDependencies and peerDependencies, and it does make sense to me to have a duplicate there, because it means both that I need thatgrunt
package for my own use, but also that the users of my library can use their own version, as long as it respects the peerDependencies version lock. Is that correct? Or is the OP example a very bad one?JacksModule
depends onJillsModule ^1.0.0
withJillsModule
being a peer dependency ofJacksModule
andYourCoolProject
were usingJacksModule
andJillsModule ^2.0.0
, we will get the peer dependency warning by NPM, which will advise us to installJillsModule ^1.0.0
as well. But what happens then?YourCoolProject
will now have two versions ofJillsModule
importable throughimport jillsModule from "..."
? And how do I remember that when I useJacksModule
I need to pass it an instance ofJillsModule v1.0.0
?JacksModule
repo, upgrade it to depend onJillsModule ^2.0.0
and offer a PR to the project maintainer. It may help to submit a bug first saying this dependency is outdated and you would like to help update it. If you make a good PR, most library maintainers will merge it and thank you for it. If maintainers are unresponsive, you can publish your fork to NPM namespaced under your name and use your fork instead. In any way, there are solutions butpeerDependencies
does not solve it on it's own.