I am trying to use Yarn Workspaces for my app that I am splitting into multiple packages such that I can share code between the mobile and web version of the app. Let me explain what I am trying to do with an example.
Let's say I currently have a mobile app called awesome-app. I am refactoring it by taking out the shared code and creating three packages as follows:
awesome-web
awesome-mobile
awesome-shared
Let's say I want to add new functionality to awesome-mobile for which I have to install depA to awesome-mobile. How can I do that such that yarn only installs depA and updates the package.json for the awesome-mobile. I tried using command yarn package <package-name> add depA but it ended up installing all the dependencies again which I want to avoid.
Also, let's say I want to use awesome-shared in awesome-web. Is there a yarn command such that it installs and updates the package.json for awesome-web automatically. Currently, I do it by hand and then I do yarn install in the root folder which ends up installing all the dependencies again.
In my monorepo, I have 3 packages package1, package2, package3, each package contains a npm script named build.
However, these packages are not linked together. I.e. there are no require() in any of those packages linking to a sibling package.
From the root folder, I run lerna run build. It seems to run build of the packages in the alphabetically order.
Is there a way to specify the order to run the build commands of these packages?
--sort won't work because they are not linked.
You don't specify the order, you specify the topology by including a package as a dependency of another.
If package1 needs to be built before package2 you add package1 to the dependencies of package2 in the latter's package.json file. If you do not want package2 to directly depend on package1 (e.g. on production) you can still add it into devDependencies and Lerna will understand the dependency.
From lerna -h:
--sort Sort packages topologically (dependencies before dependents).
Pass --no-sort to disable. [boolean] [default: true]
Note Some commands can be ran ignoring this topology, for example from lerna exec's --parallel option documentation:
completely disregards concurrency and topological sorting
lerna run build --include-dependencies --stream
--include-dependencies this flag can help
You can first build your shared package and after that trigger another build.
Ex :
yarn workspace #shared run build && yarn lerna run build
I have been using npm for a personal project and just recently stumbled across yarn. Would there be any harm or "intended side effects" to switching to yarn's package manager in the same project where I had been using npm?
Although a few commenters here say its ok to mix both yarn and npm on the same project, after using yarn and npm and then yarn again, this is what yarn has to say about it:
warning package-lock.json found. Your project contains lock files generated by tools
other than Yarn. It is advised not to mix package managers in order to avoid resolution
inconsistencies caused by unsynchronized lock files. To clear this warning, remove
package-lock.json.
Since to me it is not any harm to using both them into one project.
I use npm and yarn (50/50) in dev environment.
But on ci/di i use only yarn because it is faster, and i reduce build minutes thanks yarn.
Also they both create different .lock file names.
Nobody told about the lock files.
Imagine you use yarn on dev environment, and yarn on your build/production servers. When you install a package using yarn, and your project works on your computer, you probably would want to keep it working on a production environment (your server).
That being sad, you would commit you yarn.lock file, that "saves" the exact versions of each package you have, when the project ran on your computer.
On your buid/production server you should call yarn install, but asking to keep all the same versions with --frozen-lockfile parameter. Some even say "yarn install --frozen-lockfile should be the default behavior", and I agree.
Then... another dev jump in the project you are working and install a package using npm (other than yarn). That new package will not be included in your yarn.lock file, but, a new package-json.lock file would be created, telling the exact packages versions it is using.
When that commit arrives on your build/production server, it will crash, fail, because that new package doesn't exist on yarn.lock file. Someone would need to pull that changes, call a yarn to install the dependences and update the lock file with the new package dependences, and push it again to the repo.
A quick point about using the lock file or not. If you call a 'yarn install' on your build/production server some weeks after the last install on your machine, the server would have many other new versions than your last "stable" version. It already happened to me many times.
I published recently the package-locks-checks, which help ensure you have not just one lock file but also locked each package version on your project.
There will be a point that one or both will no longer work and your project will be stuck at only using the existing lock file. Meaning, the issue probably will involve installation fails if you opt to reinstall without a lock file. And that also means failure to create a new lock file, so you are stuck with the existing one that you are trying to get rid off in the first place. We are actually encountering this issue in one of our projects. Because it is so big, no one tries to fix the issue and just rely on the existing lock file.
So, even if we say it's a rare case that it won't cause harm. Mixing npm and yarn should be avoided.
Here https://classic.yarnpkg.com/en/docs/migrating-from-npm/ we may find a confirmation that Yarn's resolution algorithm is compatible with NPM resolution algorithm.
Inside a npm project (with package.json) if you run yarn it will read your node_modules folder (using the resolution algorithm) and create a yarn.lock file with your project's locked dependency tree.
Based on that I assume that they are compatible inside the same project.
Update 30/04/2021
My original reply refers to yarn 1 (classic), although I've just created a React app with create-react-app tool and it creates the project's repository with package.json + yarn.lock by default. Again, another demonstration that it's fine (even with the warning mentioned by Dave Pile).
At the end of the day this is a matter of putting both together to work and checking yourself...
Plus you get a warning from yarn as Dave Pile said because we have to push *-lock.json files changes you have to consider using npm version >= 7 to make sure whenever you install packages by npm it will update your yarn-lock.json file too.
Because whenever you install the packages either by npm or yarn depends on what you have chosen for updating a dependency in the package.json (Using tilde ( ~ ) which gives you bug fix releases and caret ( ^ ) gives you backward-compatible new functionality) it will update you.lock file and since you have to push it might happen that you have different version of lock files.
I am just learning about yarn and npm.
I want to use a particular package that crashes my app. I found a project issue which looks like fix has been implemented and merged with master. A comment says npm package has not been updated yet.
I have added the package to my project:
yarn add react-widgets
I thought I could use yarn to add the package via git repository to get fixed version.
The package I want is
https://github.com/jquense/react-widgets
So I tried
yarn add https://github.com/jquense/react-widgets.git
I get error:
error Package "undefined#undefined" doesn't have a "name".
Firstly, is this error a problem with my use of yarn or a problem with react-widgets repository? I am assuming it should work to add it like this so please correct me if I am wrong.
Also, can I assume that if there is an updated npm package, that I will be able to update with yarn at that time?
I'm afraid it is impossible to do now. This git repo is actually several packages inside(main package is here) one repo and root package.json is not valid one - it doesn't have name description and this is fail reason. There is discussion about supporting this in yarn, but it hasn't been implemented yet(in npm too).
We have a project which have to be packaged as a zip so we can distribute it to our cliens. With the normal node_modules directory i have no problems. I just put the directory and the node.exe together in my project folder and can start our project on every other computer without installing node or running any npm command.
But now i have a dependecy on phantomjs which needs to be installed as a global package npm install -g phantomjs.
How do i pack modules like this into our project? I first thought of copying phantomjs into the local node_modules directory and set the path variable NODE_PATH to this directory. It doesn't find phantomjs.
Development and client platforms are both windows.
Well, generally it is fine to install global dependencies with the --save flag and call their bins like ./node_modules/phantomjs/bin/phantomjs /*now executes*/ (just as an illustrative example).
However, with with Phantom it's not that simple, since it's downloading binaries and/or even compiling. You would have three options:
ssh into target and just npm install -g phantomjs before or define it in a manifest e.g. Dockerfile just like that, if you are using containers.
Compile it from source as advised here.
If you are using the CLI, then just the --save approach.
So I hardly advise just making a Docker image out of it and ship it as tarball. You can't zip the platform dependent Phantom installation, unfortunately.
Also lots of dependencies like karma-runner-phantomjs look for the path of the global dependencies to resolve it for their use.