The motivation behind mondorepo
The Node.js package ecosystem has been traditionally developed following the one package per repository rule, which is a workable solution for developing packages that are small in size or complexity and live in relative isolation.
As projects’ scale increases this approach has a couple significant problems.
- The amount of code in a complex project often increases beyond what would ideally live in a single package.
- Most times, multiple teams of developers need to collaborate on concurrently developed packages managed in separate repositories.
Traditional approaches
Traditional approaches to solve these problems are:
- Use relative paths all across the project
- Include jointly developed packages inside the main project’s
node_modulesdirectory. - Following the
monorepoapproach (mononotmondo).
Using relative paths across a huge codebase
The most common approach to develop a big project is to rely on the usage of relative paths all across the project’s modules.
Inevitably, this means that all modules have a bunch of require statements
that look like this:
var myLib = require('../../../lib/my-lib');
var otherLib = require('../../../lib/other-lib');
While this is a common way to approach the problem, it is very difficult to manage because it requires developers to keep track of these relative paths all across the project.
Developing from inside node_modules
Some projects may choose to jointly develop a private subpackage by placing its sources
directly inside the main project’s node_modules directory.
An example of this approach would look like the following:
Repository: 'awesomecorp/MyAwesomeProject'
MyAwesomeProject/
index.js // <- contains "require('my-pkg')"
node_modules/
my-pkg/
//sources for my-pkg live here and are added to source control
The hypothetical developers of this example chose to jointly develop my-pkg and AwesomePackage
from the same repository. This is problematic because:
node_modules/my-pkghad to be force-added to the repository (asnode_modulesis usually gitignored)npm installneeds to be run both in the root ofAwesomePackageandnode_modules/my-pkgas part of setting up the project.- The removal of the
node_modulesis a common practice as part of a clean build process, a developer unfamiliar with the project may inadvertently remove this directory and download an outdated copy from the registry, causing all sorts of difficult-to-track-down errors.
This approach is clearly problematic, which is why several complex projects follow the
monorepo structure (detailed below).
monolithic repos (monorepos)
Larger projects tend to adopt an approach where multiple packages live inside a
single repository. The most well-known example of such a project is perhaps Babel.
The generalization of this approach is often called a monorepo (note, no d on this
one - mono not mondo).
Such a project looks like this:
Repository: 'awesomecorp/MyAwesomeProject'
MyAwesomeProject/
MyAwesomeProjectMain/
index.js // <- contains "require('my-pkg')"
my-pkg/
//sources for my-pkg live here
my-other-pkg/
// sources for my-other-pkg live here
Without the use of external tools, the first problem with this approach is that a require
statement would not be able to find any of the subpackages as they’re not located inside
the main project’s node_modules directory. Manual solutions like symlinking them can
provide some aid but they are vulnerable to similar problems as the previous approach.
There are several npm packages available that attempt to resolve this problem as well
as provide management for these monorepos. The most widely used is Lerna.
Even with the help of a external tool, one of the main downsides of this approach is that to work on a single package you need to download the whole repository which includes all other packages. This strongly resembles the old development approach of monolithic applications.
Our approach
While there are potential advantages to each of these approaches, here at Sencha we decided to tackle the problem in a way that projects remain modular and sub packages can be developed on their own.
We call these mondorepos.
mondorepos (“mondo: large, big”)
As an alternative to monolithic repositories, mondorepo enables teams to collaborate
on big complex projects that span across multiple repositories. Each subpackage can be a
mondorepo on its own and so on.
An example of this project structure would look like this:
Repository: 'awesomecorp/MyAwesomeProject'
MyAwesomeProject/
index.js
package.json // <- contains a reference to "awesomecorp/My-pkg" under "mondo.uses.My-pkg"
Repository: 'awesomecorp/My-pkg'
MyAwesomeProject/
index.js
package.json
Running mondo install will connect all used repositories (declared inside mondo.uses) and make My-pkg
available to be used on a simple require('My-pkg') statement, isn’t that neat?
This effectively means that each subpackage can be developed on its own if needed, but also can be included as part of any other project that wants to jointly develop a bunch of its own requirements.