How Monolithic Repository in Open Source saved my Laziness
Fortunately, I noticed talk from Fabien about monorepo and Symfony. I said to myself: "I don't know a thing about it. Let's try it out. I can always return if it sucks."
I never did. Today I will show you why I see monorepo approach in open-source so awesome.
If you don't have 50 minutes to watch the talk (my case), here are 40 slides from it.
Fabien introduces a tool, that helps you with splits - splitsh. Do you understand it? Me neither. Splitsh is fast yet complex tool to maintain Symfony and Blackfire ecosystem. All we need is one git command.
But we'll get to that later. First things first.
Monorepo (for monolithic repository) is single repository, which contains code for group of packages (framework, application...).
Not only Symfony, but also:
Opposed to this is manyrepo approach (for many repositories), meaning every package is in his own repository.
It is used by:
and almost everybody else.
So how does this monorepo work?
Imagine flow of a oxygenated blood from heart to arteries. All blood that was in heart, will drift to all arteries.
From 1 heart to many arteries in one direction. Same relation is between 1 monolithic repository and many repositories.
This is related to commits and tagging:
- Every commit that does changes is Symfony\Console code in monorepo, is also in Symfony\Console manyrepo.
- Every tag in monorepo, is also in all manyrepos.
You have one repo to maintain.
- New package? Not a problem.
- Changing name of method that is used by 10 other packages? Easier than ever.
- Starting a framework? Ideal.
That's the way Nette, Symfony and Laravel started and grew so much, even if only one person was behind majority of commits.
It's easy to make changes that affect all packages - bump to PHP 7.1, Symfony 3.2 or Nette 3.0 is a matter of minutes. And I know it works on all packages. Before monorepo, I had to upgrade every package manually, which resulted in dissonance: one package used Symfony\Console 3.2, but other only 2.8 and it got messy for no reason.
Another thing I love is testing both monorepo (all packages together) and manyrepo level. Once I spend 5 hours fixing a bug in Symfony\Process. It was difficult to find, because all tests were passing. Even for testing just the Symfony\Process directory made it pass. But when I copied only the Symfony\Process code to the standalone directory, it failed. Few hours later I found out, it's somehow related to having Symfony\Stopwatch package. Yea, WTF.
That could be caught testing on monorepo level.
On the other hand, monorepo testing is also important. When Nette was split from monorepo to manyrepo only, all tests were passing packages were standalone. But in combination some of them didn't.
This is not issue of the code itself, but of the testing architecture.
When maintaining 15 own packages, ApiGen and co-maintaining few more repositories, I spent a lot of time by package management and not coding. It wasn't fun and I contributed less and less.
Again, it's a matter of project architecture rather than the code or maintainers.
These are the best for me, but there are many more described in those slides by Fabien.
/packages /Symplify /DefaultAutowire /src /tests composer.json
Inspired by Laravel, all we really need is one git command. To split:
git subsplit init [email protected]:symplify/symplify.git git subsplit publish --heads="master" packages/DefaultAutowire:[email protected]:Symplify/DefaultAutowire.git rm -rf .subsplit/
It says: take code from
packages/DefaultAutowire directory and put it to
[email protected]:Symplify/DefaultAutowire.git repository.
As a result, we have this:
/src /tests composer.json
That's it! Could it be simpler?
I recommend to read Monolithic Repositories with PHP and Composer by Benjamin Eberlei (Doctrine Core Maintainer). It's surpassed by composer local packages, but points remain the same. Don't worry, I will write about local packages in the near future.
If you like monorepo approach, but prefer own tagging per package, Martin Zlámal recently wrote about git submodule approach (in Czech).
This is my point of view.
Do you maintain lot of repositories? How do you make it fun to code and care for all of them?
Let me know in the comments. Thank you.