Old codebase, outdated tool chain - how to limit the blast?

I have a 5-6 year-old enterprisey React codebase at work with a bunch of dependencies that for the most part haven't been upgraded since the day they were first installed. Most of the build chain is two+ major versions behind current: webpack 3, babel (& all loaders/plugins) v6, from before the monorepo reorg, Jest v23, and Flow 🤦 version 0.37. For those who haven't been keeping up with Flow (and who has?) the current version is 0.195.0. It's so old that the flow-language-server in vs code won't start, and the Stripe flow-to-typescipt-codemod works only partially and leaves several thousand compiler errors that it can't address. Oh, and it's also an ejected create-react-app codebase with a lot of custom build scripts and a really bespoke end-to-end testing setup that I don't pretend to understand. Anyone who understood this codebase has long since moved on. Right now I'm as expert as anyone else here, and I have the team's blessing to take the reins and upgrade this all to a modern stack. But as I experiment with it, it's a pile of pickup sticks, because anything I pull on seems to require me to upgrade everything else. Target state is Typescript replacing Flow, and either Vite and Vitest replacing the rest of the toolchain— or if that's impossible, at least modern versions of Webpack and Jest. Question for the channel, as per title: is there any way to avoid a big rewrite that takes a month and breaks a bunch of things I don't know about? Anybody have experience with something similar?
GitHub
GitHub - stripe-archive/flow-to-typescript-codemod: Codemod Stripe ...
Codemod Stripe used to migrate 6.5m+ lines of code from Flow to TypeScript - GitHub - stripe-archive/flow-to-typescript-codemod: Codemod Stripe used to migrate 6.5m+ lines of code from Flow to Type...
11 Replies
Unknown User
Unknown User•2y ago
Message Not Public
Sign In & Join Server To View
benten
benten•2y ago
I'd probably just rewrite it and save the headache.
Unknown User
Unknown User•2y ago
Message Not Public
Sign In & Join Server To View
Ambushfall
Ambushfall•2y ago
Highly depending on how much money this current thing is making for the company, if it isn't profitable now, chances are it won't be profitable after.
cje
cje•2y ago
if you want to be working in a clean react codebase with good tooling a month from now, my advice would be to send out some CVs ive been in a similar situation before minus flow my experience was the same, so many pieces are connected that you get to a point where you cant really see if it works or not before you're done porting the entire thing sorry i don't really have much good advice
paul
paulOP•2y ago
Thanks for the suggestions all. Rewrite (of the build scripts, that is, not the whole app) may truly be the only option, but I can't call it "saving the headache". @lee do you have experience with what you outlined? The way I read it, in the middle of the port I would have two React apps, each with part of the functionality, which wouldn't really give me a way to deploy my changes midway. Might as well be a full rewrite. It's not a money-making tool, but it's very valuable. I thought by saying it was 5-6 years old I was implying that the company has kept using it for that long. We wouldn't have done that if it didn't do a useful job
kwargs
kwargs•2y ago
I work on a similar project, with the difference being it's written in Javascript. There are some requirements to change the authentication and since it's all mangled up some parts need to be rewrited. My first step was to get the hang of the overall structure of the code, add todos for very messed up things and add type annotations with JSDoc. Starting with low hanging fruits and seperating stuff to be able to later switch out some parts seems the way to go. I would refrain from a rewrite, it's probably not a fruitful endeavor for a complex application. And I would not migrate from one library to another if there is no very big value to gain from doing so. For example migrating away from jest might not be the best idea. It's pretty stable at this point and it works.
paul
paulOP•2y ago
I like my tests to run fast.
kwargs
kwargs•2y ago
Yeah who doesn't but that's not a good enough reason if it does not cause a huge problem.
paul
paulOP•2y ago
True(*) but that one's the easiest to upgrade. You can have tests that run in jest and tests that run in vitest, and migrate over time. Nor is the interface difference a big change. Bigger worry is Enzyme is basically unsupported on React 17, and completely unsupported in 18, so there will be a lot of test rewriting where we use it. * Edit: actually no, dev workflow speed is a good reason to change most anything. Not a good reason to make unsafe changes, not a good reason to sacrifice customer experience (outside of a startup), but speed of feedback is the root of pretty much every good technical practice. We need to take it seriously and not treat it as the luxury of greenfield teams
robotkutya
robotkutya•2y ago
my 2022 was wasted on something similar project is in a pretty OK-ish place now, but it was not woth the time and effort to do it, would have made so much more sense to just rewrite from scratch, but hey, it is what it is this is one of those cases, when I think it makes a lot of sense to start writing lots of E2E tests it makes all the refactorings a whole lot easier think of creating a mold via the E2E tests, and then pour the new app into it so my strategy would be to 1) replace the bespoke E2E test setup with something modern, add bunch of tests to cover all the critical product stuff and edge cases 2) replace build tools --> move to vite 3) establish a new feature based project structure (something like bulletproof react https://github.com/alan2207/bulletproof-react ) 4) rewrite stuff route by route / feature by feature
Want results from more Discord servers?
Add your server