This item is a work in progress. It is subject to change radically at any time—up to and including being deleted entirely if I change my mind about it! Because it is a draft, it is not included in the site’s feeds; if you want to watch its progress, you will need to do that very old thing and bookmark this page and return to it over time until it is published.
Given its preliminary, work-in-progress status, I would appreciate it very much if you do not share this to social media, Hacker News, etc.—call it part of the contract for “working with the garage door open”. Thanks!
types or Ruby’s Sorbet etc.; I am talking to people who are already interested in adopting them.
- TODO: some variety of
strict: falseand/or convert-files-you-touch
- TODO: leaves-in,
Most developers (myself included, the first time I did this!) are very much tempted to do the “just convert a file when you touch it, in loose mode or with lots of
// @ts-expect-error and
any scattered around” thing — for at least the three following reasons:
That pattern usually works with other kinds of migrations.
It feels more tractable, in that you can just do it “as you go”.
It actually works pretty well for sufficiently-small codebases — it’s very good for <1,000LOC and pretty good for <10,000LOC.
Unfortunately, it is a trap. When working on a codebase of any significant size, it actually means that you will end up having to propagate changes to various files over and over and over again. (I have a mediocre illustration of this I made a while back that I will try to dig up.) Worse, you cannot rely on the things you have already converted actually being safe: They feel safer than JS types-wise because they are in TS… but they are not, because they have lots of
// @ts-expect-error and
any scattered around. It can end up being quite demoralizing and frustrating to have errors coming out of your “but we already converted this!” modules. It can be a super frustrating thing to explain to other stakeholders, too: Well, yeah, we converted this to TypeScript, but not all the way, so it still has these issues…
The “walk the dependency graph leaves-in with maximum strictness” approach avoids that set of problems entirely. There is no free lunch, though, of course. Avoiding those problems requires being a bit more disciplined — you need to carve out some dedicated time to do the work by tackling a couple modules each week or something like that. Ultimately, though, it means that you neither have to revisit already-converted modules nor have the extremely annoying experience of getting errors TS can and should catch coming from already(-but-only-partially)-converted modules.
One additional note here: you can sometimes do a mix of both of these approaches quite effectively. If you have a larger app/etc. broken into a set of smaller packages, you can do the “iteratively work within a small library” approach within the packages, while avoiding publishing the types until you get them to full strictness. That ends up having some of the advantages and disadvantages of both approaches.