Type-Safe Printf() in TypeScript
typescriptlang.orgWord of warning: the typescript compiler is not a particularly fast evaluator of recursive list manipulation programs, which is what these kinds of types are.
They’re great in small doses where you really need them, but overuse or widespread use of complex types will make your build slower. It’s much better to avoid generics or mapped types if you can. The typings for a tagged template literal (without digit format specifiers like %d4) don’t require any generics.
I love to write code like this, but I’m guilty of over using fancy types and I flinch when I see a typescript build profile showing 45s+ spent on generic types I wrote without realizing the cost.
The nature of ts also means that if you make your files slower to build via type/list nonsense, the language server is going to bog down and the editing experience will mysteriously become bad for everyone. Strongly discourage doing slow stuff with the type system.
While I certainly agree, I've found that this is often an indication of too-complex an architecture, and a fundamental re-think being necessary. I've had projects that depend on [fp-ts], which end up incredibly generic-heavy, but still make it entirely through a typecheck(not build- typescript's just worse at that than other tools like esbuild) in seconds-at-worse.
Obviously depends on your organization/project/application, but I do like these things as complexity-smells.
[fp-ts]: https://gcanti.github.io/fp-ts/
How large in lines of typescript are the projects you've used fp-ts or similar with?
We have about 3 million; when I discuss a slow type, i mean a type that contributes ~1 min of checking or more across all the uses in 3 million lines, analyzed from a build profile using Perfetto. I've looked at a generic-heavy library that's similar (?) to fp-ts, effect-ts (https://effect.website/), but I worry that the overhead - both at compile time with the complex types, and at runtime with the highly abstracted control flow that v8 doesn't seem to like - would be a large net negative for our codebase.
Nothing that large admittedly- but I have gotten near the 1 million mark(prolly ~800k?) in one project. But I'd also say that at that size(honestly these days, I reach for it pretty much by default) I'd go toward a monorepo that only runs CI on the packages that have changes, as that much JS to even just go through a typical eslint is gonna be a real chore. As a result, the 'complex types' don't end up impacting as much.
As to runtime: While v8 doesn't like it, what it doesn't like even more is having code to run in the first place- and I've found that my FP-heavy projects often have fewer lines of code by factors of 3 at worse, often as high as 15. So in general I didn't get much in the way of perf issues, and when there would be a place that perf mattered, I'd then rewrite it to not use the FP stuff and instead be written to purpose.
Basically, I use FP(and by extension fp-ts) as a good default(as it increased velocity by enormous factors, and more as time went on), then reach for the toolbox when the situation called for it.
BIG ASTERISK however: I don't use `fp-ts` much in React however. With it primarily depending on `Object.is` for comparison, the pure nature of the libraries creates a need for a lot of tools I wasn't able to find a satisfactory answer to. So most code like this was either accomplishing things outside of components(components would often call them though), or was backend-focused(ie, Node.js).
Minor nit: I’ve found types like these—that is, iterative recursive types—benefit from using terminology common to map/reduce. And by “benefit from”, I mean become more understandable by a wider audience—not necessarily the HN audience per se, but quite likely teammates and future selves.
Which is to say, these names almost always make types like this more clear:
- Head: the first item in the input type you’re iterating through
- Tail: the remaining items or unprocessed structure you’ll likely recurse on next
- Acc (or pick your favorite “reduced” idiom): a named type for the intermediate product which will become the final type when you finish iterating. This can be provided as an optional parameter with an empty tuple as its default, largely modeling a typical reduce (apart from inverting the common parameter order).
It also helps, IME, to put a “base case” first in the type’s conditions.
When all of these names and patterns are utilized, the resulting type tends to look quite a lot like an equivalent runtime function you could encounter for producing the value equivalent to its type. This is great because you can even write the runtime function to match the type’s logic. This demonstrates both what the type is doing for people who find these “complex types” intimidating, and that the type accurately describes the value it’s associated with.
this sounds similar to prolog!
Cool.
There is a way to make this easier to extend, though: https://tsplay.dev/WGbEXm
Can't tell off the top of my head if there are any disadvantages to this approach though.
Neat, but this is basically a ripoff of this post from a few years ago (even to the point of not including the runtime implementation):
https://www.hacklewayne.com/a-truly-strongly-typed-printf-in...
Reminds me of Idris: https://gist.github.com/chrisdone/672efcd784528b7d0b7e17ad9c...
Recently though, I've been wondering whether advanced type system stuff is the right approach. It usually becomes pretty complicated, like another language on top of the regular language. Maybe it would be easier to have some kind of framework for compiler plugins that do extra checks. Something that would make it easy to check format strings or enforce rules on custom attributes, like Linux's sparse does, using plain imperative code that's readable to the average dev. Large projects would have an extra directory for compile time checks in addition to the tests directory they have now.
But I haven't seen any language community do something like that. What am I missing?
> Maybe it would be easier to have some kind of framework for compiler plugins that do extra checks. [...] But I haven't seen any language community do something like that. What am I missing?
Go has adopted a similar approach to this - they've made it fairly easy to write separate plugins that check stuff like this. The plugins aren't executed as part of the compiler though, they're standalone tools. For example, see golangci-lint, which bundles together a load of plugins of this kind.
Some of these plugins are shipped within the go command directly, as part of the "go vet" subcommand. (including a printf format check, which is similar to what's described in this post, i.e. it checks that arguments are of the correct type).
Sounds like comptime from Zig. There are a few others that does something similar, but Zig probably has most mind share right now.
You parse the string and then iterate over the passed arguments and check if everything adds ups. Rather straightforward.
Expressing it in the type system like TS did is impressive, but not simple.
I wonder if we should have a kind of "hidden type system", where we still take advantage of having a single type system to reason about, but the extra-specific "weird-ish" types can be hidden, almost like private variables, where visibility is literally hidden from the programmer unless obtained from debug modes or errors.
You mean like the C++ auto keyword but everywhere?
auto is just type inference, doesn't change any visibility
> another language
With the property of verifiably correct behavior
> compiler plugin
A number of languages allow it (Haskell being the most prolific example, but also Java, Scala, gcc, many others)
Maybe check out Clojure spec?
Unless I'm mistaken, Clojure spec operates at run-time, not at compile-time.
I don’t see why static assertions wouldn’t be enough in this case.
The interesting thing here is that the typesafe printf has its function arguments inferred from a string literal, at compile time. You can change the 9 to a "9" and see the type error even before running the code.
This is something that most mainstream language's type system cannot do.
(This may be obvious, but a lot of commenters here might have missed that.)
not sure i understand the utility of this when format strings and string template types already exist.
you can also use typescript-eslint/restrict-template-expressions if you find yourself running into problems with that
https://typescript-eslint.io/rules/restrict-template-express...
I think this is less about the utility and more about showing off unusual ways to use the TypeScript type system.
I've been kind of curious why tricks like this aren't used more to make sql and such. Heck, you could do similar tricks for shell execution. Or any general "string that is parseable." Seems we always take the route of not parsing the string as much as we can?
This is a pretty neat application, but most embedded languages like SQL have a way more complicated grammar that would require a really complicated set of types to parse. This can tank the performance of your type checking step and it also means that the error messages you get out of the parser-in-types are going to be nearly useless.
A more common solution is to parse the string at runtime with a proper parser with decent error handling and then have the parser return a branded type [0] which you can use elsewhere to ensure your strings are well formed.
[0] https://egghead.io/blog/using-branded-types-in-typescript
I'd expect that for most SQL that is being inlined in code, you could get by with only supporting a subset of what is possible. I also have no doubt you could easily sabotage the effort by trying to support every dialect of sql in a smart way. :D
I'm also curious on the idea of having a string parsed at runtime and why that is necessarily better? Sounds like this is essentially dynamic typing? Where they are calling it branded, instead of dynamic? At first, I confess the idea sounded close to a tagged union. You have to have something in the data to indicate the tag; but I guess it is missing the union part? Definitely looks close to the idea of treating "objects" as maps.
Neat idea, thanks for sharing!
I can't find it now, but someone actually built that for SQL in Typescript as an experiment. The problem folks run into is IDE and compiler performance. These sorts of features are what make your system turing complete, so they start stressing the compiler pretty quickly
I think, from having it used recently, that supabase's TS library does this. I had to write a wrapper around it a few months ago at $dayjob and was really surprised when select/from parts of a "query" (not really a SQL query, because it's just a postgrest query) actually got parsed at compile time and spit out the right types. And since our code is pretty type heavy, I was gonna have to do that anyway, so I really appreciated it
There is actually efforts in the Typescript community attempting to do just that. Personally I think it'll end up being a waste, but these sorts of experiments, even when they fail, often can help along new discoveries.
And on the off-chance they get it right, then damn that's pretty great.
Fully agreed that failure is expected! I think it can still make great learning for all involved. And, 100% agreed that proving us wrong on this would be great!
There is an implementation of SQL that operates on a table shaped type, entirely at type level. For your amusement: https://github.com/codemix/ts-sql
There are a bunch of more practical takes that codegen types from your database and generate types for your queries, eg: https://github.com/adelsz/pgtyped
To me the second approach seems much more pragmatic because you don’t need to run a SQL parser in your typechecker interpreter on every build
Cool! I'm not sure how I feel on the idea of keeping it out of the "typechecker." At large, the amount of work being done by what we call the compiler nowadays is already far more than folks would have called for decades ago. I don't know that I think there is a solid line of "this should never happen during compilation."
Your implication that it is a tradeoff, btw, I fully endorse/agree with. And I know there will be pathological projects out there where the code gen will take ages to complete. I'd hazard a guess that most people wouldn't notice the codegen happening on every build for most projects. Especially on modern build machines.
> why tricks like this aren't used more
Some languages don't support this.
The languages that do would require extensive systems to implement this feature. It may simply not be a priority over other requirements like thread safety, atomicity, etc.
> similar tricks for shell execution
Shell only supports strings, integers and lists. The type system is too limited for this level of type-checking.
This works in typescript due to the advanced type operations built into the language.
Apologies, I meant the equivalent of `os.popen` in python. You'd almost certainly only support a subset of what a shell actually supports, but that would almost certainly be for the best.
Basic point being that the equivalent of named/delimited parameters with pretty much forced support for escaping such that you have to go out of your way to send raw strings.
I think, bottom line, it bemuses me that the default "convenience" methods are almost always "send this string over to another process to evaluate it" instead of any processing locally on it. That feels it would be far better as the power "escape hatch" instead of the "convenience method" that it is often pitched as.
Am I missing something? This is just a toy implementation of a function prototype, that only includes integers and strings?
As a general rule, if something is on the HN homepage and you find yourself asking "am I missing something?", the answer is almost by definition "yes" :)
It's just a cool use of some of typescript's more advanced features that many developers probably don't use on a day-to-day basis (likely for good reason, as other comments have pointed out!)
I really enjoy how people try to dispel their outright attempts at bullying behavior with an emoticon. :)
Meanwhile, the HN homepage is not some carefully guarded display of exceptional merit, and no serious "hacker" would take the things posted here to be above reproach.
I think it was an attempt to add a cheeky or comical tone to the response, instead of outright saying "Yes, you're missing something" or the more curt "Yes". But if I helps,
Yes, you're missing something.
Nice!
Now do ReScript. :D
Honestly, if you spend that much code on a single `printf`, I will reject your PR and we will have a conversation about code maintenance and cost.
Please don't adopt this.
printf is about 700 lines in musl libc https://git.musl-libc.org/cgit/musl/tree/src/stdio/vfprintf....
and there's no language-level type safety, although plenty of tools lint printf now
Except missing the pesky runtime implementation. We don't need though, right? As long as the types say it's right.
I think the point is safely typing the pattern of having variadic functions with a format string argument.
The function implementation itself isn’t that interesting, or “pesky” to be honest
The static types depicted in typescript are entirely fictitious. Any similarity to runtime types is purely coincidental.
Yeah. I see a lot of "typescript is more readable" arguments out there, but I find this code dense and verbose. The more words you use to explain something the more likely you are to be misunderstood. What we're looking at here is basically a restricted wrapper for console.log and a regex implementation meant to simulate a logger in another language. Why not just write a cross-compiler for that language? There's no learning curve for the syntax then, only the target platform.
<rant>The invention, support and defense of Typescript baffles me. It feels like an intensely wasteful work-around for poorly written interpreter error messages concocted by comp-sci grads who think compiled languages are superior to interpreted ones in all situations and they want to bring this wisdom to developers of loosely typed languages. </rant>
This isn't typical application code, but either proof-of-concept or library code. What you'd instead get is the type error, which in turn would explain itself well enough- that it expected a certain type, and got a different type.
The reason you wouldn't want a cross-compiler for that other language is that there are true semantic differences between languages, and many languages simply cannot fully support the JS runtime- and if your primary application is still in Typescript, trying to cross-compile for a single feature is downright ridiculous.
> The invention, support and defense of Typescript baffles me. It feels like an intensely wasteful work-around for poorly written interpreter error messages concocted by comp-sci grads who think compiled languages are superior to interpreted ones in all situations and they want to bring this wisdom to developers of loosely typed languages.
I sincerely don't think statically typed languages are superior, but I'd argue a large part of the increase in quality over the last few years of the ecosystem is due in large part to Typescript.
Is it going to fix all, or even a majority of the issues? Probably not. But if it can improve upon the situation I don't see why we'd make perfect the enemy of good.
Hmm, let me try to defend TypeScript, then. I think it’s a terrific language, and more importantly manages to salvage JavaScript into a very decent language.
Coming from a mostly C/C++ background, I had been very skeptical of “gradually typed” languages like Dart (and now Python), but I’ve come around to the view that for many purposes it’s better than a completely statically typed compiled language.
You don’t need to compile TS at all, just bundle it, which is lightning fast in current tools, so it feels almost as nimble as pure JS. But you have almost-instant type checking in tooltips and code completions, and you can run a full project check whenever you want (that’s slow, but still way faster than compiling Rust).
The type system isn’t perfect but it’s incredibly expressive. Almost anything you could imagine doing in straightforward JS can be typed (admittedly sometimes with a lot of effort and head-scratching) and once typed it’s generally easy to use with confidence that most runtime errors will be avoided.
The fact that JS and TS are separate languages is a bonus -- it keeps TS honest, and the competition between JS engines means runtime performance is great. If you were designing TS from scratch, I think you’d be tempted to add some kind of runtime type reification, but TS is better off without that. Completely erasing types means compilation will always be fast, and in my experience RTTI causes more architectural problems than it solves.
> Hmm, let me try to defend TypeScript, then. I think it’s a terrific language, and more importantly manages to salvage JavaScript into a very decent language
Counterpoint: It's a desperate attempt to make Javascript useable and nearly does so, but ends up being weird in itself to get around the limitations of the underlying language.
I use Typescript most days and I hate it. Part of that is that npm is a dumpster fire but a lot of it is that Typescript is a rubbish version of much better languages and it hurts to use it. I am so desperate to get back to something sane like C# that I will quit this job for less money.
I'm sure we'll never fully agree, but I am surprised that you think TypeScript "ends up being weird in itself". To the contrary, I think the type system feels very natural; what feels weird is when you switch to another language and realise it doesn't have e.g. sum types.
One big problem I do bump into with TS is that the tooling of TS itself is pretty bad -- the way it manages compiler options, import paths, etc. Luckily you don't need to use TSC very much in practice, and other TS-aware tools like esbuild and Bun are vastly better.
(Likewise, I find myself much happier with NPM than with other package managers like Maven or PyPI, at least when you switch the NPM tool itself for a better drop-in replacement like PNPM.)
Anyway, each to their own! I hope you're able to find a good C# gig or similar.
This is such a cynical take. The point is to model what types exist at runtime in the type system, so that you can reason about those same runtime types statically. They’re only “fictitious” if they’re defined incorrectly, or if the type system can’t sufficiently express certain of their nuances. The former is usually only the case when developers intentionally work around the safety provided by the type system; the latter is possible, but at this point it’s usually only ever the case for patterns that are hard to reason about regardless of the type system or even the presence of types at all.
I find myself inclined to the opinion you're disagreeing with in all honesty.
When defining types in other languages, the task is prescriptive (you specify what fields there are in a type and the runtime accepts this as law), but in Typesxript the task is meant to be descriptive (as you say, one models the types that exist at runtime which is the inverse).
I was excited about Typescript when I learned it, but found myself disillusioned by actual experience when using it (of course others love it and have good reason to). Had defined classes in Typescript so I can have some of my types reflected at runtime.
Curious what your issue was with duck-typing. Were you effectively looking to create ADTs that are required to go through a specific step-by-step process, not simply 'look like' the thing that was expected?
If so, you might be interested in [newtype-ts].
[newtype-ts]: https://github.com/gcanti/newtype-ts
Thanks for the link. That was exactly my use case and I should remember your helpful suggestion next time I use TS.
Guardrails won't keep you on the road if you intentionally steer into them at full speed either.
I mean, all types are "entirely fictitious" as far as the computer is concerned. Yeah they usually have fewer layers than JS does, but that's a pretty arbitrary line to draw.
If they're "coincidental" you missing the point, and not doing them right.
What do you mean? `console.log` supports `%d` and `%f` already.
console.log("This is a %s and a %d!", "string", 4.3) -> This is a string and a 4!