I don’t like how dang busy it gets. What I do is if I need more than a few args, I accept only an object. And offer only type information in the signature, and resolve default values at the start of the function. It even works nicely if all args are optional.
They are passing "prefix", "ending", and "extraNames" in an object. It's just defined inline in the function and using destructuring so the function can access the keys by name without having to put the object in a variable (and the default values are part of the object destructuring).
I agree the problem, but the given solution feels a bit clunky to me. From a syntax perspective, passing an options object and applying it over a set of defaults feels a lot cleaner. The reason kwargs work in Python is primarily because the syntax is clean on both the caller and callee's side of things. This puts a lot of additional non-obvious syntax on the callee side.
I personally kind of feel like no existing language provides great syntax for structured function parameters. I recently wrote a post about how I think languages could make this better in general (although one thing I didn't tackle in my post is default values for arguments, which is very important): https://samestep.com/blog/parameter-syntax/
Oh interesting; is that actually true? I'm not aware of Haskell features that allow it to do better than any of the three other languages I compared to in my post. What would my running example look like in Haskell?
{-# LANGUAGE RecordWildCards #-}
main =
transport
("2025-09-05T09:00:00Z", "2025-09-05T11:00:00Z")
foo123
kwargs
where
foo123 = Foo
kwargs = Kwargs {..}
bar = Bar
person = Person {..}
name = "Alex Smith"
phone = "+1 (555) 555-5555"
options = Opts {..}
fragile = True
window = PM
data Foo = Foo
data Bar = Bar
data Kwargs = Kwargs
{ bar :: Bar
, person :: Person
, options :: Opts
}
data Person = Person
{ name :: String
, phone :: String
}
data Opts = Opts
{ fragile :: Bool
, window :: HalfDay
}
data HalfDay = AM | PM
transport (start, end)
foo
Kwargs { person = Person{ name = who, ..}, options = Opts{..}, ..}
= do
doStuff who foo
moreStuff phone end window
otherThings start fragile bar
doStuff who foo = print "called doStuff"
moreStuff phone end window = print "called moreStuff"
otherThings start fragile bar = print "calle otherThings"
I see, so similar to the other languages I mentioned in my post, it looks like Haskell doesn't let you colocate the types with the leaves of the binding form?
The `Kwargs{..}` alone does imply a non-ambiguous type in the function signature (without explicit but optional annotations), and it binds locally scoped names too. Why doesn't it colocate in the same sense?
It's similar to the Rust example from my post: there are no anonymous records, so in order to be able to construct something with named fields, you need to first define those record types outside. In contrast, my proposed syntax depends on already having anonymous record types like OCaml and TypeScript do, but avoids the extra boilerplate required by separating types from binding forms as shown in my TypeScript example.
To address the example towards the end of your post, if you want patterns + struct binding at the same time, prefix patterns with `bindName@`, for instance:
transport
myTuple@(start, end)
foo
kwargs@(Kwargs { person = Person{ name = who, ..}, options = Opts{..}, ..})
= do
...
My personal philosophy, which isn't shared by many but I'm yet to find it failing me: (almost) never use default arguments. It's bad design and causes real issues, between bugs and incidents I've been involved in dozens of situations where I've found that "this would never have happened if this argument didn't have a default value".
First, there's very rarely a value that really makes sense to be the "default". People often use a default value when it's a "commonly used" value: I don't think it's a good reason, other values usually are just as valid and important as the most-commonly-used value. It's like saying someone in Sweden is "white by default" because it's the most common case: it's going to be often correct but _why would you do that_ when you could just make it explicit. The typing you're saving isn't worth it.
Second, it makes your API less easy to use. Devs _should_ read through all the available arguments anyway to make your function do what they want, including these default arguments (even if they decide not to set them), so what did you gain from making them optional? A bit of typing, a bit of screen real estate? Sure, but devs _do not actually behave like that_ and often don't read or don't pay enough attention to optional arguments. They just (consciously or not) hope that the default values will be reasonable for their specific use-case. Well, too often it's not and that result in a bug or incident: the typing and screen estate you gained are not worth it. FORCE devs to think about what's the correct value, that's good design.
Third, it's much less readable. The reader doesn't know that there are invisible arguments with an implicit value, despite these arguments modifying the behaviour of the function. Make everything explicit and no-one needs special implicit knowledge to know what a function is doing (even Python says "explicit is better than implicit").
Fourth, what would make sense as a default at some point might not make sense later. You pick a default that seems to make sense, people use this default value all over the place, things change and actually new use-cases usually need a different value... well sucks to be you. Now you have a footgun, as it's not even a good default for new use-cases and it's more likely to be misused, or you need to go over all previous use-cases to give an explicit value so that you can change the default... might as well have never made it a default.
There's exceptions of course: I wouldn't want `key` or `className` to be explicit in React, but for 99% of backend stuff, default values are a bad design. I hate Go for this: zero values have caused _countless_ issues to almost everyone I know, whether they acknowledge it or not.
I was going to disagree with you; and I was going to use a very simple function to do it `randString(length, alphabet)` where alphabet is defaulted - as azAZ09 is a good default.
However, on many non-trivial projects I've worked on we ended up making the default explicit and then creating wrappers for specific use cases (i.e. randFileName, randFilePath not being case sensitive on some platforms and randTestFilePath including additional characters for tests but production variants sticking to portable sets).
Its further worth noting that rarely do I allow the default global random to be used and instead dependency inject it via params/factories.
JS is a fun language where argument defaults can accidentally shoot you in the foot because of no runtime type checking of argument application. With your example, you might want to map a bunch of lengths through it, which in simplest form might look:
someLengthArray.map(randString)
The classic surprise is that map provides 2 arguments with the second argument being the index of the array. Depending on how `randString(length, alphabet)` uses its alphabet it might just use the string form of the index so you accidentally get random strings of n length '0' then '1' then '2', etc.
(The most common case of this problem in JS is probably `parseInt(string, radix = 10)`. Accidental arbitrary radix from an index parameter is a fun way to get unexpected numbers back from your array of strings.)
I don't really like this at all. I'm not familiar with Python but this kwargs sounds like it leads to inconsistency throughout usages of a function.
In JS/TS it's a lot more conventional to use objects once you go past 2-3 arguments. Sometimes patterns from other languages are great in other languages (like, functional ways of modelling errors, making invalid state impossible, not using nulls, etc), but other patterns like the one shown here don't seem great.
Hmm? “Use objects once you go past 2-3 arguments” seems to be exactly what this post is arguing for. (Which, as you say, is not novel and is common in browser APIs and other JS libraries.) Not sure what you’re disagreeing with.
TLDR: Typescript has objects and destructuring. If you squint it kinda looks like Python's kwargs, because both JS objects and Python kwargs are simply key value pairs.
I don’t like how dang busy it gets. What I do is if I need more than a few args, I accept only an object. And offer only type information in the signature, and resolve default values at the start of the function. It even works nicely if all args are optional.
They are passing "prefix", "ending", and "extraNames" in an object. It's just defined inline in the function and using destructuring so the function can access the keys by name without having to put the object in a variable (and the default values are part of the object destructuring).
Yeah but to have half-decent typings you then have to define the entire type, again, in the signature.
^^ this
I would probably do the same before using kwargs.
I agree the problem, but the given solution feels a bit clunky to me. From a syntax perspective, passing an options object and applying it over a set of defaults feels a lot cleaner. The reason kwargs work in Python is primarily because the syntax is clean on both the caller and callee's side of things. This puts a lot of additional non-obvious syntax on the callee side.
Object destructuring with default parameters is a JavaScript feature.
The first code example on the article is wrong. It would print "!, Bob." and not "Hello, Bob!".
I personally kind of feel like no existing language provides great syntax for structured function parameters. I recently wrote a post about how I think languages could make this better in general (although one thing I didn't tackle in my post is default values for arguments, which is very important): https://samestep.com/blog/parameter-syntax/
You can do most of it in default Haskell syntax, plus `RecordWildCards`: https://ghc.gitlab.haskell.org/ghc/doc/users_guide/exts/reco...
Oh interesting; is that actually true? I'm not aware of Haskell features that allow it to do better than any of the three other languages I compared to in my post. What would my running example look like in Haskell?
Here's the sample you can play with (interactive: https://play.haskell.org/saved/TWBEYFzH):
I see, so similar to the other languages I mentioned in my post, it looks like Haskell doesn't let you colocate the types with the leaves of the binding form?
> doesn't let you colocate the types with the leaves of the binding form
Can you elaborate on the effective difference between colocations in your example and this:
The `Kwargs{..}` alone does imply a non-ambiguous type in the function signature (without explicit but optional annotations), and it binds locally scoped names too. Why doesn't it colocate in the same sense?It's similar to the Rust example from my post: there are no anonymous records, so in order to be able to construct something with named fields, you need to first define those record types outside. In contrast, my proposed syntax depends on already having anonymous record types like OCaml and TypeScript do, but avoids the extra boilerplate required by separating types from binding forms as shown in my TypeScript example.
To address the example towards the end of your post, if you want patterns + struct binding at the same time, prefix patterns with `bindName@`, for instance:
I use that approach to group options in a single parameter. But then not only the options parameter must be nullable, but also its properties:
function findAll(tableName: string, opts?: { where: Record<string, any>; limit?: number; sort?: "DESC"| "ASC" }) {}
Just make it a well-defined class.
Like option structs in Go. Simple.
[dead]
My personal philosophy, which isn't shared by many but I'm yet to find it failing me: (almost) never use default arguments. It's bad design and causes real issues, between bugs and incidents I've been involved in dozens of situations where I've found that "this would never have happened if this argument didn't have a default value".
First, there's very rarely a value that really makes sense to be the "default". People often use a default value when it's a "commonly used" value: I don't think it's a good reason, other values usually are just as valid and important as the most-commonly-used value. It's like saying someone in Sweden is "white by default" because it's the most common case: it's going to be often correct but _why would you do that_ when you could just make it explicit. The typing you're saving isn't worth it.
Second, it makes your API less easy to use. Devs _should_ read through all the available arguments anyway to make your function do what they want, including these default arguments (even if they decide not to set them), so what did you gain from making them optional? A bit of typing, a bit of screen real estate? Sure, but devs _do not actually behave like that_ and often don't read or don't pay enough attention to optional arguments. They just (consciously or not) hope that the default values will be reasonable for their specific use-case. Well, too often it's not and that result in a bug or incident: the typing and screen estate you gained are not worth it. FORCE devs to think about what's the correct value, that's good design.
Third, it's much less readable. The reader doesn't know that there are invisible arguments with an implicit value, despite these arguments modifying the behaviour of the function. Make everything explicit and no-one needs special implicit knowledge to know what a function is doing (even Python says "explicit is better than implicit").
Fourth, what would make sense as a default at some point might not make sense later. You pick a default that seems to make sense, people use this default value all over the place, things change and actually new use-cases usually need a different value... well sucks to be you. Now you have a footgun, as it's not even a good default for new use-cases and it's more likely to be misused, or you need to go over all previous use-cases to give an explicit value so that you can change the default... might as well have never made it a default.
There's exceptions of course: I wouldn't want `key` or `className` to be explicit in React, but for 99% of backend stuff, default values are a bad design. I hate Go for this: zero values have caused _countless_ issues to almost everyone I know, whether they acknowledge it or not.
I was going to disagree with you; and I was going to use a very simple function to do it `randString(length, alphabet)` where alphabet is defaulted - as azAZ09 is a good default.
However, on many non-trivial projects I've worked on we ended up making the default explicit and then creating wrappers for specific use cases (i.e. randFileName, randFilePath not being case sensitive on some platforms and randTestFilePath including additional characters for tests but production variants sticking to portable sets).
Its further worth noting that rarely do I allow the default global random to be used and instead dependency inject it via params/factories.
JS is a fun language where argument defaults can accidentally shoot you in the foot because of no runtime type checking of argument application. With your example, you might want to map a bunch of lengths through it, which in simplest form might look:
The classic surprise is that map provides 2 arguments with the second argument being the index of the array. Depending on how `randString(length, alphabet)` uses its alphabet it might just use the string form of the index so you accidentally get random strings of n length '0' then '1' then '2', etc.(The most common case of this problem in JS is probably `parseInt(string, radix = 10)`. Accidental arbitrary radix from an index parameter is a fun way to get unexpected numbers back from your array of strings.)
I don't really like this at all. I'm not familiar with Python but this kwargs sounds like it leads to inconsistency throughout usages of a function.
In JS/TS it's a lot more conventional to use objects once you go past 2-3 arguments. Sometimes patterns from other languages are great in other languages (like, functional ways of modelling errors, making invalid state impossible, not using nulls, etc), but other patterns like the one shown here don't seem great.
Hmm? “Use objects once you go past 2-3 arguments” seems to be exactly what this post is arguing for. (Which, as you say, is not novel and is common in browser APIs and other JS libraries.) Not sure what you’re disagreeing with.
So, an entire article just telling you to put objects as the last parameters to functions? Wow thanks, I'd have never thought of that.
[dead]
TLDR: Typescript has objects and destructuring. If you squint it kinda looks like Python's kwargs, because both JS objects and Python kwargs are simply key value pairs.
TL;DR + Some extra:
* 1-2 args: fn(arg1, arg2)
* More args: fn({ arg1, arg2, arg3, ... })
* Sometimes: fn(mainArg, { ...optionObject })
[dead]