This question is a near-duplicate of Apply a proxy using traits. However, that question dealt with applying a proxy to an Attribute, and I would like to do the same thing for a Variable. From Jonathan's answer, I understand that I
need to arrange for the Proxy to be bound into the attribute, so that there's a Proxy there rather than a Scalar container that is usually created by class initialization logic.
However, I can't seem to bind successfully to a Variable:D, even at compile time. (Including with nqp::bind). I'd greatly appreciate any pointers in the correct direction.
(Ideally, I'd like to support using the variable/trait with assignment syntax. In a perfect world, I'd have syntax like:
my $thing is custom-proxy = 42;
And the result of that would be that $thing is containerized inside the Proxy, but not in a Scalar. But if that's not possible, I'd settle for getting it working with binding via :=.
[EDIT: building on the accepted answer below, it is possible to mostly do this with the following code:
multi trait_mod:<is>(Variable \v, :$tom) {
v.block.add_phaser(
'ENTER',
v.willdo(<-> $_ {
$_ = Proxy.new:
STORE => -> $, $v { say "store $v" },
FETCH => { say "fetch!"; 42}
}, 1))
}
This works for variables that are not initialized to a different value or for state variables on calls to the function other than the first.
You can always bind.
my $actual-thing = 42;
my $thing := Proxy.new(
FETCH => anon method fetch () {
say 'fetch';
$actual-thing
},
STORE => anon method store ($new) {
say 'store ',$new;
$actual-thing = $new
}
);
say $thing;
$thing = 5;
say $thing;
Which currently results in the following.
fetch
fetch
fetch
fetch
fetch
fetch
fetch
42
store 5
fetch
fetch
fetch
fetch
fetch
fetch
fetch
5
(The repeated FETCH calls are a known limitation.)
If you wanted to have syntax like
my $thing is custom-proxy = 42;
You would need to start with
multi trait_mod:<is> ( Variable:D \var, :$custom-proxy! ){
…
}
The problem is that currently doing it this way requires a lot of deep Rakudo/nqp knowledge that I do not possess.
For example the code behind my $var is default('value') looks a bit like this:
multi sub trait_mod:<is>(Variable:D $v, Mu :$default!) {
my $var := $v.var;
my $what := $var.VAR.WHAT;
my $descriptor;
{
$descriptor := nqp::getattr($var, $what.^mixin_base, '$!descriptor');
CATCH {
my $native = $v.native($what);
…
}
}
…
$descriptor.set_default(nqp::decont($default));
# make sure we start with the default if a scalar
$var = $default if nqp::istype($what, Scalar);
}
Why does that have $what.^mixin_base?
I have no idea.
Why isn't $!descriptor accessible something like $v.var.descriptor?
I have no idea.
How do we change $v.var.VAR from a Scalar to a Proxy?
I have no idea.
Is that last one doable? (From within a trait_mod:<is>)
I am fairly certain that the answer is yes.
My 2d[1]:
I'd settle for getting it working with binding via :=.
sub custom-proxy is rw { Proxy.new: FETCH => { 42 }, STORE => { ... } }
my $variable := custom-proxy;
say $variable; # 42
In a perfect world, I'd have syntax like:
my $thing is custom-proxy = 42;
Aiui, that's #Larry's intent.
But, as you presumably know, if a type (eg role custom-proxy { ... }) is applied using an is trait to a scalar variable (eg my $variable is custom-proxy) then the compiler emits a compile time error message (is trait on $-sigil variable not yet implemented).
I can't seem to bind successfully to a Variable:D, even at compile time
First, let's clarify what a Variable is, and what you would need to successfully bind to:
multi trait_mod:<is>(Variable \var, :$foo!) { say var.var.VAR.WHAT } # (Scalar)
my $variable is foo;
You might think you could bind to var. But the compiler is passing an lvalue, so you're not going to be able to alter it.
You might think you could bind to var.var, which is an attribute of a Variable. (I explain what a Variable is, and its var attribute, and why I had to write "varvarVAR!" in the above code, here.)
The SO you linked shows how to alter the value bound to an attribute in some object:
$a.set_build: -> \SELF, | {
$a.set_value: SELF, Proxy.new:
STORE => -> $, $val { say "store $val" },
FETCH => { say "fetch!"; 42 }
}
So perhaps you could use that approach to alter the .var attribute of a Variable?
Unfortunately, "setting build logic" is used to "bind the attribute ... at each object creation", (hence "you'll be overriding any initial default value").
So I don't think this technique is going to help in this case because the Variable, and hence its .var attribute, has presumably already been built by the time the Variable is passed to the is trait.
In summary, while a trait is called at compile-time, I think it's called too late because the var attribute has already been permanently bound.
My guess is that altering Raku(do) so that the Variable's .var attribute becomes writable, or using metaprogramming to dive underneath Variable's public API to force through a change, would be beyond fraught, unreasonably complicating the compiler's variable handling code and/or swapping out codegen optimization logic for pessimization logic.
This may be behind #Larry's speculation that a more controlled is type on scalar variables will one day be implemented.
Footnotes
[1] My two (pennies | dogecoin).
Let’s say I have an options variable and I want to set some default value.
What’s is the benefit / drawback of these two alternatives?
Using object spread
options = {...optionsDefault, ...options};
Or using Object.assign
options = Object.assign({}, optionsDefault, options);
This is the commit that made me wonder.
This isn't necessarily exhaustive.
Spread syntax
options = {...optionsDefault, ...options};
Advantages:
If authoring code for execution in environments without native support, you may be able to just compile this syntax (as opposed to using a polyfill). (With Babel, for example.)
Less verbose.
Disadvantages:
When this answer was originally written, this was a proposal, not standardized. When using proposals consider what you'd do if you write code with it now and it doesn't get standardized or changes as it moves toward standardization. This has since been standardized in ES2018.
Literal, not dynamic.
Object.assign()
options = Object.assign({}, optionsDefault, options);
Advantages:
Standardized.
Dynamic. Example:
var sources = [{a: "A"}, {b: "B"}, {c: "C"}];
options = Object.assign.apply(Object, [{}].concat(sources));
// or
options = Object.assign({}, ...sources);
Disadvantages:
More verbose.
If authoring code for execution in environments without native support you need to polyfill.
This is the commit that made me wonder.
That's not directly related to what you're asking. That code wasn't using Object.assign(), it was using user code (object-assign) that does the same thing. They appear to be compiling that code with Babel (and bundling it with Webpack), which is what I was talking about: the syntax you can just compile. They apparently preferred that to having to include object-assign as a dependency that would go into their build.
For reference object rest/spread is finalised in ECMAScript 2018 as a stage 4. The proposal can be found here.
For the most part object assign and spread work the same way, the key difference is that spread defines properties, whilst Object.assign() sets them. This means Object.assign() triggers setters.
It's worth remembering that other than this, object rest/spread 1:1 maps to Object.assign() and acts differently to array (iterable) spread. For example, when spreading an array null values are spread. However using object spread null values are silently spread to nothing.
Array (Iterable) Spread Example
const x = [1, 2, null , 3];
const y = [...x, 4, 5];
const z = null;
console.log(y); // [1, 2, null, 3, 4, 5];
console.log([...z]); // TypeError
Object Spread Example
const x = null;
const y = {a: 1, b: 2};
const z = {...x, ...y};
console.log(z); //{a: 1, b: 2}
This is consistent with how Object.assign() would work, both silently exclude the null value with no error.
const x = null;
const y = {a: 1, b: 2};
const z = Object.assign({}, x, y);
console.log(z); //{a: 1, b: 2}
I think one big difference between the spread operator and Object.assign that doesn't seem to be mentioned in the current answers is that the spread operator will not copy the the source object’s prototype to the target object. If you want to add properties to an object and you don't want to change what instance it is of, then you will have to use Object.assign.
Edit: I've actually realised that my example is misleading. The spread operator desugars to Object.assign with the first parameter set to an empty object. In my code example below, I put error as the first parameter of the Object.assign call so the two are not equivalent. The first parameter of Object.assign is actually modified and then returned which is why it retains its prototype. I have added another example below:
const error = new Error();
error instanceof Error // true
const errorExtendedUsingSpread = {
...error,
...{
someValue: true
}
};
errorExtendedUsingSpread instanceof Error; // false
// What the spread operator desugars into
const errorExtendedUsingImmutableObjectAssign = Object.assign({}, error, {
someValue: true
});
errorExtendedUsingImmutableObjectAssign instanceof Error; // false
// The error object is modified and returned here so it keeps its prototypes
const errorExtendedUsingAssign = Object.assign(error, {
someValue: true
});
errorExtendedUsingAssign instanceof Error; // true
See also: https://github.com/tc39/proposal-object-rest-spread/blob/master/Spread.md
There's a huge difference between the two, with very serious consequences. The most upvoted questions do not even touch this, and the information about object spread being a proposal is not relevant in 2022 anymore.
The difference is that Object.assign changes the object in-place, while the spread operator (...) creates a brand new object, and this will break object reference equality.
First, let's see the effect, and then I'll give a real-world example of how important it is to understand this fundamental difference.
First, let's use Object.assign:
// Let's create a new object, that contains a child object;
const parentObject = { childObject: { hello: 'world '} };
// Let's get a reference to the child object;
const childObject = parentObject.childObject;
// Let's change the child object using Object.assign, adding a new `foo` key with `bar` value;
Object.assign(parentObject.childObject, { foo: 'bar' });
// childObject is still the same object in memory, it was changed IN PLACE.
parentObject.childObject === childObject
// true
Now the same exercise with the spread operator:
// Let's create a new object, that contains a child object;
const parentObject = { childObject: { hello: 'world '} };
// Let's get a reference to the child object;
const childObject = parentObject.childObject;
// Let's change the child object using the spread operator;
parentObject.childObject = {
...parentObject.childObject,
foo: 'bar',
}
// They are not the same object in memory anymore!
parentObject.childObject === childObject;
// false
It's easy to see what is going on, because on the parentObject.childObject = {...} we are cleary assigning the value of the childObject key in parentObject to a brand new object literal, and the fact it's being composed by the old childObject content's is irrelevant. It's a new object.
And if you assume this is irrelevant in practice, let me show a real world scenario of how important it is to understand this.
In a very large Vue.js application, we started noticing a lot of sluggishness when typing the name of the customer in an input field.
After a lot of debugging, we found out that each char typed in that input triggered a hole bunch of computed properties to re-evaluate.
This wasn't anticipated, since the customer's name wasn't used at all in those computeds functions. Only other customer data (like age, sex) was being used. What was goin on? Why was vue re-evaluating all those computed functions when the customer's name changed?
Well, we had a Vuex store that did this:
mutations: {
setCustomer(state, payload) {
// payload being { name: 'Bob' }
state.customer = { ...state.customer, ...payload };
}
And our computed were like this:
veryExpensiveComputed() {
const customerAge = this.$store.state.customer.age;
}
So, voilá! When the customer name changed, the Vuex mutation was actually changing it to a new object entirely; and since the computed relied on that object to get the customer age, Vue counted on that very specific object instance as a dependency, and when it was changed to a new object (failing the === object equality test), Vue decided it was time to re-run the computed function.
The fix? Use Object.assign to not discard the previous object, but to change it in place ...
mutations: {
setCustomer(state, payload) {
// payload being same as above: { name: 'Bob' }
Object.assign(state.customer, payload);
}
BTW, if you are in Vue2, you shouldn't use Object.assign because Vue 2 can't track those object changes directly, but the same logic applies, just using Vue.set instead of Object.assign:
mutations: {
setCustomer(state, payload) {
Object.keys(payload).forEach(key => {
Vue.set(state.customer, key, payload[key])
})
}
NOTE: Spread is NOT just syntactic sugar around Object.assign. They operate much differently behind the scenes.
Object.assign applies setters to a new object, Spread does not. In addition, the object must be iterable.
Copy
Use this if you need the value of the object as it is at this moment, and you don't want that value to reflect any changes made by other owners of the object.
Use it for creating a shallow copy of the object
good practice to always set immutable properties to copy - because mutable versions can be passed into immutable properties, copy will ensure that you'll always be dealing with an immutable object
Assign
Assign is somewhat the opposite to copy.
Assign will generate a setter which assigns the value to the instance variable directly, rather than copying or retaining it.
When calling the getter of an assign property, it returns a reference to the actual data.
I'd like to summarize status of the "spread object merge" ES feature, in browsers, and in the ecosystem via tools.
Spec
https://github.com/tc39/proposal-object-rest-spread
This language feature is a stage 4 proposal, which means it's been merged into the ES language spec, but not yet widely implemented.
Browsers: in Chrome, in SF, Firefox soon (ver 60, IIUC)
Browser support for "spread properties" shipped in Chrome 60, including this scenario.
Support for this scenario does NOT work in current Firefox (59), but DOES work in my Firefox Developer Edition. So I believe it will ship in Firefox 60.
Safari: not tested, but Kangax says it works in Desktop Safari 11.1, but not SF 11
iOS Safari: not teseted, but Kangax says it works in iOS 11.3, but not in iOS 11
not in Edge yet
Tools: Node 8.7, TS 2.1
NodeJS has supported since 8.7 (via Kangax). Confirmed on 9.8 when I tested locally.
TypeScript has suported it since 2.1, current is 2.8
Links
Kangax "property spread"
https://davidwalsh.name/merge-objects
Code Sample (doubles as compatibility test)
var x = { a: 1, b: 2 };
var y = { c: 3, d: 4, a: 5 };
var z = {...x, ...y};
console.log(z); // { a: 5, b: 2, c: 3, d: 4 }
Again: At time of writing this sample works without transpilation in Chrome (60+), Firefox Developer Edition (preview of Firefox 60), and Node (8.7+).
Why Answer?
I'm writing this 2.5 years after the original question. But I had the very same question, and this is where Google sent me. I am a slave to SO's mission to improve the long tail.
Since this is an expansion of "array spread" syntax I found it very hard to google, and difficult to find in compatibility tables. The closest I could find is Kangax "property spread", but that test doesn't have two spreads in the same expression (not a merge). Also, the name in the proposals/drafts/browser status pages all use "property spread", but it looks to me like that was a "first principal" the community arrived at after the proposals to use spread syntax for "object merge". (Which might explain why it is so hard to google.) So I document my finding here so others can view, update, and compile links about this specific feature. I hope it catches on. Please help spread the news of it landing in the spec and in browsers.
Lastly, I would have added this info as a comment, but I couldn't edit them without breaking the authors' original intent. Specifically, I can't edit #ChillyPenguin's comment without it losing his intent to correct #RichardSchulte. But years later Richard turned out to be right (in my opinion). So I write this answer instead, hoping it will gain traction on the old answers eventually (might take years, but that's what the long tail effect is all about, after all).
As others have mentioned, at this moment of writing, Object.assign() requires a polyfill and object spread ... requires some transpiling (and perhaps a polyfill too) in order to work.
Consider this code:
// Babel wont touch this really, it will simply fail if Object.assign() is not supported in browser.
const objAss = { message: 'Hello you!' };
const newObjAss = Object.assign(objAss, { dev: true });
console.log(newObjAss);
// Babel will transpile with use to a helper function that first attempts to use Object.assign() and then falls back.
const objSpread = { message: 'Hello you!' };
const newObjSpread = {...objSpread, dev: true };
console.log(newObjSpread);
These both produce the same output.
Here is the output from Babel, to ES5:
var objAss = { message: 'Hello you!' };
var newObjAss = Object.assign(objAss, { dev: true });
console.log(newObjAss);
var _extends = Object.assign || function (target) { for (var i = 1; i < arguments.length; i++) { var source = arguments[i]; for (var key in source) { if (Object.prototype.hasOwnProperty.call(source, key)) { target[key] = source[key]; } } } return target; };
var objSpread = { message: 'Hello you!' };
var newObjSpread = _extends({}, objSpread, { dev: true });
console.log(newObjSpread);
This is my understanding so far. Object.assign() is actually standardised, where as object spread ... is not yet. The only problem is browser support for the former and in future, the latter too.
Play with the code here
Hope this helps.
The object spread operator (...) doesn't work in browsers, because it isn't part of any ES specification yet, just a proposal. The only option is to compile it with Babel (or something similar).
As you can see, it's just syntactic sugar over Object.assign({}).
As far as I can see, these are the important differences.
Object.assign works in most browsers (without compiling)
... for objects isn't standardized
... protects you from accidentally mutating the object
... will polyfill Object.assign in browsers without it
... needs less code to express the same idea
Too many wrong answers...
I did a search and found a lot of misinformation on this.
Summary
Neither ...spread nor Object.assign is faster. It depends.
Object.assign is almost always faster if side-effects/object mutation is not a consideration.
Performance aside, there is generally no up/downside to using either, until you reach edge cases, such copying objects containing getters/setters, or read-only properties. Read more here.
Performance
Whether Object.assign or ...spread is faster, it depends on what you are trying to combine, and the runtime you are using (implementations and optimisations happen from time to time). For small objects, it does not matter.
For bigger objects, Object.assign is usually better. Especially if you do not need to care about side-effects, the speed gains come from saving time by just adding properties to the first object, rather than copying from two and creating a brand new one. See:
async function retrieveAndCombine() {
const bigPayload = await retrieveData()
const smallPayload = await retrieveData2()
// only the properties of smallPayload is iterated through
// whereas bigPayload is mutated.
return Object.assign(bigPayload, smallPayload)
}
If side-effects is a concern
In cases where side-effects matters, such as if an object to be combined with another is passed in as an argument.
In the example below, bigPayload will be mutated, which is a bad idea if other functions/objects outside of retrieveAndCombine depends on bigPayload. In this case, you can swap the arguments passed to Object.assign, or just use {} as the first argument to create a new object:
async function retrieveAndCombine(bigPayload) {
const smallPayload = await retrieveData2()
// this is slightly more efficient
return Object.assign(smallPayload, bigPayload)
// this is still okay assuming `smallPayload` only has a few properties
return Object.assign({}, smallPayload, bigPayload)
}
Test
See if for yourself, try the code snippet below. Note: It takes awhile to run.
const rand = () => (Math.random() + 1).toString(36).substring(7)
function combineBigObjects() {
console.log('Please wait...creating the test objects...')
const obj = {}
const obj2 = {}
for (let i = 0; i < 100000; i++) {
const key = rand()
obj[rand()] = {
[rand()]: rand(),
[rand()]: rand(),
}
obj2[rand()] = {
[rand()]: rand(),
[rand()]: rand(),
}
}
console.log('Combine big objects using spread:')
console.time()
const result1 = {
...obj,
...obj2,
}
console.timeEnd()
console.log('Combine big objects using assign:')
console.time()
Object.assign({}, obj, obj2)
console.timeEnd()
console.log('Combine big objects using assign, but mutates first obj:')
console.time()
Object.assign(obj, obj2)
console.timeEnd()
}
combineBigObjects()
function combineSmallObjects() {
const firstObject = () => ({ [rand()]: rand() })
const secondObject = () => ({ [rand()]: rand() })
console.log('Combine small objects using spread:')
console.time()
for (let i = 0; i < 500000; i++) {
const finalObject = {
...firstObject(),
...secondObject(),
}
}
console.timeEnd()
console.log('Combine small objects using assign:')
console.time()
for (let i = 0; i < 500000; i++) {
const finalObject = Object.assign({}, firstObject(), secondObject())
}
console.timeEnd()
console.log('Combine small objects using assign, but mutates first obj:')
console.time()
for (let i = 0; i < 500000; i++) {
const finalObject = Object.assign(firstObject(), secondObject())
}
console.timeEnd()
}
combineSmallObjects()
Other answers are old, could not get a good answer.
Below example is for object literals, helps how both can complement each other, and how it cannot complement each other (therefore difference):
var obj1 = { a: 1, b: { b1: 1, b2: 'b2value', b3: 'b3value' } };
// overwrite parts of b key
var obj2 = {
b: {
...obj1.b,
b1: 2
}
};
var res2 = Object.assign({}, obj1, obj2); // b2,b3 keys still exist
document.write('res2: ', JSON.stringify (res2), '<br>');
// Output:
// res2: {"a":1,"b":{"b1":2,"b2":"b2value","b3":"b3value"}} // NOTE: b2,b3 still exists
// overwrite whole of b key
var obj3 = {
b: {
b1: 2
}
};
var res3 = Object.assign({}, obj1, obj3); // b2,b3 keys are lost
document.write('res3: ', JSON.stringify (res3), '<br>');
// Output:
// res3: {"a":1,"b":{"b1":2}} // NOTE: b2,b3 values are lost
Several more small examples here, also for array & object:
https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Operators/Spread_syntax
The ways to (1) create shallow copies of objects and (2) merge multiple objects into a single object have evolved a lot between 2014 and 2018.
The approaches outlined below became available and widely used at different times. This answer provides some historical perspective and is not exhaustive.
Without any help from libraries or modern syntax, you would use for-in loops, e.g.
var mergedOptions = {}
for (var key in defaultOptions) {
mergedOptions[key] = defaultOptions[key]
}
for (var key in options) {
mergedOptions[key] = options[key]
}
options = mergedOptions
2006
jQuery 1.0 has jQuery.extend():
options = $.extend({}, defaultOptions, options)
⋮
2010
Underscore.js 1.0 has _.extend()
options = _.extend({}, defaultOptions, options)
⋮
2014
2ality published an article about Object.assign() coming to ES2015
object-assign published to npm.
var objectAssign = require('object-assign')
options = objectAssign({}, defaultOptions, options)
The Object Rest/Spread Properties syntax proposed for ES2016.
2015
Object.assign is supported by Chrome (45), Firefox (34) and Node.js (4). Polyfill is required for older runtimes though.
options = Object.assign({}, defaultOptions, options)
The Object Rest/Spread Properties proposal reaches stage 2.
2016
The Object Rest/Spread Properties syntax did not get included in ES2016, but proposal reaches stage 3.
2017
The Object Rest/Spread Properties syntax did not get included in ES2017, but is usable in Chrome (60), Firefox (55), and Node.js (8.3). Some transpilation is needed for older runtimes though.
options = { ...defaultOptions, ...options }
2018
The Object Rest/Spread Properties proposal reaches stage 4 and the syntax is included in ES2018 standard.
Object.assign is necessary when the target object is a constant and you want to set multiple properties at once.
For example:
const target = { data: "Test", loading: true }
Now, suppose you need to mutate the target with all properties from a source:
const source = { data: null, loading: false, ...etc }
Object.assign(target, source) // Now target is updated
target = { ...target, ...source) // Error: cant assign to constant
Keep in mind that you are mutating the target obj, so whenever possible use Object.assign with empty target or spread to assign to a new obj.
This is now part of ES6, thus is standardized, and is also documented on MDN:
https://developer.mozilla.org/en-US/docs/Web/JavaScript/Reference/Operators/Spread_operator
It's very convenient to use and makes a lot of sense alongside object destructuring.
The one remaining advantage listed above is the dynamic capabilities of Object.assign(), however this is as easy as spreading the array inside of a literal object. In the compiled babel output it uses exactly what is demonstrated with Object.assign()
So the correct answer would be to use object spread since it is now standardized, widely used (see react, redux, etc), is easy to use, and has all the features of Object.assign()
I'd like to add this simple example when you have to use Object.assign.
class SomeClass {
constructor() {
this.someValue = 'some value';
}
someMethod() {
console.log('some action');
}
}
const objectAssign = Object.assign(new SomeClass(), {});
objectAssign.someValue; // ok
objectAssign.someMethod(); // ok
const spread = {...new SomeClass()};
spread.someValue; // ok
spread.someMethod(); // there is no methods of SomeClass!
It can be not clear when you use JavaScript. But with TypeScript it is easier if you want to create instance of some class
const spread: SomeClass = {...new SomeClass()} // Error
The spread operator spread the Array into the separate arguments of a function.
let iterableObjB = [1,2,3,4]
function (...iterableObjB) //turned into
function (1,2,3,4)
We’ll create a function called identity that just returns whatever parameter we give it.
identity = (arg) => arg
And a simple array.
arr = [1, 2, 3]
If you call identity with arr, we know what’ll happen
libc's error handling is usually to return something < 0 in case of an error. I find myself doing this over and over:
let pid = fork()
if pid < 0 {
// Please disregard the fact that `Err(pid)`
// should be a `&str` or an enum
return Err(pid);
}
I find it ugly that this needs 3 lines of error handling, especially considering that these tests are quite frequent in this kind of code.
Is there a way to return an Err in case fork() returns < 0?
I found two things which are close:
assert_eq!. This needs another line and it panics so the caller cannot handle the error.
Using traits like these:
pub trait LibcResult<T> {
fn to_option(&self) -> Option<T>;
}
impl LibcResult<i64> for i32 {
fn to_option(&self) -> Option<i64> {
if *self < 0 { None } else { Some(*self) }
}
}
I could write fork().to_option().expect("could not fork"). This is now only one line, but it panics instead of returning an Err. I guess this could be solved using ok_or.
Some functions of libc have < 0 as sentinel (e.g. fork), while others use > 0 (e.g. pthread_attr_init), so this would need another argument.
Is there something out there which solves this?
As indicated in the other answer, use pre-made wrappers whenever possible. Where such wrappers do not exist, the following guidelines might help.
Return Result to indicate errors
The idiomatic Rust return type that includes error information is Result (std::result::Result). For most functions from POSIX libc, the specialized type std::io::Result is a perfect fit because it uses std::io::Error to encode errors, and it includes all standard system errors represented by errno values. A good way to avoid repetition is using a utility function such as:
use std::io::{Result, Error};
fn check_err<T: Ord + Default>(num: T) -> Result<T> {
if num < T::default() {
return Err(Error::last_os_error());
}
Ok(num)
}
Wrapping fork() would look like this:
pub fn fork() -> Result<u32> {
check_err(unsafe { libc::fork() }).map(|pid| pid as u32)
}
The use of Result allows idiomatic usage such as:
let pid = fork()?; // ? means return if Err, unwrap if Ok
if pid == 0 {
// child
...
}
Restrict the return type
The function will be easier to use if the return type is modified so that only "possible" values are included. For example, if a function logically has no return value, but returns an int only to communicate the presence of error, the Rust wrapper should return nothing:
pub fn dup2(oldfd: i32, newfd: i32) -> Result<()> {
check_err(unsafe { libc::dup2(oldfd, newfd) })?;
Ok(())
}
Another example are functions that logically return an unsigned integer, such as a PID or a file descriptor, but still declare their result as signed to include the -1 error return value. In that case, consider returning an unsigned value in Rust, as in the fork() example above. nix takes this one step further by having fork() return Result<ForkResult>, where ForkResult is a real enum with methods such as is_child(), and from which the PID is extracted using pattern matching.
Use options and other enums
Rust has a rich type system that allows expressing things that have to be encoded as magic values in C. To return to the fork() example, that function returns 0 to indicate the child return. This would be naturally expressed with an Option and can be combined with the Result shown above:
pub fn fork() -> Result<Option<u32>> {
let pid = check_err(unsafe { libc::fork() })? as u32;
if pid != 0 {
Some(pid)
} else {
None
}
}
The user of this API would no longer need to compare with the magic value, but would use pattern matching, for example:
if let Some(child_pid) = fork()? {
// execute parent code
} else {
// execute child code
}
Return values instead of using output parameters
C often returns values using output parameters, pointer parameters into which the results are stored. This is either because the actual return value is reserved for the error indicator, or because more than one value needs to be returned, and returning structs was badly supported by historical C compilers.
In contrast, Rust's Result supports return value independent of error information, and has no problem whatsoever with returning multiple values. Multiple values returned as a tuple are much more ergonomic than output parameters because they can be used in expressions or captured using pattern matching.
Wrap system resources in owned objects
When returning handles to system resources, such as file descriptors or Windows handles, it good practice to return them wrapped in an object that implements Drop to release them. This will make it less likely that a user of the wrapper will make a mistake, and it makes the use of return values more idiomatic, removing the need for awkward invocations of close() and resource leaks coming from failing to do so.
Taking pipe() as an example:
use std::fs::File;
use std::os::unix::io::FromRawFd;
pub fn pipe() -> Result<(File, File)> {
let mut fds = [0 as libc::c_int; 2];
check_err(unsafe { libc::pipe(fds.as_mut_ptr()) })?;
Ok(unsafe { (File::from_raw_fd(fds[0]), File::from_raw_fd(fds[1])) })
}
// Usage:
// let (r, w) = pipe()?;
// ... use R and W as normal File object
This pipe() wrapper returns multiple values and uses a wrapper object to refer to a system resource. Also, it returns the File objects defined in the Rust standard library and accepted by Rust's IO layer.
The best option is to not reimplement the universe. Instead, use nix, which wraps everything for you and has done the hard work of converting all the error types and handling the sentinel values:
pub fn fork() -> Result<ForkResult>
Then just use normal error handling like try! or ?.
Of course, you could rewrite all of nix by converting your trait to returning Results and including the specific error codes and then use try! or ?, but why would you?
There's nothing magical in Rust that converts negative or positive numbers into a domain specific error type for you. The code you already have is the correct approach, once you've enhanced it to use a Result either by creating it directly or via something like ok_or.
An intermediate solution would be to reuse nix's Errno struct, perhaps with your own trait sugar on top.
so this would need another argument
I'd say it would be better to have different methods: one for negative sentinel values and one for positive sentinel values.