Name: PINFuture
Owner: Pinterest
Description: An Objective-C future implementation that aims to provide maximal type safety
Created: 2016-12-05 01:48:32.0
Updated: 2018-05-23 08:12:32.0
Pushed: 2018-05-22 22:05:41.0
Size: 625
Language: Objective-C
GitHub Committers
User | Most Recent Commit | # Commits |
---|
Other Committers
User | Most Recent Commit | # Commits |
---|
PINFuture is available through CocoaPods. To install it, simply add the following line to your Podfile:
"PINFuture"
PINFuture is an Objective-C implementation of the asynchronous primitive called “Future”. This library differs from other Objective-C implementations of Future primarily because it aims to preserve type safety using Objective-C generics.
A Future is a wrapper for “a value that will eventually be ready”.
A Future can have one of 3 states and usually begins in the “Pending” state. “Pending” means that the final value of the Future is not yet known but is currently being computed. The Future will eventually transition to either a “Fulfilled” state and contain a final value, or transition to a “Rejected” state and contain an error object. “Fulfilled” and “Rejected” are terminal states, and the value/error of a Future cannot change after the first fulfill or reject transition.
Callback style
oid)logInWithUsername:(NSString *)username
password:(NSString *)password
success:( void (^)(User *user) )successBlock
failure:( void (^)(NSError *error) )failureBlock;
Future style
INFuture<User *> *)logInWithUsername:(NSString *)username
password:(NSString *)password;
Callback style
f showSpinner];
r logInWithUsername:username password:password success:^(User *user) {
[Posts fetchPostsForUser:user success:^(Posts *posts) {
dispatch_async(dispatch_get_main_queue(), ^{
[self hideSpinner];
// update the UI to show posts
});
} failure:^(NSError *error) {
dispatch_async(dispatch_get_main_queue(), ^{
[self hideSpinner];
// update the UI to show the error
});
}];
ilure:^(NSError *error) {
dispatch_async(dispatch_get_main_queue(), ^{
[self hideSpinner];
// update the UI to show the error
});
Future style
f showSpinner];
uture<User *> *userFuture = [User logInWithUsername:username password:password];
uture<Posts *> *postsFuture = [PINFutureMap<User *, Posts *> flatMap:userFuture executor:[PINExecutor main] transform:^PINFuture<Posts *> *(User *user) {
return [Posts fetchPostsForUser:user];
tsFuture executor:[PINExecutor main] completion:^{
[self hideSpinner];
tsFuture executor:[PINExecutor main] success:^(Posts *posts) {
// update the UI to show posts
ilure:^(NSError *error) {
// update the UI to show the error
Callback style
tub([fileMock readContentsPath:@"foo.txt"
success:OCMOCK_ANY
failure:OCMOCK_ANY]).andDo(^(NSInvocation *invocation) {
(void)(^successBlock)(NSString *) = nil;
[invocation getArgument:&successBlock atIndex:3];
if (successBlock) {
successBlock(@"fake contents");
}
Future style
tub([fileMock readContentsPath:@"foo.txt"]).andReturn([PINFuture<NSString *> withValue:@"fake contents"]);
To access the final value of a Future, register success
and failure
callbacks. If you only want to know when a Future completes (and not the specific value or error), register a complete
callback.
Callbacks will be dispatched in the order that they are registered. However, depending on your specified executor
, the blocks might execute in a different order or even execute concurrently.
Whenever you pass a callback block you must also pass a required executor:
parameter. The executor
determines where and when your block will be executed.
executor:
[PINExecutor main]
Executes a block on the Main GCD queue.[PINExecutor background]
Executes a block from a pool of background threads. Careful: With this executor, it's possible that two callback blocks attached to a single future will execute concurrently.A good rule of thumb: Use [PINExecutor background]
if work that your callback block does is thread-safe and if the work doesn't need to be executed from the Main thread (e.g. because it's touching UIKit).
PINFuture makes use of Objective-C generics to maintain the same type safety that you'd have with callbacks.
Future<NSNumber *> withValue:@"foo"]; // Compile error. Good!
In Objective-C, type parameters are optional. It's a good practice to always specify them for a PINFuture.
Future withValue:@"foo"]; // This compiles but will likely blow up with "unrecognized selector" when the value is used.
PINFuture is non-blocking and provides no mechanism for blocking. Blocking a thread on the computation of an async value is generally not a good practice, but is possible using Grand Central Dispatch Semaphores
PINFuture does not capture Exceptions thrown by callbacks. On platforms that PINFuture targets, NSException
s are generally fatal. PINFuture deals with NSError
s.
withValue:
Construct an already-fulfilled Future with a value.
uture<NSString *> stringFuture = [PINFuture<NSString *> withValue:@"foo"];
withError:
Construct an already-rejected Future with an error.
uture<NSString *> stringFuture = [PINFuture<NSString *> withError:[NSError errorWithDescription:...]];
withBlock:
Construct a Future and fulfill or reject it by calling one of two callbacks. This method is generally not safe since because there's no enforcement that your block will call either resolve
or reject
. This is most useful for writing a Future-based wrapper for a Callback-based method. You'll find this method used extensively in the PINFuture wrappers of Cocoa APIs.
uture<NSString *> stringFuture = [PINFuture<NSString *> withBlock:^(void (^ fulfill)(NSString *), void (^ reject)(NSError *)) {
[foo somethingAsyncWithSuccess:resolve failure:reject];
executor:block:
Construct a Future by executing a block that returns a Future. The most common use case for this is to dispatch some chunk of compute-intensive work off of the the current thread. You should prefer this method to withBlock:
whenever you can return a Future because the compiler can enforce that all code paths of your block will return a Future.
uture<NSNumber *> fibonacciResultFuture = [PINFuture<NSNumber *> executor:[PINExecutor background] block:^PINFuture *() {
NSInteger *fibonacciResult = [self computeFibonacci:1000000];
return [PINFuture<NSNumber *> withValue:fibonacciResult];
In order to achieve type safety for an operation like map
that converts from one type of value to another type, we have to jump through some hoops because of Objective-C's rudimentary support for generics. map
and flatMap
are class methods on the class PINFutureMap
. The PINFutureMap
class has two type parameters: FromType
and ToType
.
map
and flatMap
only preform a transformation is the source Future is fulfilled. If the source Future is rejected, then the original error is simply passed through to the return value.mapError
and flatMapError
only preform a transformation is the source Future is rejected. If the source Future is fulfilled, then the original value is simply passed through to the return value.map
uture<NSString *> stringFuture = [PINFutureMap<NSNumber *, NSString *> map:numberFuture executor:[PINExecutor background] transform:^NSString *(NSNumber * number) {
return [number stringValue];
flatMap
uture<UIImage *> imageFuture = [PINFutureMap<User *, UIImage *> flatMap:userFuture executor:[PINExecutor background] transform:^PINFuture<NSString *> *(User *user) {
return [NetworkImageManager fetchImageWithURL:user.profileURL];
mapError
uture<NSString *> *stringFuture = [File readUTF8ContentsPath:@"foo.txt" encoding:EncodingUTF8];
ngFuture = [fileAFuture executor:[PINExecutor immediate] mapError:^NSString * (NSError *errror) {
return ""; // If there's any problem reading the file, continue processing as if the file was empty.
flatMapError
uture<NSString *> *stringFuture = [File readUTF8ContentsPath:@"tryFirst.txt"];
ngFuture = [fileAFuture executor:[PINExecutor background] flatMapError:^PINFuture<NSString *> * (NSError *errror) {
if ([error isKindOf:[NSURLErrorFileDoesNotExist class]) {
return [File readUTF8ContentsPath:@"trySecond.txt"];
} else {
return [PINFuture withError:error]; // Pass through any other type of error
}
gatherAll
ray<NSString *> fileNames = @[@"a.txt", @"b.txt", @"c.txt"];
ray<PINFuture<NSString *> *> *fileContentFutures = [fileNames map:^ PINFuture<NSString *> *(NSString *fileName) {
return [File readUTF8ContentsPath:fileName];
uture<NSArray<NSString *> *> *fileContentsFuture = [PINFuture<NSString *> gatherAll:fileContentFutures];
eContentsFuture executor:[PINExecutor main] success:^(NSArray<NSString *> *fileContents) {
// All succceeded.
ilure:^(NSError *error) {
// One or more failed. `error` is the first one to fail.
gatherSome
Experimental. This API may change to improve type safety.
ray<NSString *> fileNames = @[@"a.txt", @"b.txt", @"c.txt"];
ray<PINFuture<NSString *> *> *fileContentFutures = [fileNames map:^ PINFuture<NSString *> *(NSString *fileName) {
return [File readUTF8ContentsPath:fileName];
uture<NSArray *> *fileContentsOrNullFuture = [PINFuture<NSString *> gatherSome:fileContentFutures];
eContentsFuture executor:[PINExecutor main] success:^(NSArray *fileContents) {
// fileContents is an array of either `NSString *` or `[NSNull null]` depending on whether the source future resolved or rejected.
ilure:^(NSError *error) {
// This can't be reached. If any of the source futures fails, there will be a `[NSNull null]` entry in the array.
chainSuccess:failure:
This is similar to success:failure
except that a new Future is returned that does not fulfill or reject until the side-effect has been executed. This should be used sparingly. It should be rare that you want to have a side-effect, and even rarer to wait on a side-effect.
etch a user, and return a Future that resolves only after all NotificationCenter observers have been notified.
uture<User *> *userFuture = [self userForUsername:username];
Future = [userFuture executor:[PINExecutor main] chainSuccess:^(User *user) {
[[NSNotifcationCenter sharedCenter] postNotification:kUserUpdated object:user];
ilure:nil;
rn userFuture;
executeOnMain
/executeOnBackground
We've observed that application code will almost always call with either executor:[PINExecutor main]
or executor:[PINExecutor background]
. For every method that takes an executor:
there are 2 variations of that method, executeOnMain
and executeOnBackground
, that are slightly more concise (shorter by 22 characters).
The following pairs of calls are equivalent. The second call in each pair demonstrated the convenience method.
rFuture executor:[PINExecutor main] success:success failure:failure];
rFuture executeOnMainSuccess:success failure:failure];
rFuture executor:[PINExecutor background] success:success failure:failure];
rFuture executeOnBackgroundSuccess:success failure:failure];
uture<Post *> *postFuture = [PINFutureMap<User, Post> map:userFuture executor:[PINExecutor main] transform:transform];
uture<Post *> *postFuture = [PINFutureMap<User, Post> map:userFuture executeOnMainTransform:transform];
uture<Post *> *postFuture = [PINFutureMap<User, Post> map:userFuture executor:[PINExecutor background] transform:transform];
uture<Post *> *postFuture = [PINFutureMap<User, Post> map:userFuture executeOnBackgroundTransform:transform];
run
.These decisions are possibly controversial but deliberate.
success:failure:
and completion:
methods. A reader could easily be mislead into thinking that the chained operations are guaranteed to execute sequentially.success:
method or a failure:
method. We think it's a better for the site of any side-effects to make it explicit that they don't want to handle a value or that they don't want to handle an error by passing a NULL
argument.value
and error
as parameters to the completion
block. If a caller needs to consume value
or error
, they should be using success:failure:
. If they need to execute cleanup code without consuming the value, then completion
is more appropriate. If a value
and an error
are passed to completion
, it's very easy for callback code to misinterpret whether the future resolved or rejected.Copyright 2016-2018 Pinterest, Inc
Licensed under the Apache License, Version 2.0 (the “License”); you may not use this file except in compliance with the License. You may obtain a copy of the License at http://www.apache.org/licenses/LICENSE-2.0
Unless required by applicable law or agreed to in writing, software distributed under the License is distributed on an “AS IS” BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. See the License for the specific language governing permissions and limitations under the License.