For secret #3, I think another option if you really need to catch general exceptions could be to catch CancellationException in its own, dedicated catch block to rethrow it. Not much different, but I like the clarity of using separate catch blocks for each type of exception that can be caught
Please never ever ever ever copy anything byte by byte. You made the function as inefficient as possible. Copy single byte then do ensureActive and your file is copying like 1000 times slower? Always use buffers so such operations start with 4 kb buffer and measure, experiment. But please, never do byte by byte...
I've already wrote this in another comment: I did a benchmark for readBytes and reading byte by byte with 5.2Mb file and the results are: readBytes - 41..58ms, byte by byte - 26.6..27.5 seconds
On tip 3, you can have the exception handling outside the loop (if you want the behaviour of cancelling the loop) i.e. try { while {} }catch {}. This advice may not be pertinent to every exception, but swallowing general exceptions in a loop is rarely advisable.
Hey Philipp, Thanks for the video! Could you please make a new video on "How to use Room with KMP"? Previously, your video covered only Android and iOS, but I would like to see how this can be done for desktop as well. Also, it would be great if you could create a full 3-4 hour production ready project based on KMP that runs on Android, iOS, and desktop. Thanks!
I think I've made one mistake relating to secret number 2 (Supervisor Job) in my last project. I have a bug (not a crasher) in the app but it happens rarely.
Hi Philipp. I would like to tell you how much your videos like this one helped me in the recent past and I'm very grateful to you because of it. Also I'd like to give you something back: for Secret 1 there is actually a Service implementation that has a built in lifecycle and lifecycleScope just like a viewModel has and it's called LifecycleService.
Facts about Kotlin that annoy me: - They added null safety. But removed exceptions safety. - They could have built a cancellation mechanism into the generated code. But they invented CancellationException.
You'll be shocked what code you'll find when looking into the toByteArray() function 🤯 TL;DR: No, that won't have an impact on performance, it's the same code with an extra check for cancellation which is nothing else than an extra if-statement being evaluated for each byte. Unless you're running your code on phones from 1995, that won't have an impact ;)
@@PhilippLackner I see there's a System.arraycopy(). A slightly redundant array copy. Not great, not terrible. I was working on a project more than 5 years ago, I wrote the same code in some networking part of the app. Changing it to the read(array, position, length) call and copying 1KB at a time improved performance by like a 1000x. It was a noticeable difference. Anyway, Kotlin should handle this better. There should be a standard library call to do this. There is runInterruptible(), but that's not perfect either ...
I don't have any data on this, but you could get rid of that arraycopy() by creating the array on your own and appending each newly read byte in each iteration
Exactly! I would read input stream by chunks or at least wrap input stream to BufferedInputStream. The second thing I would pre-allocate byte array in ByteArrayOutputStream as the readBytes() function do. This prevent internal resizing byte array after adding bytes.
Yes, these tips are all features of the KotlinX Coroutines library. The only platform-specific parts of this video were the Android app components lifecycle and the specific file operations.
Ive 7 years of experience in Android app dev, But I never paid any attention to these! Thank you so much!
For secret #3, I think another option if you really need to catch general exceptions could be to catch CancellationException in its own, dedicated catch block to rethrow it. Not much different, but I like the clarity of using separate catch blocks for each type of exception that can be caught
Please never ever ever ever copy anything byte by byte. You made the function as inefficient as possible. Copy single byte then do ensureActive and your file is copying like 1000 times slower? Always use buffers so such operations start with 4 kb buffer and measure, experiment. But please, never do byte by byte...
I've already wrote this in another comment:
I did a benchmark for readBytes and reading byte by byte with 5.2Mb file and the results are: readBytes - 41..58ms, byte by byte - 26.6..27.5 seconds
@@BelokonRoman really?
When I see hardcoded dispatchers in the code, it's usually a sign that people don't care about testing functionality with coroutines in general 🙂
I have to go check my app for 2 & 3.
Thank you. I love learning about Coroutines
note about 5 example, if you read/write data byte-by-byte it could be very slow...
For #4, in case of Dagger, you could use qualifier, and inject Dispatcher directly. Or without DI, simply set it to some default value.
The secrets 2 and 5 🤯
Great Video! Philipp Lackner
Philipp,thanks for the video!
Thanks for such a great video ;)
You can use isActive in your while loop that will solve the problem isn't it ? 14:44
I agree with everything you've read, colleague
On tip 3, you can have the exception handling outside the loop (if you want the behaviour of cancelling the loop) i.e. try { while {} }catch {}. This advice may not be pertinent to every exception, but swallowing general exceptions in a loop is rarely advisable.
@@TheAndreArtus that will still break the cancelation mechanism and can execute code that shouldn't be executed
What if in #3 we do not re-throw, but replace while (true) with while (isActive)?
Hey Philipp,
Thanks for the video! Could you please make a new video on "How to use Room with KMP"? Previously, your video covered only Android and iOS, but I would like to see how this can be done for desktop as well.
Also, it would be great if you could create a full 3-4 hour production ready project based on KMP that runs on Android, iOS, and desktop.
Thanks!
love it!
This is awesome
I knew about supervisor, it is used also in viewmodelScope
I think I've made one mistake relating to secret number 2 (Supervisor Job) in my last project. I have a bug (not a crasher) in the app but it happens rarely.
Hi Philipp. I would like to tell you how much your videos like this one helped me in the recent past and I'm very grateful to you because of it. Also I'd like to give you something back: for Secret 1 there is actually a Service implementation that has a built in lifecycle and lifecycleScope just like a viewModel has and it's called LifecycleService.
Facts about Kotlin that annoy me:
- They added null safety. But removed exceptions safety.
- They could have built a cancellation mechanism into the generated code. But they invented CancellationException.
Fifth solution is not a solution,
But new problem
Hi please help me i want to do work in kotlin not jetpack. I want to do data transfer using local only hotspot is that possible
Secret 5 is going to destroy the performance though ...
You'll be shocked what code you'll find when looking into the toByteArray() function 🤯
TL;DR: No, that won't have an impact on performance, it's the same code with an extra check for cancellation which is nothing else than an extra if-statement being evaluated for each byte. Unless you're running your code on phones from 1995, that won't have an impact ;)
@@PhilippLackner I see there's a System.arraycopy(). A slightly redundant array copy. Not great, not terrible.
I was working on a project more than 5 years ago, I wrote the same code in some networking part of the app. Changing it to the read(array, position, length) call and copying 1KB at a time improved performance by like a 1000x. It was a noticeable difference.
Anyway, Kotlin should handle this better. There should be a standard library call to do this. There is runInterruptible(), but that's not perfect either ...
I don't have any data on this, but you could get rid of that arraycopy() by creating the array on your own and appending each newly read byte in each iteration
😂@@PhilippLackner
Exactly! I would read input stream by chunks or at least wrap input stream to BufferedInputStream. The second thing I would pre-allocate byte array in ByteArrayOutputStream as the readBytes() function do. This prevent internal resizing byte array after adding bytes.
Wow can I use it in multiplatform kmp or not?
@@meshoraouf8929 Absolutely👍
Yes, these tips are all features of the KotlinX Coroutines library. The only platform-specific parts of this video were the Android app components lifecycle and the specific file operations.
plsss speak sloww
Without SupervisorJob() on scope , still other coroutines continues to run
val scope = CoroutineScope(Dispatchers.IO)
Button(onClick = {
scope.launch {
launch {
try{
println("child 1 coroutine")
delay(2.seconds)
throw Exception("Failure in Child 1")
} catch (e : Exception){
throw CancellationException()
}
}
delay(4.seconds)
launch {
println("child 2 coroutine")
}
}
}) {
Text(text = "Scope check")
}