I have come to this same conclusion after many years. It's not cost-effective to have any bespoke business logic (models, controllers, etc.) shared between the two mobile platforms (don't get me started on sharing UI code). If you have some incredibly tricky low-level algorithm/library and/or need for speed, think database, crypto, intense graphics, etc., then fine, you may be able to swing a shared module in C++ or something. Other than that, it's almost like the collective consciousnesses of Google and Apple conspire to make cost-effective code-sharing of typical CRUD apps almost impossible.
Here's a fine alternative to code-sharing: well first and foremost you need to have a good requirements spec. After that, implement it on one platform first in a platform agnostic way. So heavy use of delegates/interfaces/injections to handle platform-specific functionality, even for one-liners like getting current system time. So the code should be pure, boring, plain old Java, Kotlin, Swift, whatever. Minimal use of fancy language features that may not port cleanly. After one platform has the module in place and well-tested, do a copy/paste port to the other platform. You will find that the copy/paste port will actually uncover bugs, optimizations, edge cases, etc., that can then be ported back to the original platform. Things you would not have found if you were only writing the code once. It's like the most intense kind of PR code review you can get.
So I've found that copy/paste-port-based code sharing between platforms is not just a lesser evil, but actually a strategy that has its own unique benefits. If you get into a groove you can build up a set of regexes to give you a sloppy transpiler that does most of the tedious porting for you.
The article says it all - they chose the wrong language with no community support for the trail they were blazing and that technical debt destroyed their productivity.
I achieved 90+% code share in Xamarin including UI across windows - ios - andriod. 100% buisness logic shared. You can write cross-platform UIs, and swap out to native code on a widget by widget basis. Threading is seemless. Sure it has its own set of issues, but everything is a trade-off. Look at your use case, and define what trade-offs you are happy with.
The drop-box use-case is also an outlier - to achieve high levels of responsiveness, it must integrate with low level platform specific APIs to the point their entire app may be some kind OS specific service. Look at blogs about their desktop apps and you see my point.
If you going to drink cool-aid from companies just because they are dominant in an area, atleast drink the cool-aid of a startup which was created for the purpose of building cross-platform apps.
I took a look at Xamarin recently to see how far it had come, and the tooling is really impressive. They seem to do a great job of keeping up to date with iOS/Android SDK changes, too.
But when trying to find an example of a native-feeling Xamarin-based app for iOS, I was really underwhelmed. I tried to find one that was (a) well-known and (b) third-party (not developed by Xamarin or Microsoft). The 4 contenders were:
- UPS - 1/4 stars - All views of the app are awkward embedded web views. Every page has a full screen modal spinner.
- Outback Steakhouse - 1/4 stars - Not updated for modern iPhones, so the whole app is letterboxed. UI is completely custom, so didn't attempt to feel native. Crashed as soon as it tried to get my location.
- Alaska Airlines - 3/4 stars - Feels native-ish, nice "vanilla" design. Mix of native forms and web views. All UI related to searching flights & buying tickets is a web view. Every tap on the tab bar causes a full screen modal spinner.
- FOX Sports - 4/4 stars - Wow, impressive. Lots of native-feeling views with their own branded styling. The custom widgets feel native.
It was great to find one "hero" app for Xamarin (FOX Sports), but the fact that it's an outlier is a little concerning.
They specifically mentioned when I was interviewing there that the reason their Xamarin apps are so good is because they already know how to make good native iOS apps.
Also, I specifically asked about whether or not they would be using more technologies like Xamarin in the future and they didn't seem too keen on it.
According to them, making a truly native seeming app in Xamarin isn't much less work than making two separate native apps.
Many businesses go with Xamarin because some guy in a management mandates it because he feels it will be a silver bullet that will let them get around having to hire skilled native mobile developers.
The reality is that to do Xamarin well you’ll have to wander outside out Xamarin Forms, and to do that well not only do you need native mobile developers but you need far higher caliber native mobile devs who are able to mentally translate swift/objc and android calls into their C# equivalent, and who have an idea of what’s going on the os when there’s a crash at a level lower than the .NET runtime caused by a native part that’s not apparent in the crappy Xamarin stack trace.
Xamarin really only makes sense if your company has a mountain of existing complicated .NET code with priceless business logic that you absolutely have to reuse.
I used to work for them. Great company, great technical talent. If anyone wants to move to Charlottesville or Durham I encourage you to apply! I live in CVille and I love it.
The Xamarin folks are native platform engineers first. Xamarin teams comprise Android experts and iOS experts who did then went on to learn Xamarin on top.
Thats the same experience with us except instead of Xamarin we used more popular cross platform mobile framework Flutter from google. Given most of business logic is maintained in back-end. Fron-end client be it mobile or web based needs UI and state management with some local data for offline storage. We thought it will be easy to share code across platform and we were so wrong.
Given advances in mobile platform and Native SDK a company like Google, Microsoft or Facebook will not be able to keep up with native SDK and will be one or two generations behind. You can see the result in issues on flutter project. Also given Swift has Python like syntax and ease of use with powerful toolchain and Kotlin with nice syntax with functional programming efforts are not that different working with native SDK.
Obviously this cross platform mobile framework are ok for simple crud apps. But when it comes to building a mobile app using modern vision, audio, keyboard, touch and gesture input along with modern sensors in the platform and provide a consistent UX is a big challenge. Also UI and UX have enough subtle differences across iOS and Android the code written to manage them is also substantial. So its nice to go native SDK route for mobile apps.
I think there is a little bit of bias here. Companies that are prioritize look and feel over cost probably are going native. And companies that prioritize cost over look feel are free are going xamarin.
Xamarin is indeed the closest there is to the holy grail of cross-mobile development. I too had some success with it for a small project at first, but just like the theme of the blog post, the challenge is in the long-term maintenance being cost-effective, not just the initial development. Especially in the face of these adversarial platforms constantly changing things underfoot. For example in Android recently I've been forced to switch to the AndroidX library, and iOS has Dark Mode coming out. Will Xamarin support both of those platform changes cleanly? Maybe, but it's a risk.
I'd be happy to hear if the project you worked on has been cost-effective to maintain in the long term. Then at least someone has seen the mountain top!
there is a difference between Xamarin and Xamarin.Forms, one thing is just C# => ObjC binding / C# => Java binding (C# to AOT native shared libraries and binds it into the native things), which basically SUPPORTS ALL of the native features.
so basically you can either have one application on top of a custom ui framework (xamarin.forms) or you can only share business logic, and write controllers, which feels native in c# or the native language.
i.e. without forms you can build a controller with objc or you can have the implementation in c#, and call into a shared business logic library.
unfortunatly the docs are really really weak when you consider what you really can do with xamarin.
with forms you can build a ui that works on both platforms so basically you have 100% shared code.
It was years ago, but Rdio went the Xamarin route, regretted it, and then did exactly what Dropbox just did and went full native.
The reasons for regretting it were similar: lack of available engineers on the market, incompatibilities between runtimes, and lagging support for platform features (although in the Xamarin case it's not even in your hands to fix it)
Xamarin on paper seems like the holy Grail, but the fact that Microsoft itself chose to use react native instead of its own solution for Skype made it seem like I should not expect long term commitment to xamarin from MS.
I would be careful to attribute framework decision making to careful thought. This is more the regular organizational dysfunctionality. Either:
- The entire organization uses a single framework (and improves on it), which will lead to a bunch of teams grumbling that "We are only using X because of organizational pressure, while Y is much better"
- You leave freedom to the org to pick their technology, and hope/steer people towards certain parts. But you will have to accept some divergence from the norm
I prefer the second org, and luckily I am in such a place where we can use Go while the team next to us is on Scala, and we weigh in-house developed technology against publicly available technology on ROI.
Office also has to make sure that things work on the web given they also support all the Office.com O365 versions of their apps. That's another reason they'd prefer React Native (for React Native Web) to Xamarin.
But the Office team is also a right tool for each job group and has both Xamarin and React Native apps and telling the difference between them is getting tougher to do. You can already integrate React Native components inside a Xamarin app and vice versa because they each can look like native components to each other, and it's not entirely an either/or decision.
Years ago I wrote a programming language a bit like Haxe, except that it let you 'inject' snippets of your target language wherever you wanted. So you could keep your major architecture in one common language and still implement specific cases on the spot (ie you don't have to go to other files, and the various implementations for different target languages were right next to each other, making it easier to keep them in sync).
I implemented targets for JS, php and C++. It worked okayish but in the end it was more work to maintain the separate language than just write everything 3 times with some help of a simple macro/template language. I still sometimes wonder if the concept would be developed further, with a proper team and good tooling etc, if it would be a practical way of doing large scale cross platform software development.
This is a fascinating idea, and reminds me of the dreaded [1] rewrite, where you rewrite the application because the new system will always be better in the new framework...
But this is different, if you write things simply, so the code can be ported easily. Then you're essentially making the code easier to share, without tightly coupling it to a specific platform... seems like a big win to me
Agreed that it is both fascinating, and from the experience I’ve had with the approach, it has turned out to be quite an effective way to build very maintainable software.
I think that kills a lot of interesting projects right there. If I could write a solid spec, that would mean I believe I wouldn't learn much from shipping the product. And also that my competitors aren't learning or advancing.
Nobody says you have to write the spec beforehand.
As long as you don’t blurt a lazy “use the source Luke” and take the effort to document and edit the spec during or right after implementing something, it’s ok.
Nah, it’s just this whole manichean polemic between two grotesque caricatures of processes; agile improvisation vs upfront waterfall planning à la nazi wehrmacht going to Russia.
One just has to have an idea where to go before starting, and a good description where one went when you’re done.
Whatever you first create has some kind of vision behind it. Even if that vision is a rough spec, it might be precisely the little bit of spec that’s necessary to get things started.
A good requirements spec doesn’t mean it has to be heavyweight with all the corners and edges figured out. Instead, it can be just lightweight enough to get things moving in the proper direction... but some little direction is certainly needed.
Basically the same process as when you first started. Implement the changes on Platform A, get it stable, port it over to Platform B, get it stable, port back any refinements to platform A. Platform changes themselves should ideally be quarantined behind said abstractions/injections, so your actual ported code is very simple and "timeless": no fancy language features, no preprocessor magic, no annotations, no excessive syntactic sugar.
Approaches like this could increase the second-class-platform effect. Where, for example, Android gets updates long after IOS. (Unless both releases are synchronized.)
In theory the shared code approach means platforms could be tested in parallel.
The whole point of the post here is that the theory doesn't apply very well to modern mobile development, and trying to share code effectively slowed them way down.
We've had a mostly different (positive) experience doing a similar thing at FullStory for mobile instrumentation, though I think I know where some of the key differences are.
Our core "business logic" lives in Rust. This is a shared bit of code between Android & iOS that mainly deals with orchestration, serialization, and server communication. We managed to extract ~1/2 of each platform's native code into this shared orchestration. What's left on each platform is more like a platform driver.
I fully admit that things are different for us because we're not dealing with a bunch of UI that we manage (instead we're dealing with UI that our customers have written in Java/Swift/Objective-C/etc), but I think that there's a few places where we've made this work in a better way:
- We've chosen flatbuffers to serialize most (not all) of the data between the platform drivers and the shared library.
- We're using a language that's a little more hip. It's still not very easy to hire Rust developers, but developers interested in Rust are _very_ interested.
- We built tooling early on to make this pretty seamless. Our Rust code builds in XCode and gradle, so there's no real friction to a developer getting up and running.
Our team is about a year in on this transition to Rust and so far it's helped us move far quicker, especially given the size of our team.
As someone with plenty of C++ knowledge on mobile, I would expect there to be more C++ devs available with good mobile skills than there are rust devs?
Maybe my own viewpoint is twisting the amount of devs I think are available, but I don't think the choice dropbox originally made was crazy. C++ is a great way to share complex code over multiple platforms (i.e. desktop + mobile) if you're running a relatively small engineering team with limited resources.
Now, if you have semi unlimited resources, it makes sense to rebuild things natively for each platform in many cases. But Dropbox might not have been around, had they not maken the choice to share part of their codebase cross platforms.
As a first approximation, there are no C++ developers available. You're probably right that there are more of them than there are Rust devs, but filling a team with either one is going to be a very hard problem.
That’s super interesting. I really like Rust, definitely my favourite language recently. I’ve been thinking a lot about the pattern you are using with Rust for the shared layer. Have you got an engineering blog or something with more details?
There is a FullStory engineering blog, but as we've been pretty much heads down for a while on our product we've been unable to write up some of our internal discoveries. I'm hoping that we can open-source some of the mobile<->Rust glue (iOS + Android), and write up what we believe is effective sharing of business logic between platforms.
I would love GitHub to have a way to publish "example" code that is for reading, but not for using or branching.
I have code that contains useful parts that are worth publishing, but it also contains horrorshow code and library/tooling specific dependencies that are useless to anyone else.
I don't want to make the code useful for others, but I would remove comments/keys/PI, and parts of the code would remain useful to others to copy or use as a reference for how to do something.
The main issue is avoiding getting roasted by some random dev blogger for some of the technical debt (the embarrassing code you remove or fix if publishing properly). Maybe a click-through agreement before you can read the code?
I'm planning on learning and writing mobile apps in Rust. I remember reading that some things aren't supported on Android and iOS. Has that caused you any issues?
So far we haven't hit any roadblocks with the platforms being unsupported. I'm not sure exactly what others referred to, but in our case, our Rust code is effectively "headless".
Under the hood, Android is very much a Linux-looking system and iOS is very much a Darwin/OSX-looking system, so as long as you avoid moving too far into the UI layers, they'll behave exactly as you expect.
Oh yeah! Thanks for linking that. You are correct that Android/iOS are second tier, but TBH we have never seen anything to suggest that the stdlib is anything but 100% stable on those platforms.
I did run into a bug with the rand crate running out of randomness on Android, but that's a different story. :)
Uh oh. So immediately you do understand that you have taken on the risk of potentially being unable to find any good(or bad) Rust engineers in the future, correct? Also they won’t be cheap as there won’t be many of them.
A smarter, business-friendly, future-proof approach would be to have done that in Java. Future you will want you to have done that in Java, but if you’re trying to retain good engineers maybe I get the Rust thing.
I think many of the best programmers would jump at the opportunity to work with Rust. That might change if the language de jour changes but working with rust is a positive for many of the best people right now.
I see you got downvoted here, but I think it is worth replying. We evaluated using Java on both iOS + Android (transpiled to iOS in some form), but what tipped the scales towards Rust was the "fearless concurrency" (that turned out to match the marketing) and the risk around the current state of transpilers.
I was a fan of Java for a very long time. It was my language of choice from early 2000s to mid 2010s. It does _amazing_ things with keeping memory management out of mind (while still giving you the the "backdoors" to make things work well if you need a non-gc approach), but thread safety continues to be a glaring hole.
From a former Java super-fan, I honestly think the future is Go + Rust.
FWIW we also evaluated Go as a shared language, but there were risks around integrating our Go shared library with apps that may include some pre-existing Go code.
I was the engineering manager for a time at a company who launched a competing product to dropbox (its dead now). During my time there we deployed a common c++ core that did the business logic, and we wrote the uis in Qt with Qml for the forms. We also used swig to export the c++ objects to java. Our ui forms looked native, and they ran on every mobile os at the time (Windows phone, iphone, android), and we even used some of the Qml forms on our desktop ui.
Yes it was kind of a pain in the ass to share code like this, but it worked, and the mobile guys mostly just dealt with the specific problems relative to the mobile domain and not api integration issues with our cloud.
That being said, I can tell you for certain our mobile devs (most of them anyway) would've preferred writing their own client api in the native language of the platform. Maybe Dropbox just got tired of dealing with that, they have the money now to hire fleets of engineers to write just about anything on any platform, why would you even bother sharing a client library in that case? But in our case, we were a small team, so we really didn't feel comfortable just letting the next mobile dev who walked in the door come up with another foundation library.
Coding is hard, coding with a team is even harder.
I often struggle trying to explain to more junior developers that there are times when it's OK to write code more than once. There is a mindset that you should only ever write code once. It exists for (very) good reason. But if you follow it dogmatically, you may end up with an unmaintainable tangle of dependencies that can only be resolved with a rewrite.
Sometimes, it's OK to write the same thing twice if the alternative is a major refactor - such as inventing your own stack.
You are fortunate if you only have to explain it to juniors.
This seems pretty prevalent among senior developers too, in my experience.
The most egregious case was, this guy I worked with, who would just straight up start with abstractions to avoid any duplication and enable other use cases. Now, 9 out 10 times the use cases catered for by the abstraction failed to materialise (bloody customers) but boy did he let you know when he hit the abstraction lottery and what he'd done neatly fitted in with a new use case.
I think the term for what he created was ravioli code although maybe some of it was lasagne code
I like the rule I read somewhere called Rule of 3. Goes something like, don't bother to write a shared version of some piece of code until you've already written it in at least 3 different places. This simultaneously ensures that
1. There is actually a need to use this code in multiple places
2. You know where the places that need this code actually are, and what they need, so what you write will actually be practical for all of them to use
3. Or maybe you can tell that, despite the seeming similarity, the different places are too far apart or too different for sharing code to actually make sense
At the very least you need to repeat yourself in as much as repeating the names you defined for constants, variables, macros, and functions when later invoking them. A program written in the limit of the style of don't repeat yourself either repeats these names many times and/or has too many layers.
Hi there, sorry to hijack this post. We actually had a brief exchange a few months ago here (https://news.ycombinator.com/item?id=18133835). I didn't notice you had responded until now and that is now "locked." Are you still working on RPA? If so can you reach out at george@soroco.com? Thanks a lot!
Recently worked at a company doing similar to this to support Windows and macOS.
To add to the difficulty, they used Chromium as a UI front end much like Electron. All built in house though.
They hired me as a JS/HTML/CSS dev, but same as the article states, the C++ work was the true bottleneck.
They knew I had some C++ chops and set me to work doing C++ with a sprinkle of JS here and there.
Due to incompatibilities in the compilers, there was a rats nest of dependencies, thousands of #ifdef’s to cover differing OS implementations, and we had to use some of the oldest versions of C++ to be compatible with both XCode and Visual Studio.
As the article states, we had custom background job libraries and debugging them was a nightmare.
Also similar to the article, we spent more time than one ever should setting up builds to handle our ludicrously complicated build configurations.
And lastly, as the article stated, we had a hard time finding new talent to cover the growing amount of work required to deliver our product. There just aren’t many C++/JS devs out there.
> And lastly, as the article stated, we had a hard time finding new talent to cover the growing amount of work required to deliver our product. There just aren’t many C++/JS devs out there.
Reading the article and now this, I really can't grok this perception. There are plenty of us out here, I think, it's just we're buried in the heap caused by mass resume farming, aka modern recruiting. Recruiters and hiring managers don't think of me as a senior dev with 15 years writing software and delivering value to end-users... I'm just YA candidate who doesn't have any experience in the latest JS framework du jour, so I might as well not exist.
> Due to incompatibilities in the compilers, there was a rats nest of dependencies, thousands of #ifdef’s to cover differing OS implementations, and we had to use some of the oldest versions of C++ to be compatible with both XCode and Visual Studio.
My current work has me on a single C++ code-base for Linux, Windows and MacOS (not electron though, just native). You can avoid a lot of headaches by avoiding the platform specific tools and instead cross compile using clang with a build system that works on all platforms (eg. ninja + meson).
+1 for Clang. It integrates with everything. But like most open source options, it plays second fiddle to each company’s primary languages, and so thee’s no one IDE so perfectly integrated into these mobile dev C++ use cases as to make it easy to work with, the way, say, you can be productive on .NET framework with either C# or C++. Google and Apple are both C++ compatible, or have native bindings, but... they target their examples and ease of use at the languages that best integrate with the respective IDEs. (Swift for Xcode, Kotlin for Android Studio) The end result is possible, but it’s like picking Cordova/PhoneGap/Electron for your native app. You’re going to struggle to use native technologies at some point, and there are few examples to follow. Allowing for the realization that 80% of your app code is likely platform-specific even if the UI doesn’t vary much means you can adopt new approaches and native UI conventions faster. It’s like the difference in programming speed when you simply target Chrome for your web development vs when you’re trying to write code that runs on Internet Explorer also. It’s getting to the point where platform functionality diverges so much that writing it but from the same design/spec/data model is faster than trying to maintain compatibility. Especially if you need to support multiple versions of iOS and Android in a backwards compatible way and with different UI conventions in different OS releases. Native is native, nothing else looks or feels like native, when that’s what you need, and there’s more hand-holding for apps when they follow the latest spec from the platform, rather than try to chase it with their custom bindings/implementation...
Note re. targeting a platform — often learning and reading some C++ is inescapable. For me, reading Chrome source code makes it easier to work with Puppeteer in JS, or deliver a better web app because I know how things work under the hood and it’s easier to read the C++ than the read the web standards themselves. Similarly, C++ really is portable and fast, so it shows up in many places. But on mobile, the mix of C++ to Swift/Kotlin should be 90%-platform native if your app is a traditional CRUD app that deeply integrates with the platform. If it’s simpler than that, build a website, or if it’s a game engine or some custom UI, use that, sure, but if you want to benefit the most from Apple and Google implementations, native (with shared spec/data models) is the way to go...
I agree you can avoid a lot of headaches by not using platform specific tools.
Unfortunately, once you get into users wanting OS specific feature support, all that goes out the window.
You now are forced to maintain multiple versions of your product that differ greatly per operating system.
Pair that with the fact the codebase was almost 20 years old, you get some fun legacy code support issues that a lot of C++ projects naturally have due to the fact that C++ is not a new language.
I'm currently working on some code that is for VS/Windows, Clang/macOS, and GCC/Linux. I'm just setting the std to C++17 on all of them with no extensions. I have had some warnings that have sprung up on some platforms, but not others. But I haven't run into any show-stopping incompatibilities for vanilla C++17. What should I be looking out for? Or was this a matter of using different standards on different compilers?
First of all, agree with your decision to use C++17. Great idea!
However, in our case (like with many C++ code bases) the code was almost 20 years old. Even with the Herculean efforts made by the teams to keep the code base up to date and refactored, you run into issues of backwards compatibility with older libraries.
C++17 is definitely the go to for new projects! However, I've never personally worked on a pure green field project written in C++ that didn't have to latch onto some older libraries that require older C++ std versions.
It seems like the real issue was that Dropbox lost all of their senior C++ engineers. That’s a real mistake on their part, losing the only people who truly understand your product can be a death sentence for a company. I know my employer is very conscious of who knows what part of our products, and does their best to ensure that we never have any knowledge gaps.
You’re definitely warm. I worked at Dropbox around this time and there was very little C++ or systems experience in the building. I think it’s expertise they tried to build but for whatever reason didn’t. I work at Facebook now and the difference is night and day. FB does share a ton of code across platforms, and it doesn’t even feel controversial.
I don’t think Dropbox made a mistake in cutting out their shared C++ code, but I would caution against generalizing code sharing as a bad idea.
It was partly organizational imho. Oftentimes non mobile teams would write C++ modules. After writing the module, that team would refocus on server but the module would have remaining hard to find bugs (sometimes due to lack of mobile expertise) or could just use additional love and care. Sometimes these modules needing care and attention were even written by mobile specific teams. These were also due to insufficient organizational attention.
I've been using C++ to share code between iOS and Android for years and have come to the same conclusion. I developed the C++ code on desktop operating systems which was easier to debug, but once you move it to mobile then the debugging gets a lot harder (on Android at least).
The overhead of the JNI interface is quite a lot of work too. Luckily I had a UML model that I generated the interfaces out of, so adding JNI was just a matter of templating this out of the UML, but it was still quite a bit of work.
On iOS the situation has gotten worse in that the interface between Swift and C++ has to go through C, so if you are using C++ you need a special interface layer, which is almost like the JNI situation on Android.
So you have the trade-off of on one hand having two code bases (Swift, Kotlin) and the associated headache of keeping them in sync, versus the extra work of maintaining the C++ interface layers and testing and debugging on each platform. In the end I now believe that it is better to work with the platform native development tools rather than trying to share C++.
As far as finding developers goes, because of the rarity of using C++ on mobile there will not be a lot of developers with this experience. I work remotely and have found that any remote C++ work is very rare, Swift and Kotlin are a lot easier to find work with (it's still more difficult that you would think to get remote work of any type).
>On iOS the situation has gotten worse in that the interface between Swift and C++ has to go through C, so if you are using C++ you need a special interface layer, which is almost like the JNI situation on Android.
I’m curious why you don’t add an Objective c layer for C++ interoperability, I’ve done this in the past and it works a great.
It's mainly because I'm trying to write everything in Swift now rather than Objective-C. I still think Objective-C is a great language, but Swift is the future of iOS development.
It was easier in Objective-C as you could just rename your code file with an extension .mm rather than .m and that made it Objective-C++ and then you could import your C++ headers and use your library as you should. With Swift you don't have that option, and even the interface to C is more difficult.
Even with Objective-C there is still some work in converting data formats (NSString to std::string etc.) and you need to be careful of C++ exceptions and NSException.
It looks like that took very poor decisions such as bad frameworks and in-house made libraries vs what was already existing.
First, it's not true that there's no C++ mobile community. There are plenty of forums, blog posts, projects, using C++ in mobile field. Xamarin and Qt are helping C++ Devs go mobile since years.
Second, it depends on which standard they've used. It looks like they choose a pre-C++11 standard, making things more complicated (in 2013 there were already full C++11 compliant compilers, even for mobile), especially threads. They probably forgot that Boost itself has modules for JSON, checks for non null objects and so on.
Third, they probably kicked off the project with some average C++ developers and they left. They haven't found cheap experts (compared to Kotlin/Swift, which are really cheaper) and didn't want to pay high salaries.
The story that those Kotlin/Swift developers didn't want to learn C++ says a lot of the quality of those developers as well. Sure, it's matter of interest, but it was also a great opportunity to earn much more money as well.
As for build systems, it's true that it's an unsolved issue, but CMake is pretty much a standard and it can handle dependencies, if someone knows how it works. Sorry, but this is a story of poor choices and unskilled people for this huge job.
The cross platform landscape is pretty daunting for a solo dev. I ended up just writing my app natively, time will tell if that was the right decision or not. Most of my business logic is handled by my backend. I like the performance and being able to implement platform specific features as soon as they're released. I might look into doing some codesharing in Rust if it becomes more straightforward. Sometimes it's OK to write things twice.
This article hits close to home. I've been working with Xamarin for almost two years now. It has done its job and provided a quick avenue to prototype a concept and deliver a product. However, as the article mentions, the maintenance is a bear. Here are some of the issues I face:
- The developer experience is awful. Visual Studio for Mac causes so many issues (crashes, mangled project files) it hinders productivity. About once every two weeks I spend half a day wrestling with a VS issue.
- As AirBnB's article mentioned, to be effective in cross-platform development, you need to know three platforms well which is a difficult task.
- Xamarin's performance on Android has notoriously been poor and only has marginally improved. Try using a Xamarin app on a cheaper Android device or an older version of Android.
- No mobile engineers want to work on Xamarin.
That said, the good news about working with Xamarin is that it exposed me to C# and the .NET ecosystem. I'm truly impressed with these technologies.
EDIT
I forgot to mention that I think the complexity of an app ought to determine whether a cross-platform technology is a good idea. Xamarin is a great solution for prototypes and simple apps.
I’ve come to almost exactly the same conclusions. And it really is sad: Xamarin developer tools used to be decent, almost even good.
Today? I’d rather try to build mobile apps in COBOL than Xamarin. The developer tools are so broken, so buggy, and so poorly documented that it’s just not worth it.
I thought Microsoft purchasing Xamarin would have made things better at the lowest level of the dev tools stack - hey, finally some investment will go toward shoring up the dev stack! Nope, apparently not.
Years later, trying to connect VS for Windows to a Mac build server is still astoundingly, soul-crushingly broken. No way to get support. No straightforward way to file bugs.
Having worked in multiple places that ship iOS and Android, the real thing to focus on is cross-platform requirements, not code.
Writing mobile apps is fairly trivial - writing good requirement documents is black magic I have yet to see in the real world.
Trying to share code is like trying scratch your left foot when your right foot itches. The issue with two codebases that are supposed to do the same thing is keeping them in sync over time. The only way to do that is by having actual requirement documents.
This made me think: I've found it hard enough to share code effectively on the backend, even when we're all using the same language and frameworks.
And people think it'll be easy to share code when you're targeting two completely different platforms with completely different frameworks in completely different languages?
Add more fuel to the "write once, run anywhere" dumpster fire. Let's peek in there - I see C, C++, Java, JavaScript, stored procedures, wxWidgets, RubyMotion, GTK, QT, object oriented programming, functional programming, PhoneGap, React Native, Flutter, ... It is getting full in there.
What's that about the definition of insanity? Doing the same thing over and over and expecting a different result?
Sever side rendered HTML is the closest we've come, but that's not cool anymore.
You've listed a lot of technologies that people have used to create successful products with long life-cycles. In particular C, C++, Java, JavaScript, Qt, OOP... if anything is notably missing from your dumpster it's probably Adobe AIR.
Nothing in this industry lasts forever. I'd certainly rather inherit a Java/Swing or Qt app from the early '00s than an MFC or Carbon dinosaur.
I hated Swing the bit that I worked with it.. but I'm always blown away when I see the Jetbrains IDE's in the wild (I'm a VS Code guy but a few people I work with prefer the Jetbrains stuff). They managed to make a really slick, and pretty performant (if you ignore startup time) app with it.
I guess you could say VS Code is a similar exception to the rule in Electron land.
Dropbox was able to ship two mobile applications for 5-6 years using this strategy, so it's premature to dismiss it.
Furthermore, it looks like their team changed a lot and they were unable to retain the C++ talent that implemented the original strategy, nor were they able to find interested developers.
The mobile landscape has also changed enormously during those 5 years, even the official programming languages have changed.
I like doing projects with ionic, which have the great advantage to be ported to the web and also packaged as desktop apps. From all the apps i have developed with web tech or even i know of no users are complaining about lower performance or similar. Why developers have this opinion?
What specifically isn't elegant about Dart in your opinion? Seems like a perfectly good language to me, and although it's nothing fancy, it does have a bit of sugar, and a solid standard library. I also like how codegen and static analysis are so accessible through the build libraries.
It's evolving pretty quickly, too. Non-nullable by default is coming up, along with extension methods, FFI. Possibly implicit conversions.
I've been working with Flutter/Dart for the past 8 months or so, and I have to say...I'm pretty into it.
The nice thing about React Native compared to Flutter is that RN uses native UI widgets whereas Flutter implements its own rendering system. Flutter engineers basically reinvent a large portion of the work already done by Apple and Google in their UI kits and ultimately you have to rely on them to keep doing that work whenever changes are introduced to iOS and Android UI kits. Since RN uses native widgets they get those updates "for free". There is also the Flutter "uncanny valley" where very subtle but still perceptible differences between the Flutter engine and the native UI kit cause Flutter apps to sometimes feel slightly off. I have especially noticed this on lower end phones where the FPS in Flutter apps sometimes drops and causes the whole UI to feel sluggish in a way that native UI widgets do not.
In my experience it generally does, but on older devices, especially older androids, when the FPS does drop, the entire UI appears to be affected, whereas native widgets seem to perform better, or at least, don't seem to drag down the performance of the entire app UI if one widget is misbehaving.
This has been exactly my experience. I don't particularly like Dart, but as an SDK Flutter is clearly ahead of RN. RN just keeps feeling too fragile. And the tooling is simply not there whereas Flutter has excellent tooling considering how new it is.
wow! I didn't know that. I am currently learning Flutter, because I think it has a strong potential to unify mobile and desktop development, however, I didn't know that React Native was "so bad".
I had the opposite experience with Flutter personally, the problem is that Dart is a niche language with limited exposure (Flutter is basically the only big project in Dart realistically) and it's also a new project as well, it makes a bad combination.
Could you describe your experience more in details? What exactly didn't work out? I would like to know to understand whether it makes sense for me to invest my time in Flutter/Dart instead of a more mature solution like React Native.
Actually, I don't think we should consider React Native more mature. It's been around for longer sure, but it's significantly more fragile, keeps releasing with regressions, has poor tooling around it except for as far as React is concerned etc.
I don't know what you are trying to build on mobile but I'd definitely learn Flutter instead of RN, which I have. Others here have said you should just go full native instead, and they're probably right.
I wasn't aware of the fact that react native is fragile - it seems I did the right choice by learning Flutter. I hope it evolves even better, and also the criticism to Dart for me is so far quite irrelevant, as there are already tons of libraries and it's really easy to learn.
May I ask what kind of stuff are you developing with it? I've been working on a WebRTC app myself with it, basically porting my app from RN to Flutter. So far I've encountered no major obstacles, just a little friction which is a part of the learning curve.
The flutter approach is interesting, however i tried really hard to become friend with dart, which seems impossible to me... I only hope somebody is building a transpiler or something.
wxWidgets does actually get close to that mark. But regardless of the library/programming environment, there will always be platform-specific code you have to deal with.
I don't think wxWidgets is a strong example for cross-platform success. While admittedly I haven't touched wxWidgets in years, back when I used it, I marvelled that it worked at all, as the source code was such an utter mess. For example, I found a bug where a class constructor would raise lifecycle change events before the constructor even finished, and the issue was brushed away on the mailing list. Argh.
I really despise Electron apps. They always seem bloated and antagonistic toward the native user experience on my platform (eg. inferior keyboard support, hard to integrate with custom tooling like AutoHotkey, etc.). Have I just not used the right ones?
I'm not a fan of them, but they do make desktop linux a bit more viable insofar as you will have desktop binaries for a lot more services than you otherwise would without electron
If the user interface is usable and snappy, it is probably fine that it consumes twice as many resources as a similar native app would require for the same result.
Example: Slack.app is a huge memory hog (though not as bad as it was before their rewrite), but it fits decently into OS X and does not lag on me.
I have lots of RAM, so the extra consumption is fine.
Sure, it annoys some programmers that it could be more efficient, but other than them, who cares? Those guys are probably using Emacs or a CLI program as their client anyway.
(I must be a rare breed, since I do email, writing, and programming in Emacs, but not chat)
I think it's a bit of an unfair characterization. I don't use emac or CLI as my client, I enjoy a beautiful UI as much as the next guy but electron is a real bother to me.
I switched from Sublime to VS Code and went from 20MB of RAM usage to 500. Add slack, add discord, you now have GBs of RAM getting eaten up just for the UIs. Most mid-range laptops still only have 8GB of RAM, add a browser to the mix and it's a constant battle to have everything running smoothly, the programs frequently crash due to memory issues.
It's a real problem and not solely programmer efficiency-OCD.
While it's fair to assume most Electron apps are probably running old Electron (and thus Chromium), Electron itself has done a bunch of work to get caught up. Their 6.0 release was the first release to be shipped on the same day as the equivalent Chrome release https://electronjs.org/blog/electron-6-0 their plan is to stay at most one version behind https://electronjs.org/blog/12-week-cadence
My laptop has 8 gigs of RAM, and I run many programs simultaneously (usually including several codebase analysis servers behind Emacs, a VirtualBox VM or two, Slack, and of course tens of Chrome tabs, several in development mode), but I don't have anything like the problems you describe. I do use The Great Suspender aggressively in Chrome and that does help a lot.
My Emacs instance itself is extremely tricked out with extensions and usually weighs in between 100 and 200 MB of RAM. Maybe that's the difference - sounds like VS Code eats a lot more.
I'm on a MacBook Pro from 2015 at work, if memory serves.
What setup are you running that handles your needs so poorly?
I'm impressed you can run all that especially multiple VMs on 8gb, I have to close Firefox when I need VMs.
I'm on windows 7, on a Dell Inspiron with a RAM and disk upgrade, no special configuration or software. I didn't think to look for an addon to do that, that's pretty cool, I downloaded the Firefox equivalent, might do the trick!
God I wish it was only twice the memory. Twice would be acceptable. Discord by itself likes to push to a full gigabyte to show me 50 lines of text with avatars and 0-5 thumbnails of images.
Even worse it's using significant amounts of video card memory, and I have a tenth as much of that.
Not if it results in only this one application feeling snappy and everything else hanging because of resource consumption. That's especially annoying when you work on a shared server. I used to work in a C++ team where many people used CLion (on a shared development server) and it made us all miserable, because the server would hang from time to time, because all the CLions ate all the RAM (sometimes just one instance could consume more than 50GB). Similar things happened with Eclipse, but a lot less often.
And then the IDE of another guy does its indexing really fast zillions of times per second, great, but then another guy sits there waiting for "git status" to complete for several minutes.
1.5 years ago Dropbox was looking into rewriting their shared mobile codebase in Rust. It's a shame this blog post doesn't touch upon that at all, instead focusing on C++.
> That being said, C/C++ are the only languages with a compiler supported by both Google and Apple, so using a different language would have created a whole host of other problems to deal with.
1st tier tooling supported by the platform won't break yearly on every new OS version and you don't have to wait until your 2nd tier tooling developers gets around to fixing it.
The platform developers will do their best to keep 1st tier tooling working and fix it when it breaks.
Likely too much friction. Native environments benefit from using the expected IDE and tooling integration to build UI apps on respective platform. C or C++ is the best common supported language, and they didn't even like that.
Also the UI frameworks are being written in said preferred language these days, eg SwiftUI, and adding one additional language could be perceived as a pain.
Apple ships a C++ compiler but you can't use C++ with Apple's platform SDKs. IF you want to use C++ you need to write Obj-C++ glue code to access most of the system frameworks, including UI code.
I would look at it differently, i.e, obj-c++ is a valid supported language environment, that makes interopt with C++ pleasant at least compared to other interopt choices. Developers may choose to target Obj-C++ even out of preference.
Metal shaders != Metal API. Metal API is Obj-C. Metal shaders are written in MSL (Metal Shading Language). C++ is still not involved, though MSL is admittedly based on C++.
"The Metal programming language is a C++14-based Specification with extensions and
restrictions. Refer to the C++14 Specification (also known as the ISO/IEC JTC1/SC22/WG21
N4431 Language Specification) for a detailed description of the language grammar.
This section and its subsections describe the modifications and restrictions to the C++14
language supported in Metal."
Also, the MSL and Objective-C compilers are built on top of LLVM, written in C++.
I don't know what point you're trying to make. MSL doesn't link with the rest of your program. No matter what language you use for your program, the shaders are written in MSL. For my program if I use Rust or C++ or Obj-C or Swift or a microscopic magnet and a very careful hand, I write my Metal shaders in MSL.
The fact that Metal itself requires the existence of C++ really has nothing to do with this topic though, which is whether there's any downside to writing your app in a language that Apple doesn't ship a compiler for. The fact that Metal shaders are written in a language based on C++ is irrelevant to this question, because it doesn't matter what language you pick for the rest of your app as Metal shaders don't link with your app. Swift has zero integration with C++ and yet I can write a Metal app using Swift, because the fact that the shaders are written in MSL has no impact on the fact that the app uses Swift.
And the language for writing drivers in is even more irrelevant because we're not talking about writing drivers. You can't write drivers in Swift either.
It baffles me that people think duplicating application code for every device is the way to go. Cross platform can be done correctly. I'm working on a stack that can natively target Windows, Mac, Linux, Android, iOS, MacOS, and Web. I couldn't imagine duplicating my code for each one of these devices. If something special needs to be done, I write an extension.
I have a feeling posts like these get upvoted because developers who make the native apps want to keep their jobs.
But those developers will always be needed for the native bridges, just not as many of them.
I'm very much in the cross platform is a bad idea camp. It's fine for sharing business logic and other non-UI code, but having tried several cross platform UI kits I've never found it to work well. Cross platform UI always ends up compromising on the native UIs strengths and uniqueness resulting in poor UX, poor accessibility, unaccceptable resource usage, bloated binary size etc. Just a few concrete examples:
- Non standard UI/UX is super common because cross platform UI targets the lowest common denominator. Flutter is the quintessential example of this, it throws away all native UI support and renders everything from scratch with low level graphics. In React Native I've observed several cases where iOS UI has been re-implemented, poorly, in Javascript and shipped on Android and iOS.
- Poor accessibility. Native iOS apps get a lot of accessibility for free because Apple has done a great job, with RN and especially Flutter you don't get this benefit.
- Resource consumption. Electron is terrible in this regard as is the common meme. React Native runs a JS VM for every app which surely has a negative impact on both battery and performance.
- Binary size. A basic Flutter/RN app easily pushes you past 20MB in app size on iOS, whereas a complex and complete app like Tweetbot is ~7MB.
It's certainly possible to build great UX with cross platform UI frameworks, but the amount of work required to do it well starts to get into the territory where you are better of just doing it natively in the first place.
React Native can wrap any native component and bridge it into your app. If you had bad experience with a 3rd party component, that's not RN's fault, build your own extensions you need, but for the most part you just need the core. It's still easier than rebuilding your app X times.
- React Native you most certainly do get accessibility.
- Running a JS VM isn't that bad, but it is a tradeoff. Imo it's completely worth it to prevent having to learn X different languages for each platform.
- The app size evens out, but yes the minimal is higher. However many apps are much larger than 20MB, it's better than other cross platform solutions
In no way are any of these things so bad where I would duplicate my code 6 times and then maintain them throughout the life of the project. That's just crazy.
> React Native can wrap any native component and bridge it into your app. If you had bad experience with a 3rd party component, that's not RN's fault, build your own extensions you need, but for the most part you just need the core. It's still easier than rebuilding your app X times.
Yes a great strength of React Native is that it still leverages the native UI kits and makes it possible to do so when building custom components. My comment wasn't about my experience, in fact I've built numerous bridged components in RN, but and observation about the react native community. In fact one of the most crucial lessons I've gathered when working with RN is that you must have both iOS and Android experts in your team to build good apps with RN successfully.
Here's just a few examples of popular 3rd party components(from awesome-react-native[0]) that are problematic in my eyes:
- react-native-swiper[1]. Completely custom JS implementation with a "iOS"y look. 8k stars
- react-native-gifted-chat[2]. Large completely custom JS implementation of chat UI. 8k stars
- react-native-slider[3]: A pure JS implementation of a slider. 1k stars
> - React Native you most certainly do get accessibility.
My point wasn't that it can't be done, but that it requires significantly more work that, let's face it, most developers will not do. With native UI you get a lot more for free which results in a better user experience for everyone.
> - Running a JS VM isn't that bad, but it is a tradeoff. Imo it's completely worth it to prevent having to learn X different languages for each platform.
I agree it's a tradeoff and to be fair I don't have more than a hunch that running a JS VM is bad for energy and resource usage. The tradeoff is distinctively on the user's expense in favour of the developers. As developers shouldn't we be making tradeoffs that benefit our users instead?
> - The app size evens out, but yes the minimal is higher. However many apps are much larger than 20MB, it's better than other cross platform solutions
Yes many native apps do grow quite large too, my point is the default. The most basic RN, but especially Flutter, app is much larger than the native counterpart. Again a distinct tradeoff on the user's experience in favour of the developer.
Some of your statements are partially correct, but a few of these assertions are flat out false for React Native. Let's look at them:
> - Binary size. A basic Flutter/RN app easily pushes you past 20MB in app size on iOS, whereas a complex and complete app like Tweetbot is ~7MB.
I have working examples of fairly complex production applications that are <=10 MB on iOS.
> - Poor accessibility. Native iOS apps get a lot of accessibility for free because Apple has done a great job, with RN and especially Flutter you don't get this benefit.
False statement. React Native provides you with ability to tag elements so you can use accessibility-tools to navigate the app. I have recently tested this with a RN app running on iOS 13 using the new voice control feature. Several testing frameworks also rely on these features being available.
> I have working examples of fairly complex production applications that are <=10 MB on iOS.
Fair, the problem isn't as bad for RN, for Flutter it's still very bad though.
> False statement. React Native provides you with ability to tag elements so you can use accessibility-tools to navigate the app. I have recently tested this with a RN app running on iOS 13 using the new voice control feature. Several testing frameworks also rely on these features being available.
The ability to tag elements is only a very simple basic requirment to get accessibility right. My point was more that if you use UIKit you get significant levels of accessibility for free.
Just as a quick example I did a quick ddg search for "react native how to create custom button" and all 5-7 top results that I looked at failed to set the appropriate accessibility roles so the button actually appears as button to a user with the screen reader. Yes you can still do accessibility right with RN, but it's not the easy to get it right be default and most app developers will not do this on their own.
> failed to set the appropriate accessibility roles so the button actually appears as button to a user with the screen reader
Appreciate that you looked into it. On iOS and you do not need to set anything to become screen reader friendly _in React Native_. The default TouchableOpacity with a simple piece of text inside of it will become focusable and iOS will read the text aloud for you. This works because RN uses native elements unlike Flutter, Electron, and others.
In fact, I took you up on the claim that it wouldn't read correctly. I used my app just now which has never been tested or optimized for accessibility. Covered my eyes and began navigating. It works just fine! All tappable elements are indeed read aloud. I was able to navigate through the app settings, change them, go back, and reload other screens, navigate to the updated content, and perform actions on the new content.
Please do not put RN in the same bucket as Flutter or Electron. They are fundamentally differently.
There's a lot of valid points from native advocates though. Like differences in touch features, various sensor APIs, Firebase etc. These may not be serious considerations in simple apps but they make major difference to both the end user as well as the codebase/developer in larger scale / more serious apps.
That said, myself I still write Flutter too, having jumped ship from React Native.
But on React Native there are major platform differences that you also have to bridge. Moreover, as the platform essentially uses each platform's native components through that JSON serialized bridge you'll have UI inconsistencies between platforms when sharing code.
I took this as an argument for a thin, or at least thinner, client. If something is complicated and costly to maintain - shouldn't it move to the server? What complicated processes are you doing on-device that couldn't be done somewhere else & fronted by an API?
Done correctly != react native. I've had to shut down and migrate 3 RN projects because of how much of a mess it is. It's not even 1.0 and is a horrible choice for important apps.
> Because this issue involved debugging multi-threaded code running back and forth between C++ and Java it took weeks to nail down!
I don’t think the issue here is the use of c++ — it sounds to me like it’s the use of c++ on problems that would have been far simpler solve on the native code ... not all problems are that way — but even where it makes sense to solve with c++, you have to be careful about the glue.
I developed and maintained cross platform c++ for iOS/Android in a previous role and the platform to native glue is inherently complicated — among the most complicated parts of the program to reason about if you get into the details of it. I solved by making this layer dead simple — you could almost never pass objects in or out — just primitive values.
The c++ layer exposed to platform was just static methods — including a bunch of methods for retrieving state and we added more static methods when new kinds of communication were required. Every c++ method grabbed the largest lock it could prior to doing anything natively (and in debug mode related methods associated with some kind of stateful lifecycle would fatalError if they couldn’t get the needed lock — the app code shouldn’t have been calling into native code in a way that violated simple assumptions about how the state was allowed to change).
Interestingly — this forced the platform layers above to comply with the assumptions needed to keep the native code simple — and helped those working on the UI immediately find out when they were violating these assumptions — things like click handlers that weren’t debounced or weren’t switching to a disabled mode after being clicked — when those click handlers invoke native code and that native code mutates state — you have a potential nightmare scenario for robustness ...
This model actually was quite straightforward to add new functionality to over time and we frequently moved decision authority over complicated logic into the native layer so that we could make it robust and cross platform ...
The best way I've seen this work (inside Google Apps) is to write shared business-logic in a common layer (here it (often) gets written in Java, then transpiled to Obj->C and Javascript, other teams go from C++ -> Java & JS, and other other teams embed JS inside of everything.. and other other other teams write everything in a native language (there is no one single way at Google))
then use a data-definition-language (like protobuffers) to define a bridge from inside of transpiled code land to your native layer, and compile native language bindings for your UI to read & write data models.
then stack a mechanism like GRPC which takes the DDL to the next level and makes services a declarative, language agnostic definition... and compile your service stubs for each platform.
...then write the UI in a native layer for each platform, using the native service stubs and data models compiled from your DDL.
Once you start trying to abstract the UI models between iOS, Android, and web you are stuck with a shitty status quo solution that leads to terrible hacks or boring concessions that ignore platform differences.
However abstracting the common (non-ui) mechanisms like asynchronous data structures, sockets, and storage is a relatively solved problem.
The overhead of C++ adoption actually prevented us from ever moving fully in this direction.
Today, unlike in 2013 when they started, there are other options.
For others considering code-sharing, another option not mentioned in their article is Rust for the core with Swift and Kotlin. Between bindgen, futures, serde_json and non-nullable pointers in Rust, those would satisfy their stated subcategories today.
Companion templates are explained here: (January 2019)
Medium.com/visly/rust-on-ios-39f799b3c1dd
Medium.com/visly/rust-on-android-19f34a2fb43
(No affiliation other than starting to proceed down this path myself.)
For Common Lisp heads, there's also MOCL, which seemed quite reasonable when I explored it years ago: wukix.com/mocl
I don't know enough details to say for sure, but some possibilities:
1. People may be more willing to learn Rust. I much prefer the language itself, but also it's clearly on the rise, so it may feel like a better learning investment.
2. The standard library is broader and the package ecosystem is easier to work with, so you might be able to have a more standard stack with fewer proprietary bits.
3. One complication with integrating multiple languages is making different memory management strategies coexist. Rust's type system seems to have ways of making that easier, e.g. neon-bindings.com
It is really easy to make a blunder in c++ that will require a lot of effort to find. Debugging that on mobile platforms is not fun. Like they mention a deadlocking issue that was difficult to solve.
Rust is built for memory and thread safety. Also the much better build system and open source ecosystem.
Its pretty evident from things they said in this blog post that they want to stay firmly in mainstream. They mentioned C/C++ is the only compiler supported by both platform vendors, which is important due to a "whole host of other issues" that they didn't specify. So using a compiler developed by a company that doesn't even sell commercial support would be a non-starter. Still there are commercially supported platforms that are quite mature at this point, such as Xamarin.
This is precisely the experience of my startup trying to rely on flutter. It's a constant battle. Moreover Android and iOS are different implementation with different capability with constantly evolving API. It's hard to keep cross platform code in sync many #ifdef with edge cases.
Moreover with REST API architecture majority of the common code is in back-end. So building in Swift and kotlin for respective platform is not as tedious.
I've been building a Flutter app and it has been fantastic. I'm using one shared library flutter_barcode and while it isn't perfect, but it shows how easy it is to create native solutions for both platforms and marry them within Dart/Flutter itself.
I can't see how duplicating code in Swift/Kotlin would be any better than just writing it once in Dart. Dart is also quite a nice language to work in. Easy to get up to speed and a nice mix of static/dynamic typing. The optional Provider api make state management a breeze, no more Stateful widgets.
I've also switched from REST to GraphQL. I was a long time hold out there, but now I really see the advantages.
Anyway, just throwing this out there as another data point.
I thought exactly like you and was my view, until I tried to make a small app in Flutter. My startup was trying to build something like Google Lens with AR, and no way to do it in flutter without native code easily.
I like dart lang and done some small apps in it. It's Javaesque with similar style and learning curve. But still prefer Swift which is closer to python (we use python for back-end development) and Kotlin which has some nice functional programming primitives.
Also UX and guidelines are not uniform across mobile platforms. Android and iOS works in their own way to integrate different input like voice, visual and keyboard. Also mobile platforms on Android and iPhone are constantly evolving with new sensors, capabilities. Keeping native SDK itself is a gigantic task not sure how to manage cross platform SDK's. I am sure cross platform SDK can work for some use cases, but I feel it will still be way behind native SDK.
If back-end can take care of all the business logic and front-end manages UI and states with small necessary local data for offline, the efforts spend to develop cross platform app using flutter or natively in Swift and Kotlin won't be very different.
I still prefer REST over GraphQL because I like predictable performance in query and standard interface. Zen of Python PEP-20, explicit is better than implicit.
GraphQL is a solution to a problem of trying to manage business logic on client side so that it can manage ad-hoc query and get ad-hoc results from back-end. It was designed for a problem faced by Facebook. I have seen enough GraphQL code and still stay with REST Architecture, sometimes use protocol buffers for performance intensive parts.
Could you explain what you think of the pros and cons of flutter VS IONIC?
I think flutter and it's astonishing number of issues on github + their lack of manpower + their lack of major features + the presence of major performance and behavior bugs + the extremely small lib ecosystem (dart) is a major, useless risk for a startup for the benefit of tech hype.
(just my point of view, don't take it personally)
Ionic lets me build a web app and publish it to any device pretty easily. I don't have to use their components either... in fact after using Ionic, you may not use Ionic anymore at all. All you need is Capacitor to publish it.
Flutter is it's own rendering engine, non-standard web language (Dart), and ecosystem. Ionic is just a CLI and collection of web components. Capacitor is Ionic's version of Cordova. It gives you access to device capabilities such as storage, camera, etc with very little code. So you can build a website, ask if it's Native on a device, and do more cool things if it is. Then push it to Android, iOS, and the Web.
Ionic supports Angular, React (now in RC0), and Vue (Beta). This allows you to learn popular frameworks while still developing with one code base.
Lastly, Ionic's Web Components are built using standard-compliant Web Components generated by a tool they made called StencilJS. So, now you can build a component in Stencil and publish it for simply the Web (No framework) or have them work with Angular, Vue or React. Larger corporations have started using StencilJS to manage their components when they use many frameworks. I use it because I don't know which framework I'll use next, or if I'll even use one.
Cordova is time tested and has a large community behind it. I choose Capacitor if it's available and then go to Cordova if it's not. Ionic has a collection of easy to use Cordova plugins (https://ionicframework.com/docs/native/overview). I have honestly found they they are not that well maintained. I ran into enough issues that I avoid them if I can. To be fair, the contributors are people contributing on their own free time with no financial incentive. Hence, I choose Capacitor when I can. They have a financial incentive (Ionic is open about this https://ionicframework.com/blog/ionic-2019-business-update/)
I joined when Capacitor was in beta and figured I'd use the up and coming. From research, it appeared there was a good reason Ionic created their own product. They didn't do it to simply stop using a tried and true product.
Also to be clear, Cordova is not an Ionic product. It was a great community thing for a while to talk to native devices.
I haven't tried Ionic myself, but I have tried both React Native and Flutter for a small project and found Flutter to be more enjoyable and productive.
I build a small ionic app when it was based on angularjs with dependency on Cordova and later moved to webview. I didn't like the experience and personally I still prefer Python syntax over Javascript, so that was another factor. Now I think IONIC has moved to web components using stencil.js as the base. Web Components is still an evolving standard and will still take time to gain traction, moreover the available documentation and help is limited.
So if you compare ionic vs flutter I will still go with flutter given it generate binary app store package file by compiling dart code into native platform code. But given choice between native vs flutter I think except few use cases native still makes more sense.
I get the API differences on these platforms but at least in terms of the UI Flutter seems like a good choice. I'd imagine it'd still be easier to stick to Flutter than write two separate native apps in most cases, it seems to have gotten tooling right.
So I'm also interested in hearing more elaboration on how Flutter has been a struggle for you.
That's my point, subtle difference in UX like gesture, swipe and touch controls in iOS results in additional UI code in dart to account for each platform. Not much help except for simple mobile apps doing simple crud operations.
If I use swift I get all the code examples and tons of documentation from Apple itself. In flutter I need to rely on some package which might not be supported, if my team takes over it's double the job fix the package bugs and also change the resulting changed in code, compile and test individually on iOS and Android.
Having done this quite a few times, it looks more like they are having issues to retain C++ developers, are too picky on their hiring process and the original devs have left, than anything else.
And given some of the macOS client practices, the engineering quality is also to wonder.
For me, C++ with native views, or Xamarin, will keep on being the way to do mobile apps that can't be done as PWAs.
What I don't see them saying was that it was the wrong choice in 2013 for a scrappy team and an evolving product, just that it's not the right choice in 2019 for a company with ~infinite dollars and a stable product.
I'm looking at building a mobile product solo right now, and I'm finding Flutter a pretty compelling idea. Building everything twice means being circa twice as slow, meaning I can learn about what the market needs half as fast.
I might be wrong, of course, but this isn't the article to persuade me otherwise.
I don't understand why someone at Dropbox's size and scale would go after C++ on mobile and blaze their own trail, when its probably more pragmatic to go native.
With that being said, I worked for a firm that extended the life of old ERP systems and we had to do a few mobile apps, we chose Xamarin because we were a shop of 7ish devs that had many projects to maintain and the cost of code sharing and familiarity with C# were our driving factors. It was a trade off most definitely but it was a pragmatic choice as well.
If you are capable of hiring more developers who know Java or Obj-C and can allow them to do only mobile development then its worth it to go native.
But, if you are a small company and know that Java,C# and Obj-C, C# developers are a little bit hard to find (depending on your developer market) then its probably more cost effective to go Xamarin or any other cross-platform code sharing model.
iOS requires AOT compilation as interpreted/bytecode languages AFAIK are restricted to the Nitro JS engine. With Android it compiles with a .NET runtime included IIRC.
I'm somewhat disappointed that this didn't work out, as I have usually promoted a C++ core along with a thin native wrapper as the solution for those looking to share code across platforms. At least they moved to native development, though, instead of some poor facsimile that was "easier"…
Curious about how well having that C++ core to share code has worked out for you in the past. Apart from low-level contained code that doesn't require access to too many OS interfaces (user I/O, network, threading, GPS etc) I can't imagine it would be easy or useful to have a C++ core.
Usually C++ runs core logic that cannot go on the server (that's things like your own custom client-side crypto, parsing, or other "pure computation") and then you make the user interface in Swift or Java depending on your platform.
Interesting post. I wonder if the major pain point was the fact that they were doing dev with a language that had nothing to do with either platform.
We’re currently investigating Kotlin/native and it looks really promising. We don’t have to write stuff in a foreign language, but can repurpose our existing Kotlin code and share with iOS. This model seems promising.
Does anyone have any experience with Kotlin/native?
Good points. I guess I didn’t mean that the language wasn’t valid or supported, but rather, that it wasn’t the main first-class language of the platform.
Indeed they have. What’s missing in some of the discussion here is the size of the application. For large applications like Office, a shared code base is often the correct strategy.
I work on Office and am glad someone brought up this point. Writing a C++ business logic layer is probably too much for a simple CRUD style app, but the equation changes once the view/platform layers are a tiny minority of the logic in your app.
It's funny that they have a hard time hiring experienced senior C++ devs even though the language has been around so long. Presumably it would have been easier to find someone really good at something newer like, say, Vue.
Let's be clear that senior C++ developers are masters of their domains because they enjoy solving a particular set of problems that a low overhead systems language requires.
However, in the mobile world, from my experience, hiring good senior mobile engineers means they come with a specific skillsets already. They can take a view, animate it a certain way, and then respond to touches. Those are things that are entirely platform specific domains. So they will come from Swift/Kotlin or ObjC/Java backgrounds. The ultimate problem here is those same engineers are typically turned off by C++ nor have the desire to learn it.
You can throw C++ engineers at the problem, but they will most likely be turned off by having to solve animation timing issues or custom navigation stack designs.
If you have a venn diagram of the two sets, there would probably be very little overlap.
This matches my experience - most C++ engineers I know have little desire to work on C++ for mobile after having tried it once. Most mobile developers tend to be specialized on mobile and don't focus so much on generalist C++ stuff. Some good ones do both and don't have problems finding work. :)
> it became increasingly difficult to hire replacement senior engineers with relevant C++ experience who would be interested in mobile development.
It's not just familiarity with the language, it's being willing and able to develop within the mobile development context, which as stated in the article is a whole other beast
Mobile is getting crappier, so I can see why C++ developers would be reluctant to partake.
Google and Apple keep piling feature on top, so mobile has the 2nd biggest churn after front-end. Recently even the preferred programming language changed on both platforms.
It's under pressure from web-applications, which although of inferior quality are easier to develop.
Most apps aren't anything special and require mostly UI and UI framework skills (animations, platform-specific APIs, etc).
Yeah, I have 15 years of C++ experience, but not since 2012. So with C++11 and C++17 (and soon C++20) my skills have not aged well, even though I can still write C++03 just fine. It's no wonder 'senior' devs prefer to use plain C, where even C11 is not that different from C99.
Funny, I too stopped using C++ around 2011. Recently I had to delve into Swift and learn it with great urgency... then my project changed direction toward C++ and Qt.
I find the '14 and '17 additions vastly improve C++ and in fact added things that I had just come to appreciate in Swift. So don't be too put off; learning the new stuff will help you if you need to adopt a more-recent language too.
I don't think that's necessarily the case. C++'s famed backwards compatibility, source of so many issues, is also a fabulous strength. There's nothing stopping one from writing C++03, while gradually adapting C++11 and beyond as time/circumstances permit (which is what I'm doing at my job, maintaining a massive legacy codebase while stymied by slow uptake on newer compilers by my corporate overlords...)
> It's no wonder 'senior' devs prefer to use plain C
Not this senior (in years, at least :)) dev. I've tasted RAII and I can't go back...
The problem from Dropbox's perspective is that they're not merely hiring someone to create a solo greenfield C++ codebase, but to contribute quickly to a codebase written in whatever arbitrary subset of C++ they blessed.
I used to maintain a couple of systems that between them were around 4 million lines of C++. Initially, in 2004, it would take about 16 hours to do a clean build. By the time I left that job, due to better hardware, precompiled headers and reduced dependencies, and smaller code for same functionality, clean build was down to 45 minutes.
The biggest winner, in my experience, was reducing dependencies in headers. Use forword declarations, etc. Precompiled headers helped a lot on Windows, but less so on Linux for the same code (but it still helped).
Incremental builds were fairly snappy, a minute or two (pretty good for a large C++ project).
I recently had to use a bit of Scala for some load-testing, I've never had so much trouble doing something seemingly so trivial; in this case, sending a JSON HTTP request. It was honestly a matter of hours. Pretty humbling experience!
I'm also in that approach for the same reasons. Currently I have an app I'm cross-developing with separate iOS / Android codebase. the only 'shared-code' is Firebase libs for each platform (which seems to be written in itself cpp and bind to Kotlin / Swift).
C++ itself is pretty portable.
There are actually frameworks (QT / JUCE) that allow true cross-platform builds.
They work, but you'll eventually end up having a lot of branches for specific platform.
Still if you just need basic algorithms (signal processing or anything expensive CPU based) C++ might be good enough to write simple callbacks with bindings.
Evil UX/UI:
-----------
Xamarin, Flutter, React Native -
It really depends on the app. but if the app needs to be in the "device" UX and the unique APIs provided by the platform. people can glorify the cross-platform all they want but,
You'll eventually end up with branches for each platform in the good case and bad performance in the worse case.
Using C++ for libraries was not a bad idea but I think they got the architecture wrong. They should have used a protocol like gRPC streams or flat-buffers to communicate between core libraries and the platform UX libraries. Basically the mobile app is treated as a frontend talking to a backend using a communication protocol. You can design this to be stream-based, event-based, whatever fits the boat best. Depending on the platform, you can choose what should be part of the frontend versus what should go into the backend.
This is the way a lot of the big C++ mobile apps are built. Also there are lesser debugging headaches as each part can developed independently thanks to de-coupling, so you can hire native platform developers to do the job they know best without messing with C++.
I worked in a company that makes games for both ios and android and it shares majority of the code in between. It was using libgdx which uses Java & opengl and it was working very well for us.
I would say things are not so bad when you pick your tools right.
I think the use case Dropbox has is quite different, because they have a lot of native code surface, needing lots of interactions with the shared code base. A game can get by with only a few lines of native code, AFAIK.
The common thread that I personally see between dropbox and airbnb, though completely different language sets, is adopting a shared codebase in a "brown field" manner.
Airbnb admits if they were able to greenfield React Native then there's a world where it would have worked.
Separately, the complexity of Airbnb and Dropbox as apps is very high compared to what I believe are the ideal use cases - simple interface apps that are thin wrappers on top of APIs.
How's was your experience in writing iOS and Android app in Cordova? I have a small community app written in Cordova+f7+vue2 which works pretty well on Android.
I want to port it to iOS now as there is some demand. I also want to add push notifications later. So far there is a inhouse API+server which serves as backend. Not dependant on Google for any services as of now; only if there is a non-firebase solution to push notifications?
I wish all the articles and blog post are like that. The first sentence sort of gives away what you need to know.
Until very recently, Dropbox had a technical strategy on mobile of sharing code between iOS and Android via C++. The idea behind this strategy was simple—write the code once in C++ instead of twice in Java and Objective C.
A lot of people will stop reading after this, for those interested they could continue with the details.
TLDR - it's too hard to find senior C++ mobile devs.
I'm more intrigued by AirBNB moving away from React Native - the linked article says "RN was too small a component to bother supporting, and the developer experience wasn't up to par", but I'd like more detail than that.
I've been working with Flutter for the past 6 months, and it would definitely be my "go-to" for any mobile application.
To be fair, I did spend 2 weeks going down a rabbit-hole to get Flutter talking to a .NET assembly by embedding Mono and invoking via JNI/Obj C, so I know the pain of native interop. If you're mostly doing work at the native platform level, then I can imagine why you'd stay away from a cross-platform VM.
In my case, I ended up with a brittle project that was clearly going to be a PITA to maintain & automate builds for, so I just rewrote the component in question in Dart.
AirBnB has a series of in depth blog posts about their decision to abandon RN. In particular this quote seemed extremely worrying to me:
> While debugging, React Native attaches to a Chrome Developer Tools instance. This is great because it is a powerful debugger. However, once the debugger is attached, all JavaScript runs within Chrome’s V8 engine. This is fine 99.9% of the time. However, in one instance, we got bit when toLocaleString worked on iOS but only worked on Android while debugging. It turns out that the Android JSC doesn’t include it and it was silently failing unless you were debugging in which case it was using V8 which does. Without knowing technical details like this, it can lead to days of painful debugging for product engineers.
Having been bitten by this exact issue one or two times, meh. Par for the course in the life of a Javascript dev, really.
RN would definitely still be my go to tool for the majority of green-field mobile apps. If I'd already invested in learning flutter/Dart I'd probably feel the same about that, but certainly feel no urgent need to go out and learn it. As in, I don't think there's a significant difference in what I'd be able to accomplish, and my time is probably better invested elsewhere.
That's great from a dev perspective. Not so much from a product perspective, particularly if you operate at AirBNB scale and your issue could be costing lots of money for every second of failed debugging.
The best is when the app runs faster while debugging than it does on the phone because of this exact issue. I found it nearly impossible to profile sanely
Is flutter actually a good choice for a production app right now? I've heard really bad things about its bugs, but I admit it does look great to use and develop with.
It depends. If you want to write nice looking material design app, then yes, it works really well. As iOS developer I can say that flutter is much more pleasant than storyboards & UIKit (which is probably why Apple is working on SwiftUI).
It's a different story if you want to use native features. Plugins provided by community offer only basic functionality, and are often very buggy: you will encounter iOS bugs, Android bugs, Flutter bugs and plugin bugs, all in one package. Anyway, it sort of works if you plugins provide stuff that you need. If they don't, you are forced to fork the plugin and add ad those features, learning Android or iOS during the process, and writing a lot of boilerplate for communication.
Due to C++ inherently close connection with the OS I see this as an absolute happening. But they should've go with Objective-C and Java and stay with them, as they were the matured technologies back in 2013 for their target OSes. Instead they chose another young ones. Oh well, not going to cry due to Dropbox making Kotlin and Swift better due to still lack of libraries there for their purposes. The more the merrier, that's my motto. My absolute fear is the manager's dream to only UML be as programming language and they drag'n'drop with no coding behind (the wet dream back in 2005 my manager at Siemens had, and he told me the world is going that way. Ya sure buddy, not if we let it happen and so far instead of unification we have even more diversification)
Does anyone have experience trying to do this kind of thing with LibGDX?
LibGDX is a layer on top of OpenGL (lwjgl actually), and used a lot for game development. It is based on Java, and is able to package the application for Android, iOS, Windows, Linux, Mac, and even web (targeting WebGL). For iOS it uses RoboVM, and for the web export it uses GWT to compile to JavaScript.
Obviously this is best suited for custom UIs (like games), but with some frameworks on top it seems it would be possible to build a decent interface. The framework already has support for most platform specific functionality (like text input). Performance should be great because it runs close to the metal, so to speak.
Just curious if anyone has tried this, as I am considering the approach.
Actually flutter is based on this approach. The native widgets are completely recreated with this dart based runtime. Performance is indeed great, however it seems like a hell lot of work to create all these widgets...
Code reuse on anything that is too complex will usually give you these issues as reuse doesn’t scale unless is very simple and function is super well defined and focused. Programmers will turn it into a monster as it evolves, usually
This reminds me of LLD’s approach of sharing an architecture but having separate ELF, COFF and Mach-O implementations. This made it much faster than gold which wasn’t actually slow.
Apparently sharing code isn’t necessarily a great idea.
I just launched (like, an hour ago: https://news.ycombinator.com/item?id=20700196) a cross-platform app that is almost all shared code. In my case, the solution is a web app with very lightweight native wrappers, and I'm quite happy with it. Obviously that wouldn't be the perfect fit for all apps, but none of Dropbox's "(not so) hidden costs" are relevant in my case, and I suspect many apps would be a good fit for the architecture I went with.
Most cross-platform solutions are great in the beginning and then the problems start to come to surface as teams grow, requirements change, bugs are harder to track down, code needs to be updated, the next hot cross-platform framework comes along with new promises, etc.
Sure but, these issues come to surface much quicker.
Every time I've started a new iOS project in the past ~2 years, I've reached out to a handful of developers and managers on how cross-platform is going for them. None have been completely sold on it. The ones that are, are the ones that have been doing it for a few months. I've come back around to talk to them months later and they aren't so confident in the decision anymore.
Nope, that's not accurate at all. The native apps support push notifications, in-app payments, native UI elements, proper lifecycle management (storing in-progress doodles across the app being killed in the background, or the device being restarted, for example), and more offline support than would be possible with the website. Among other things.
The things that are in the native wrappers are things that would be iOS-specific or Android-specific regardless of zero shared code or maximum shared code. In my case, the bulk of the app is shared -- everything that can be a common code base is a common code base.
According to GitHub's rough metrics, that's:
Go 44.7% Dart 32.8% Java 7.1% JavaScript 6.9% Swift 6.6%
I shudder to think how many people will read this headline and immediately takeaway that React Native/Flutter/Electron are terrible cross platform solutions.
I'm hoping more people will read this headline and at least take away that RN/Flutter/Electron should be approached with caution and care rather than "slap it in an Electron shell and it'll run great everywhere and everyone will love it," which is the implicit attitude that makes so many people, uh, not love it.
React Native, Flutter, and Electron have a lot of problems in between them. They are fairly good cross platform solutions, whether cross platform UI is actually a good idea is much more dubious, I lean towards no it's not.
I'm more excited about leveraging native platform UI with shared code in a common language like Rust. C++ as described in the blog is problematic for this.
Current state of sharing code between platforms is quite terrible. Most languages support only C FFI, so if you write shared code in C++, will require bridge from C++ to C and then bridges from C to native platforms.
For CRUD core, this kind of overhead makes no sense.
I think this situation could be improved with automated generation of bridges. Apple implemented this pretty well for their own purposes, you can call Swift from ObjC and ObjC from Swift quite easily.
When you're at Dropbox's scale it's easy to just hire a load of iOS and Android devs, and some product managers to keep them all in sync.
But when you're not at that scale - you can't afford to build the same app twice - you are often left with a choice of either Xamarin or React Native. Both of these are perfectly good frameworks on which to build software; don't let the FUD win.
As someone who has built a C++ library to share across iOS and Android I completely agree. The biggest issue for us was finding developers willing to learn and adopt C++.
For our use case we weren’t writing UI or user flow logic, it was an ‘engine’ that handled complex business logic. It used JSON as input and output.
If I were to do that same project again I would use mostly native code and write our engine in Javascript.
Since it's not stated in the article, I am going to assume they haven't tried Xamarin.
I've been really pleasantly surprised lately how much code can be effectively shared when using Xamarin Forms. I haven't encountered much problems, except when dealing with tricky stuff like the Android camera API's (which, from what I understand, is problematic in native Android code as well).
How about running a background service to synchronize a local folder of content with a remote server? Making sure the background service is launched correctly upon startup? Integrating with the OS's native sharing features with other apps? And so on.
In my experience, using a "cross-platform" framework means that you'll write the code not once, but three times: once for the cross-platform framework, and once per platform (Android / iOS) given all the quirks and unavoidable customizations that will need to be made to make your app function correctly.
For platform specific stuff it's easy to write a wrapper. Most of the time Xamarin is very closely up to date with Apple's SDK, due to some automated tooling that generates the C# classes, interfaces and such. So you should be able to either use the C# wrappers for platform specific functionality, or alternatively can quite easily create your own binding using the Sharpie tool [0].
The last 2 apps I've been working on were Xamarin apps and I'd guess the first app probably shares 80% - 85% of the code base and the second app 95%. The first app less, because the client wanted a specialised camera module, which required me to create bindings for the Fotoapparat [1] library on the Android platform.
Both apps used Xamarin Forms. The platform specific code, all things considered, is really small.
I should note I've also done a lot of Swift & Objective-C development in the past and a little bit of Java. I also tried Xamarin around 2013 and back then the experience wasn't as good as it's right now.
My major gripes with Xamarin currently are:
- The IDE becomes slow sometimes (fixed with a restart)
- The IDE sometimes has issues connecting to the Android emulator for debugging (often restarting the computer fixes this).
I should also note that if I were to write an iOS only app, I'd definitely use Swift (I love working in Swift, more than C#, Objective-C, Java, …). But for multi-platform apps, currently my tool of choice is Xamarin.
I wonder how well Xamarin will work with Swift-only frameworks, like the upcoming SwiftUI and Combine frameworks. They make heavy use of Swift generics. It’s been a long time since I’ve touched Xamarin so it’s a genuine curiosity.
I did have to make some adjustments to the library, so I guess it could be tricky to use Sharpie Bind with Swift frameworks, if you're not in control of the Framework yourself.
There's quite a few places that are still dealing with 4 codebases with varying degrees of shared code. Desktop, mobile web, native Android, native iOS.
A shame, since many of them don't really need more than one responsive web codebase. Native apps do add value for many use cases, but not all.
They add value in all cases by not needing 200mb memory to run what would be a 20mb app for native. It's crazy how my phone runs perfectly fine on 2gb ram but my laptop is swapping with 16gb just because slack, vscode and whatever else election app is running.
What's fine with wasting effort on pointless duplication? You made this false claim many times in the past, but it's as false as it was before. And your support of lock-in practices is strange to begin with.
You repeating the same thing won't change the fact that lock-in is a foul practice that simply taxes developers for the benefit of those who want to hamper competition. Only shills support such approaches. You perfectly well know it yourself.
You know what lock-in is, so no point in pretending you don't understand what its problems are. It's not about productivity, it's about anti-competitive jerks who use development tools to hamper competition.
Of course those who push lock-in like to present it as a positive thing and hide its real nature, but it's just a smokescreen which is easy to see through. Once in a while, they just can't help it, and express their real views: https://en.wikipedia.org/wiki/Criticism_of_Microsoft#Vendor_...
Anyway, I don't think explaining this to you in particular is useful. You know it all well, but pretend it's a good thing. Which is very shill-like.
Unlike you, even developers of such lock-in stuff admit, it's a bad thing for the industry. Those who are interested, can find comments by Dan Baker (from Oxide Games) for example, who is a former developer of DirectX.
Plenty of nice words for someone that advocates APIs that are portable only on paper, as the best features are all hidden away on OEM specific extensions, leading to multiple incompatible code paths.
Psst, lets not reveal the Achilles heel from the soap box.
Whitewashing lock-in is pretty much a shill thing. And pushing it on others is a jerk one.
As for Dan Baker, he meant low level APIs, not DX12 in particular. He explicitly said, that MS only API is harmful for developers, and non lock-in one is needed. Not surprisingly, Ashes of the Singularity is in the process of getting rid of DX12 for Vulkan. Have fun trying to dismiss it.
Over time as well, you cannot keep up with features. You will always be lowest common denominator, because if a system handles 6 platforms, and you want the feature of 1 platform on that platform, it will triple or more your amount of effort to implement it.
Find and hire candidates with this very specific skillset (we tried to hire for this role for over a year with no success)
Really? You weren't able to fill this role? I can almost guarantee that you could throw an extra 25-50k at someone competent to fill it.
In the end we no longer share mobile code via C++ (or any other non-standard way) and instead write code in the platform native languages.
I've started to comment on this across the interwebs. If you are a cross-platform (native) mobile developer, and haven't taken a look at Xamarin, you _need_ to give it some attention. Xamarin.Native specifically is the best solution I've come across that mostly gets out of your way, and has robust library ecosystems and sane interop.
Would like to see an analysis of the costs and benefits of two separate codebases after they are done with that and have given it time to prove itself... especially with the more complex code.
Write once run anywhere seems to be the eternally elusive holy grail of cross platform software development. It's a shame AppForge died right before mobile took off (tongue in cheek).
Nah, this is bunk. The cost of code sharing on top of something like Ionic is next to nothing. And for every platform specific feature you can just plumb in whatever plugins you want.
Just look at the code needed to support the file system on iOS 13 to write a file provider. It changed a lot, and non-native code to adopt to that would be a nightmare I imagine.
You are right. I've done a lot of development in AIR myself and it is super impressive. Easy TypeScript like language. The Flex SDK is mature with all the standard stylable controls you need. Dedicated IDE that just works great. Easy deployment. I was able to build a cross platform (iOS/Android/Web/Desktop) video chat application with it and it looked and worked great everywhere. Someone dropped the ball somewhere with AIR.. I'm amazed they're still updating it.. that's something at least.
Unless you need something very specific for the display UI. Try cordova (open source version of PhoneGap), and stuff like ionic, framework7.io, mithril.js will mostly get your most apps done needed while still sharing the code via JavaScript, HTML5/CSS. for anything native you can always write your own cordova-plugin if it doesn't already exist.
I've been considering building a pwa and using a webview as a pseudo native app I know iOS support isn't great but it seems to be improving and iOS push notifications could be built separately in the meantime. How's anyone tried this approach?
While PWA’s largely work, they provide a poor installation experience and little device integration. Most people use Apache Cordova, PhoneGap, or Ionic as wrappers to their mobile web app, providing expected installation experiences through app stores and common device integrations.
Are you really asking if someone has tried putting their website inside a webview and call it an “app”?
Have you heard of Electron, Cordova and tons of other crapware that help produce so much garbage “apps” every year, it would put to shame even the most toxic enterprises in an unregulated Republican dream?
I'm currently working on a project where we have indeed decided we needed to go Swift/Kotlin native on each platform, but with the long-term goal of building a cross-platform app with Qt.
Qt certainly has its issues, but I respect what its developers have accomplished. And it appears to offer adequate ways to do common things like parse JSON, and it abstracts things like Bluetooth usage and file storage in the apps' sandboxes; so I don't see some of the hurdles cited in this article.
I initially had a similar thought, but I don't think that applies here.
With Electron, you don't write a cross-platform base and then write all the platform-specific goo on top of it. You just write your app using web technologies, press the build button, and ship it on all the platforms you want to support.
Since web technologies are popular, you don't end up with the same issues hiring and training developers like you do when you need to teach a new hire about your bespoke cross-platform mobile solution.
With Electron, essentially your users pay for all of this by being forced to adopt a resource-hungry platform, while the development team gets a mostly-free pass.
This is wild. So Unreal/Unity and others somehow get to run their arguably vastly more complex engines on mobile desktop and consoles, but sharing some logic between iOS and Android is too complicated for Dropbox. What are they smoking?
I see you haven't worked on that level (or maybe you are young)....
First, unreal/unity work at a different level of the hardware, and they have to deal with different graphic drivers, in the same type of the OS.
But keep in mind that they are not the end product! The end product is the game that run on them.
So, the cost to building a multiplatform game, is having a whole Game Framework (which are whole companies), to deal with the lower level details and help abstracting out the differences of the platforms.
Dropbox is the end product. So, to make the code shareable, they have to build something like the equivalent of a game engine....
which is not a simple task....
I have worked at Spotify, and we had a lower level library to help (built in C++) to help out with some of the common features across platforms. It was ported from the desktop app, to iOS and Android, and it was a separate project. Over time, as features kept piling in, it turned into a giant un-manageable mess. It got so bad at some point, that we had to skip a couple of scheduled releases to clean up/fix some of its issues.
The problem became even worse when Spotify tried to build a unified UI on top of it (HTML 5 based), with a bridge layer to the C++. The idea was to write once, and (haha) run it everywhere. It failed, for the reasons many of them failed in other companies as well (facebook was doing HTML5 at the time as well).
That got removed, but the problem still remains that "Core" got overly complex as many platforms had their own features, which often were "one off", A/B testing, to see if it works or not. And separate teams (iOS, Android, and Desktop) had their own schedule on this. (It makes perfectly sense to A/B test one feature in Android, and another one in iOS, to see if people even like that feature before making is x-ross platform.
Common platform code, massively slowed down A/B testing of new experimental features, and thus it was a huge cost to the company.
The only solution to this was to keep the common "Core" (as it was called), to the most minimal features needed, and mostly related to playback (ie. song retrieval, caching, retrieval, decryption, playback, and some account features).
Do, the A/B testing on the platform level (iOS, or Android), of it was successful then roll what made sense to the rest (but still trying to keep core as lean as possible).
But since I left few years ago, things might have changed again....
I was on that core team and that's not quite what I recall regarding the core library.
We had some hard requirements on that library because of protocols used by the backend and other reasons. And when you mention features "mostly related to playback", well, it was 95% of the library from its conception (the rest being login and communication with backend).
Most of it is quite complex code and not anything you want to rewrite for each platform (mainly offline storage, remote control, playback).
The issues we had were caused by growing pain from having the library initially mixed in the desktop code (back when mobile wasn't a thing). It needed to be extracted, productified and a whole lot of tooling and testing written for it. Once we got there, the quality of the code was much nicer and the integration in the various clients easy and automated (which it couldn't be at first).
It certainly took sometime, but it wasn't a problem caused by it being cross platform, only that we needed a better infra. We were there when I left the company about the same time as you for quite a while already.
The integration with various clients got later easier once we removed the need to reach into that library for just fetching simple server side data. But that wasn't a cross-platform library issue, it was mainly a backend issue.
I appreciate the lengthy answer, but it builds on the premise that I can't possibly know what I'm talking about because otherwise I would've come to the conclusion that code sharing just sucks. My point with engines still stands: they share huge chunks of codes. Even code that has nothing to do with low level abstraction (animation, loading models, mixing audio etc.). And yes, their job is to run on all their target platforms. But if your app targets multiple platforms that's your job as well.
What you described with spotify is that you were essentially building your own phonegap. That might be the right answer if you want to achieve as much code sharing as possible. But just because you can doesn't mean you should. You can't go ahead and make wrong abstractions and assumptions about platforms and then tell me I shouldn't even be trying because you failed at it, when others somehow achieved it. And I know several non-SV companies (a Navigation vendor I worked for for example) that ported their C++ code base to mobile in a matter of days. So it might as well be possible that the engineers in charge at Dropbox, FB etc. were lacking experience.
This got me thinking: may be the right (low resource-high performance) way to do multi-platform software nowadays is to use a game engine instead of Electron.
Games are wholly enclosed binaries that render everything in open gl or whatever, their entire ui is written in the 3d space, so there's less layers I'd imagine.
The Unreal Engine is a complex beast with anything but few layers. Yes the games are just packages, but the engine itself is an multiplatform engineering feat.
I think it's exactly because they are so complex that those engines can afford to share a lot of code between platforms. For smaller and/or simpler applications, it's probably not worth it to deal with the cross-platform complexities.
Here's a fine alternative to code-sharing: well first and foremost you need to have a good requirements spec. After that, implement it on one platform first in a platform agnostic way. So heavy use of delegates/interfaces/injections to handle platform-specific functionality, even for one-liners like getting current system time. So the code should be pure, boring, plain old Java, Kotlin, Swift, whatever. Minimal use of fancy language features that may not port cleanly. After one platform has the module in place and well-tested, do a copy/paste port to the other platform. You will find that the copy/paste port will actually uncover bugs, optimizations, edge cases, etc., that can then be ported back to the original platform. Things you would not have found if you were only writing the code once. It's like the most intense kind of PR code review you can get.
So I've found that copy/paste-port-based code sharing between platforms is not just a lesser evil, but actually a strategy that has its own unique benefits. If you get into a groove you can build up a set of regexes to give you a sloppy transpiler that does most of the tedious porting for you.