Barely does 1 and 2 since 3 causes browsers to actually slow down to a crawl. The second that 18k list started to load, my browser turned ridiculously slow.
yeah the website presents it in bloated HTML unfortunately.
Here is the content:
-------------------------------
How on Earth the Facebook iOS Application is so large
Recently someone on reddit asked “How on earth is the Facebook app size so large ?”. The person asking the question realized that the ~100Mb compressed App Store archive wasn’t all assets - a very large portion was the application binary.
How do you answer that question? You quickly reverse engineer the application!
1 Download the binary to a jailbroken device from the App Store
2 Find the application on the device.
SSH into the device and find the application container on the filesystem:
find / -type d -iname "SomeApp*.app"
3 Decrypt the binary
Install dumpdecrypted on the device. Follow the instructions for using it.
4 Get the files off the device using scp or rsync.
In most cases you will want the application container, the decrypted binary, and the application data container (i.e. what it writes out to the filesystem).
At this point you have the decrypted binary and it’s files. You can now run class-dump on the binary and output a directory of header files for all the classes in the binary, or you can use a disassembler like IDAPro or Hopper to easily see all the classes, symbols, etc.
In the case of the Facebook application, there are more than 18,000 classses in the application:
> FB App is 114MB in size, but loading this page in Chrome will use a good 450MB, idk how they managed that.
Comparing the size of the binary with the size of the memory used by the live program is apples to oranges. The two metrics don't have any correlation whatsoever.
He doesn't compare the size of the binary with the size "of the memory used by the live program".
He compared the size of the binary with the size (in Chrome's memory) of the idiot's webpage complaining about it.
Which he obviously knows is Apples to Oranges, but it's still fun to know and illuminating (the guy blames FB for being too careless with their app size but is too stupid to watch for his web post's size).
I love how easily some ppl on HN call someone an idiot. Way to encourage people to share.
I found it interesting even if my browser ate 260 whatever MB of memory. Doesn't make him an idiot, just a little bit lazy about optimizing his blog (which is not used by millions like the Facebook app)
It seems none of the cool kids with lean blogs managed to post this stuff or get it on HN. I'd love to read their articles about it, but hey, they don't seem to exist.
>I love how easily some ppl on HN call someone an idiot. Way to encourage people to share.
We live in the "social sharing age". People even share what they had for lunch. If anything, people should be encouraged to share LESS or more meaninful stuff only.
>It seems none of the cool kids with lean blogs managed to post this stuff or get it on HN. I'd love to read their articles about it, but hey, they don't seem to exist.
Serving something you want to read doesn't mean there should be no criticism for serving it in a bad way.
If you had only one chinese restaurant in your city, one that served bad food, would it make sense not to complain for their good quality because "other restaurants don't serve noodles at all"?
> Serving something you want to read doesn't mean there should be no criticism for serving it in a bad way.
There's criticism of serving something in a bad way, and then there is calling someone an "idiot".
Should we instantly call every single person an idiot for making a mistake?
I wouldn't say that the food was bad here, it was just a bit cold.
> We live in the "social sharing age". People even share what they had for lunch. If anything, people should be encouraged to share LESS or more meaninful stuff only.
This is off topic, but I don't think people should share less. If you are bothered by it, you should read less. Maybe we should have better tools for filtering.
But information is everything, something as banal as "I had a bad day" has value. Tracking social sentiment for example.
>There's criticism of serving something in a bad way, and then there is calling someone an "idiot".
True.
>Should we instantly call every single person an idiot for making a mistake?
No, but what if they do that first by pointing fingers at the stupid FB devs?
>This is off topic, but I don't think people should share less. If you are bothered by it, you should read less.
This still leaves the problem of an overproduction of pointless stuff, which aside from the ecological issues and time wasted by those producing it, also makes deciding "what to read" among the avalance of BS more difficult.
I was expecting a bit more nuance and insight than '18,000 classes, lol, FB is dumb.' At the very least, it'd be interesting to note what the size of the app might be after app thinning goes live next month.
In brief, a new feature of iOS 9 that allows app resources to be delivered on a device specific basis..meaning that, for example, a user with an iPhone 5 will no longer need to download assets meant to support the 6+ or iPad. More here: http://arstechnica.com/apple/2015/06/app-thinning-will-be-a-...
This so called "investigation" of the Facebook app can be done much simpler. Locate the app in iTunes, rename the .ipa file to .zip, unzip and run otool and nm on the binary. The author list 18,000 header files which is not much of an investigation nor much interesting. Though the staggering number of classes does admittingly somewhat explain the enormous size of the FB app. I do not use the FB app, but generally it can be equally illuminating to look in the Resources/ folder of an app if you wonder where the size came from.
>This is the same information provided by using ‘otool -ov’, but presented as normal Objective-C declarations, so it is much more compact and readable.
No. You have to jailbreak and "crack" the app to get at the code segment, which makes this more of an effort than you make it out to be. Unless you know something I don't?
Seems to me that huge part of this was auto-generated (and auto-structured), and each .m/.h contains very few lines of code. I don't know what FB app does, but with 18000+ fully functional non-excessive handwritten modules you can just win the internets. Oh wait~
They are obviously very lost and reinventing wheels, rather than learning and applying the right design patterns for Apple platforms. I talked to a Facebook iOS person on here recently who was under the impression that GCD offered no help with concurrency problems.
I opened it in the opera version of chrome and it loads no problem. The browser isn't the best at everything but sometimes it works when everything else fails.
Lots of money, lots of employed engineers, everyone needs to do something.
That's the case with a lot of tech today - it's done for the sake of letting people do something. Which, I guess, is not that bad, in the end things are brought to perfection, even though they're not strictly 'needed' in the product.
The fast CPUs and enough RAM take care of the added complexity and everyone's happy.
I hope that's not the case, considering there are much more interesting things Facebook employees could be working on (not necessarily even for the betterment of society).
I heard about something similar to this once on a project a friend was working on (it wasn't of this scale) - it was actually when a designer was the lead on the project and he insisted on customizing random tiny bullshit and the engineers didn't have the power to say no (to cut on the added complexity).
Reading through just some of the classes they have it looks like they have everything from packet readers to image manipulation to RSS feeds to web page embedding to video playback in there. Its literally an OS unto itself.
If you're having difficulty loading this webpage, it's because the author decrypted Facebook's iOS binary and listed all 18,000 header files.
Yes, this webpage is merely a list of 18,000 headers.
1. The article gives the insight that the binary is sooo big, not just assets as one would think.
2. It shows how the classes are extracted
3. It presents the classes for you to browse.
The word "merely" is downplaying the information being shared by the author.
Barely does 1 and 2 since 3 causes browsers to actually slow down to a crawl. The second that 18k list started to load, my browser turned ridiculously slow.
I agree that the inline HTML code block is not optimized for such a use-case. You can read the content here: https://news.ycombinator.com/item?id=10066530
Still my point was about the content being simply a list, which it is not (for better or worse in this case, I know :-)
> causes browsers to actually slow down to a crawl.
some browsers. FF 40.0 on my 2012 rMBP barely breaks a sweat.
I opened it with FF on my phone - no problem loading or scrolling it whatsoever.
Fine on Chrome 44 running on a fairly ancient Core 2 Duo
Latest Firefox doesn't even break a sweat.
TL;DR Facebook app has 18k classes, if you want to know what they are look here: https://gist.github.com/quellish/473f513fbd1310233a8e
FB App is 114MB in size, but loading this page in Chrome will use a good 450MB, idk how they managed that.
yeah the website presents it in bloated HTML unfortunately. Here is the content:
-------------------------------
How on Earth the Facebook iOS Application is so large
Recently someone on reddit asked “How on earth is the Facebook app size so large ?”. The person asking the question realized that the ~100Mb compressed App Store archive wasn’t all assets - a very large portion was the application binary.
How do you answer that question? You quickly reverse engineer the application!
1 Download the binary to a jailbroken device from the App Store
2 Find the application on the device. SSH into the device and find the application container on the filesystem:
find / -type d -iname "SomeApp*.app"
3 Decrypt the binary Install dumpdecrypted on the device. Follow the instructions for using it.
4 Get the files off the device using scp or rsync. In most cases you will want the application container, the decrypted binary, and the application data container (i.e. what it writes out to the filesystem).
At this point you have the decrypted binary and it’s files. You can now run class-dump on the binary and output a directory of header files for all the classes in the binary, or you can use a disassembler like IDAPro or Hopper to easily see all the classes, symbols, etc.
In the case of the Facebook application, there are more than 18,000 classses in the application:
[... list of all the classes ...] https://gist.github.com/quellish/473f513fbd1310233a8e
There is a LOT of crap in there. Even a “FBFeedAwesomeizer” - which alone is a collection of 74 classes and protocols.
This is why the application binary itself is over 114Mb.
> FB App is 114MB in size, but loading this page in Chrome will use a good 450MB, idk how they managed that.
Comparing the size of the binary with the size of the memory used by the live program is apples to oranges. The two metrics don't have any correlation whatsoever.
He doesn't compare the size of the binary with the size "of the memory used by the live program".
He compared the size of the binary with the size (in Chrome's memory) of the idiot's webpage complaining about it.
Which he obviously knows is Apples to Oranges, but it's still fun to know and illuminating (the guy blames FB for being too careless with their app size but is too stupid to watch for his web post's size).
I love how easily some ppl on HN call someone an idiot. Way to encourage people to share.
I found it interesting even if my browser ate 260 whatever MB of memory. Doesn't make him an idiot, just a little bit lazy about optimizing his blog (which is not used by millions like the Facebook app)
It seems none of the cool kids with lean blogs managed to post this stuff or get it on HN. I'd love to read their articles about it, but hey, they don't seem to exist.
>I love how easily some ppl on HN call someone an idiot. Way to encourage people to share.
We live in the "social sharing age". People even share what they had for lunch. If anything, people should be encouraged to share LESS or more meaninful stuff only.
>It seems none of the cool kids with lean blogs managed to post this stuff or get it on HN. I'd love to read their articles about it, but hey, they don't seem to exist.
Serving something you want to read doesn't mean there should be no criticism for serving it in a bad way.
If you had only one chinese restaurant in your city, one that served bad food, would it make sense not to complain for their good quality because "other restaurants don't serve noodles at all"?
> Serving something you want to read doesn't mean there should be no criticism for serving it in a bad way.
There's criticism of serving something in a bad way, and then there is calling someone an "idiot".
Should we instantly call every single person an idiot for making a mistake?
I wouldn't say that the food was bad here, it was just a bit cold.
> We live in the "social sharing age". People even share what they had for lunch. If anything, people should be encouraged to share LESS or more meaninful stuff only.
This is off topic, but I don't think people should share less. If you are bothered by it, you should read less. Maybe we should have better tools for filtering. But information is everything, something as banal as "I had a bad day" has value. Tracking social sentiment for example.
>There's criticism of serving something in a bad way, and then there is calling someone an "idiot".
True.
>Should we instantly call every single person an idiot for making a mistake?
No, but what if they do that first by pointing fingers at the stupid FB devs?
>This is off topic, but I don't think people should share less. If you are bothered by it, you should read less.
This still leaves the problem of an overproduction of pointless stuff, which aside from the ecological issues and time wasted by those producing it, also makes deciding "what to read" among the avalance of BS more difficult.
I was expecting a bit more nuance and insight than '18,000 classes, lol, FB is dumb.' At the very least, it'd be interesting to note what the size of the app might be after app thinning goes live next month.
Details on app thinning?
In brief, a new feature of iOS 9 that allows app resources to be delivered on a device specific basis..meaning that, for example, a user with an iPhone 5 will no longer need to download assets meant to support the 6+ or iPad. More here: http://arstechnica.com/apple/2015/06/app-thinning-will-be-a-...
Not just assets, but also the compiled binary will be for that architecture alone. ARM6, ARM7 etc. Really cool stuff.
Not at all before iOS 9.0 general release.
This so called "investigation" of the Facebook app can be done much simpler. Locate the app in iTunes, rename the .ipa file to .zip, unzip and run otool and nm on the binary. The author list 18,000 header files which is not much of an investigation nor much interesting. Though the staggering number of classes does admittingly somewhat explain the enormous size of the FB app. I do not use the FB app, but generally it can be equally illuminating to look in the Resources/ folder of an app if you wonder where the size came from.
He didn't claim it to be an investigation. It seems to be an insight.
As for being interesting, it's quite subjective. Might not be interesting to you, but it makes other people wonder.
class-dump[0] is handy too.
>This is the same information provided by using ‘otool -ov’, but presented as normal Objective-C declarations, so it is much more compact and readable.
[0] https://github.com/nygard/class-dump
No. You have to jailbreak and "crack" the app to get at the code segment, which makes this more of an effort than you make it out to be. Unless you know something I don't?
The symbols themselves are not encrypted, so you can still list those without cracking the app.
Their binary alone is 118MiB? Even with all the auto-generated crap in there (I hope no human chose those names), how do they manage that?
I wonder how long it takes for an iPhone to even load that thing into memory before it starts executing any code..
> I wonder how long it takes for an iPhone to even load that thing into memory before it starts executing any code..
I takes my iPhone 4s about 5-10 seconds to start it.
Seems to me that huge part of this was auto-generated (and auto-structured), and each .m/.h contains very few lines of code. I don't know what FB app does, but with 18000+ fully functional non-excessive handwritten modules you can just win the internets. Oh wait~
They are obviously very lost and reinventing wheels, rather than learning and applying the right design patterns for Apple platforms. I talked to a Facebook iOS person on here recently who was under the impression that GCD offered no help with concurrency problems.
Chrome asks to kill the page after a bit. Does this content exist in a cleaner state?
here is a summary + a gist: https://news.ycombinator.com/item?id=10066530
I opened it in the opera version of chrome and it loads no problem. The browser isn't the best at everything but sometimes it works when everything else fails.
There is no Opera version of Chrome. Opera uses Blink, that's it.
Apparently Chrome uses blink too. But now I can see how one is not the other even though both use the same.
I hope it's a mechanism to make it possible for new employees to actually work on this app.
Perhabs someone with experience on large complex applications can share some insights?
Otherwise, I pity the FB employees who gets assigned to this project. "Here you go, fix that button, here's 18k files to digest".
Surely they must've decoupled it into oblivion in order to have a class that manages "just that", so
1. new employee can work in an "isolated" context (although he really isn't)
2. can locate where to write and debug in this complex codebase.
But then they can say. In 2.0 we improved performance 50%!! And repeat this process of headline news several times!
unneeded complexity
Both the app and that webpage.
to my dear downvoters, "Fools ignore complexity. Pragmatists suffer it. Some can avoid it. Geniuses remove it." Alan Perlis
Lots of money, lots of employed engineers, everyone needs to do something.
That's the case with a lot of tech today - it's done for the sake of letting people do something. Which, I guess, is not that bad, in the end things are brought to perfection, even though they're not strictly 'needed' in the product. The fast CPUs and enough RAM take care of the added complexity and everyone's happy.
I hope that's not the case, considering there are much more interesting things Facebook employees could be working on (not necessarily even for the betterment of society).
It seems to definitely be the case for designers (300+?), I wouldn't be surprised if it's the same for engineers.
10,000+ people to keep the service running?
I heard about something similar to this once on a project a friend was working on (it wasn't of this scale) - it was actually when a designer was the lead on the project and he insisted on customizing random tiny bullshit and the engineers didn't have the power to say no (to cut on the added complexity).
Don't know if this applies in FB case.
I mean, sure. Most apps have that. The question is what internal structures led to it being this way, and whether they're worth the cost.
Of course, we have no idea about any of that stuff so we sit outside and say "lol Facebook are so silly".
And its all Objective C code!
Reading through just some of the classes they have it looks like they have everything from packet readers to image manipulation to RSS feeds to web page embedding to video playback in there. Its literally an OS unto itself.
my favorite ones though:
FBWhichApplicationAreWe_DO_NOT_USE.h
FBProfileSetEventsCalendarSubscriptionStatusMutationOptimisticPayloadFactoryProtocol-Protocol.h
FBStoryCreateInputDataAttachmentsLifeEventExtrasEducationInfoEndDate.h
FBReactionAcornSportsContentSettingsSetShouldNotPushNotificationsMutationCall.h
I've never seen an iOS application quite like this. I wonder how they even manage it. Tons of subprojects perhaps?
I wonder how much of it is automatically generated.
That's a good question. Is much of this the output from some GUI design tool? There seems to be a class for every GUI element.
That's what I call a fat client :)
Makes me remember of that time they had to cheat on Dalvik because they hit the 65k methods limit 8-) https://news.ycombinator.com/item?id=8162342 (blog link is dead)
Anyway, I find this to be pointless, when the majority of that size is probably taken by assets.