NSKeyedArchiver performance and custom class alternative

NSKeyedArchiver is a common choice for saving data of medium complexity, alongside plist serialization. More complex and you might use CoreData, less and you might use NSUserDefaults. I use NSKeyedArchiver in my Voicer app to record the metadata for sound. For the metadata I tried to design the format so that it would be very responsive, by only keeping index markers to audio samples in external files. However, I noticed for long sounds that were over an hour, it could take around 10 seconds to load just the metadata.

I thought about it longer, and decided that the reason for this must be the key-based lookup. So, I went ahead and removed all the keys (this is generally safe to do as long as you encode and decode in the same order). But, performance did not increase. I then tried using NSArchiver, which is listed in the mac os x docs as deprecated, and is not even mentioned in iOS docs. It is much faster. The only reference I found to it was an older StackOverflow post from 2012. However, using it is problematic, because presumably, it counts as private API which could be rejected on app submission.

It's convenient!

It’s convenient!

I put it down for a long time, but recently came back to it – I fixed up and renamed the Cocotron NSArchiver class to NSKeylessArchiver so that it could exist along side the private one. The results are okay, but could be improved.

Writing directly to a file would be faster for my case, but I’m sticking with the initWithCoder method for now, because it is less work, but also for maintenance – the encode/decode functions are nice, and it’s easy to have backwards compatibility with NSKeyedArchiver by checking a file extension to determine the right Unarchiver to use. The NSKeylessArchiver also does some fancy but mostly-cheap stuff, such as string and object references – if you use the same string twice, it will only occur in the archiver once, and used as a reference the second time.

Here is the github repo and the readme:

How to use

Add NSKeylessArchiver and NSKeylessUnarchiver’s .h and .m files to your project. If you have an arc project, you will need to go into the "Build Phases" tab of your project settings and add the ‘-fno-objc-arc’ flag to both NSKeylessArchiver.m and NSKeylessUnarchiver.m.

Then, just #import "NSKeylessArchiver.h" and use it like an NSCoder. See the performance test project below for a sample.

When to use

If you have a moderate to large use case for NSKeyedArchiver that does not rely on keys (e.g. uses only -[NSCoder encodeObject:] and -[NSCoder encodeObjectOfObjcType: at:] as opposed to -[NSCoder encodeObject: forKey:]), then you may be able to use this. The keyed based encoding and decoding is much slower, because it allows for random access. The non-keyed method allows a faster linear, forward-only read and write because it requires the order of encoding and decoding to be fixed. Mac OS X has a deprecated NSArchiver/NSUnarchiver, and iOS has this, but is not a public API.

How do I get backward-compatibility without keys? Forward-compatibility?

A frequent cited advantage of NSKeyedArchiver is that the keys provide easy backward-compatibility. It certainly helps, but of course, backward compatibility can exist without keys.
If you store a version number, you can still add and remove new objects/data, and check the version number on load to determine whether or not to decode or not. If this is not clear I will expand on it later.

Forward-compatibility via keys is much harder, but in the ‘save user data’ case, you may not need it – the only way that you will need to load a file from a future version is for cloud-based methods.
If you do need it, it is possible to implement schema for forward-compatibility with NSKeylessArchiver, by using keys implicitly in an NSDictionary (which you should be able to use with NSKeylessArchiver!). The other options are probably too messy, and likely remnicent of file formats that specify a byte length of each component and subcomponent, in which case you should just write your own binary file.

Performance

First of, if fast-as-possible performance is your goal, you might consider something else, such as directly writing to a binary file. This is a trade off for having the convenience of having a drop-in replacement for compliant uses of NSKeyedArchiver.

For a test case, I compared NSKeyedArchiver, NSKeylessArchiver, and NSArchiver with a simple root object with 20000 ints. Please feel free modify the test repo and this class to improve performance and correctness. This was the result of running a release build on an iPhone 5S, with 10 runs per class:

encoding (min/max/avg secs) decoding (min/max/avg secs)
NSKeyedArchiver 0.2048/0.2453/0.2165 6.8919/6.9238/6.9037
NSKeylessArchiver 0.0407/0.0506/0.0451 0.0253/0.0330/0.0287
NSArchiver 0.0094/0.0114/0.0102 0.0019/0.0025/0.0020

As you can see the performance for this unrealistic use case shows NSKeylessArchiver doing much better than NSKeyedArchiver, but worse than NSArchiver. NSArchiver is a better choice, but apps that use it may not pass iOS app review due to the private API status. If you know otherwise, let me know.

In practical use, I have seen about a 2-5x speed increase for a use case that had ~20,000 ints/floats/objects being encoded/decoded.
It seems logical that the gains will be larger for encoding schemes that have a larger ratio of encoding calls to actual data being encoded.

For reference:

The Magical Bookstore

One of the joys in life is walking into a bookstore, after being away from bookstores for a good amount of time. I can’t quite say why it is so nice, but I have always felt this way. Recently, I came across an interesting book that made me think about this puzzle. The only thing I can remember about the book is its title. I don’t even remember seeing the book – perhaps I saw it as an online recommendation or reference at the bottom of an article. But it has stuck with me, in a nagging and profound sort of subtle that has me thinking there really is something magic there.

Cow Books in Naka-Meguro, Tokyo

Cow Books in Naka-Meguro, Tokyo


The title of the book is 「なぜ本屋さんに行くとアイデアが生まれるなのか?」(Why is it when you go to a bookstore, ideas are born?). I don’t plan on reading it, at least not for a while. I sort of like the mystic that it provides the bookstore, and the unanswered prompt it gives me for understanding the true values of my own consumerism. That is to say, while there is certainly some connection between books and new ideas, the author clearly asks about bookstores and not other places where you can get free books, such as a friend’s house or a public library.

This makes me recall my first two-year stay in Tokyo, just after I had graduated from college. I integrated where I wanted to, but at the end of the day the nuanced stresses of culture shock led me to foster what I believe is a queer consumerism designed to help keep me grounded in a somewhat foreign land. I would go to Jimbocho, which is the famous book’s quarter and visit many shops on a regular basis. I would search for first editions of classic literature, such as Joyce or Hemmingway. I was a student working a part-time job to fund my entire stay, so I would usually just look at the dusty old tomes through glass and protective plastic, but sometimes I would buy one at a premium rate, this being Japan. Somehow it meant something to me that the books were priced so highly. I think I felt like the money was a token of respect that I could offer to dead authors – that the price valued the book more than normal.

Christopher Alexander’s best known book is still a rare find at bookstores (amazon link)

If I were to visit Japan and go back to the same bookstores today, I would be able to buy many more books than I could have before. This buying power somehow reduces the experience, and my 25-year old self wouldn’t have liked that. Although I don’t have the desire to buy many first editions nowadays, I can still understand what I was doing then. Price labels and and implicit value (positioning and presentation) are such an important part of walking into a bookstore. This is just one part of why bookstores are special places.

Last week, I walked into Pegasus Books on Solano Ave in Albany (California). I was elated to find the architect Christopher Alexander’s The Nature of Order and The Process of Life. His books are always very expensive, even used, perhaps because they are rarer. It is sort of my automatic action to search for Alexander’s books every time I enter a used bookstore. I plan to write more about his Pattern Language, and A Synthesis of Form later. But for now, I will just mention with amazon links that these books were very influential architecture books from the 70s that mesh well into programming practice, and have played a role in influencing the Gang of Four’s Design Patterns.