Efficient String Manipulation With Objective-C

One of the best interview questions that I have ever been asked, and now ask myself to potential candidates, is create a function that will transform any set of words into an acronym. The reason why I like this question is that it shows how well the candidate understands basic string manipulation, problem solving and how efficient their technique is for the string manipulation. Regardless of the language that is chosen the two techniques use by most are to either use some sort of regular expression or split.

Regex techniques with PHP, Javascript, Python and Ruby are the most popular (I did it using PHP), but I was curious as what an implementation would look like using Objective-C. It was my theory that using regular expressions would incur more overhead and than using a “string split” technique. Apple does a fantastic job with making their framework api’s pretty darn efficient, but I had a feeling that with this type of simple manipulation using a regular expression would be like lighting a grill with an atom bomb. The results weren’t so dramatic, but none the less it does turn out that using componentsSeparatedByString is a better use of resources....for a single manipulation. When iterating and manipulating multiple strings I prefer to use regex.

When I iterated over the string 1000x the time to execute was almost the exact same.

I even tested two different regular expressions to make sure I wasn’t being to liberal with my pattern. The difference was negligible.

Test String: "hEll0 WoRLD please hElP me"
Standard Regular Expression: "(\\w)+\\s?"
Boundary Regular Expression: "(\\b\\w)"

Single (regular regex)
Time to parse using regex: 0.003947
Time to parse using componentsSeparatedByString: 0.001246

Single (boundary regex)
Time to parse using regex: 0.003747
Time to parse using componentsSeparatedByString: 0.000668

1000x (Regular)
Time to parse using regex: 0.511170
Time to parse using componentsSeparatedByString: 0.538784

1000x (Boundary)
Time to parse using regex: 0.555807
Time to parse using componentsSeparatedByString: 0.485708

Here is the test project

Export SimpleGeo Layers with Python

Urban Airship has decided to shut down the SimpleGeo services that it acquired by March 31st, 2012. I personally see this as step-backwards for the services that they offer and to the SimpleGeo customer’s who rely on the data.

* stiQRd (I am lead developer) uses it for all of our location information.

What was most shocking to me after their announcement wasn’t the cut-off date or the fact that Urban Airship had decided to sunset the service but the fact that they “punted” the solutions for helping existing customer’s migrate their data. Granted they did offer semi-replacements for Storage, but it basically amounts to you rolling your own solution. That is why we were using SimpleGeo in the first place...because we didn’t want to roll our own because of limited time and resources.

SimpleGeo was so easy to use and the company had great support. It was a no brainer.

In order to prepare and update the stiQRd app before the sunset date I need to get an export of my data. Unfortunately, there isn’t an easy way to export layer data programmaticly or through the console. I posted a question to the Google Group and got some half-ass response that didn’t help whatsoever. Not to be derailed I was hacking together a few solutions, none of which I really liked. I then came across this gem (I know...it is bad) of a Ruby script that will iterate through all of your layers and export that data to csv. Score!

The tips to retrieving all your data from a particular layer are two fold. SimpleGeo’s API doesn’t have a “get_all_records(layer_name)” type of method, but they do have “get_nearyby”, which at face value doesn’t look like it will give you the necessary results, but it does when you use (0,0) as the lat/lon and add in the “bbox” (boundary box) parameter values of ‘-90,-180,90,180’.

Unfortunately, I don’t know Ruby very well and I am spending a lot of my free time sharpening my Python skills. Since there wasn’t an implementation for export in Python I decided to port the Ruby script. With the help of Bob “I am a PEP freak” Waycott, we have a pretty elegant solution.

I am also happy to repost that Parse has taken it a few steps further and offered up the simplest/elegant solution for migrating this data over to their backend service. If you are looking for a turnkey solution then head over their for their migration tool.

If ANYONE has suggestions or updates please let me know.

Python Migration Script: (written by Cory D. Wiles and Bob Waycott)

From Months to Minutes - Mobile Ready Backends

98% of my time is spent developing iOS apps and most of them talk so some sort of web-service. Most of the web frameworks that exist today written in PHP, Python and Ruby abstract a lot of “boiler-plate” code the is tedious and error prone. They saves us valuable time and headache. However, many mobile developers still had to setup a server, database, security and create the endpoints for their app to communicate with.

This is not fun to say the least.

Below is a short, but excellent list of available mobile back-end solutions that allow users to setup, in most cases 5 minutes, a mobile back-end with out-of-the-box solutions for model relationships, PUSH notifications, user management, ACL and geo-location. I have used all of them at some point or the other and have saved me HOURS and HOURS of setup, server side coding and maintenance. In addition, any questions, comments or bugs that I sent to their support were answered promptly and with an adequate fix or recommendation.

Unless there is a pressing need for you to roll your own back-end solution I highly recommending using any of these services.

Parse
Parse is the new kid on the block and though it doesn’t have as many features compared to StackMob, what it does excel in its simplicity, outstanding support and turn key solutions for common tasks such as table lists views. There are still some features that needed to be added, such as cache invalidation, model archiving and better console management, but I have spoken to their support staff and the was told that these will be added in soon.

If you have a straightfoward model structure and need a backend setup ASAP then Parse is by far the best solution. Though they have some growing to do, but they are well are there way and their service has been a pleasure to use.

Features:

User managment
Relational Queries
iOS/Android SDK’s
Facebook Integration Wrapper
Push Notifications
GeoLocation Services

StackMob
I wish that StackMob had been around a few years for some of the larger projects that I worked on because it would have been the back-end solution I would have used...hands down. Besides all that standard features that most back-end service platforms offer they also provide tremendous additions such as API versioning, phone support, API extensions and dev/production environments for your API just to name a few.

StackMob’s solutions are where you need to me be for larger API needs. The setup is easy and the service is feature rich.

Features:

API Extensions
API Versioning
Push Notifications
Social Networking Integration
Analytics
S3 Support

Summary

No matter if your backend requirements are larger or small there is no need for most cases anymore to roll your own. Services such as Parse and StackMob have really positioned themselves as the frontrunners in the PaaS and SaaS solution providers. The advantage that they have is that they came in and built their companies from the ground up tailored for mobile, but flexible enough to server any data client.

Fast String Concatenation

Besides collection/list iterations, string concatenation is probably the most common developer task. Doesn’t matter the language. Unfortunately, this is done incorrectly, especially with large strings. I myself am guilty of taking for granted what goes on under the hood when performing this operation. After reading this article I found myself wondering about the mutability of strings in Ruby. Strings are natively mutable so you don’t have as much performance hit when doing basic concatenation, unlike Python which treats strings as immutable, thus “+=” performs a copy-on-write. No efficient at all. For small string concats doing the standard: “+=” or “+” is still the shortest distance between two points, however manipulating a string of any significant size, what seems to be the most efficient across Python and Ruby is to add each separate string into an array and then join.

Examples:

Ruby: ** Thanks to Travis Dunn for codev

Python

Source: http://www.skymind.com/~ocrow/python_string/

Do to the fact that I am doing more and more large scale projects performance is more important than ever and paying close attention to these little details allow for performance boots very quickly.

UITabBarController Subview Gotcha

By far the most view blog article that I have, and subsequent Github project, is my Universal iOS App Template. Unfortunately, it has been a tad bit neglected. There were some memory leaks, still supported 3.2 and not iOS5 compatible. Now that I am working from home I have dedicated part of my day to personal projects and the first order of business is to get the app template up-to-date. Most of the work was just cleanup and minor tweaks. One of the branches however required a little bit more work. The template that comes with TabBarController support. A few months ago I added in the feature of having a “Tweetie” style indicator slider. In 3.2 - 4.x iOS everything worked fine. In iOS5 the indicator was not initially centered in relation to the first TabBarItem, but the entire TabBar. When you selected another TabBarItem the indicator would move, but not in the correct position either.

The number of subviews are different between the iOS4 and iOS5.

--- iOS5
2011-11-17 09:14:54.290 UniversalExample[8989:f803] view frame: {{0, 0}, {768, 1024}}
2011-11-17 09:14:54.293 UniversalExample[8989:f803] view frame: {{181, 1}, {76, 48}}
2011-11-17 09:14:54.294 UniversalExample[8989:f803] view frame: {{291, 1}, {76, 48}}
2011-11-17 09:14:54.296 UniversalExample[8989:f803] view frame: {{401, 1}, {76, 48}}
2011-11-17 09:14:54.297 UniversalExample[8989:f803] view frame: {{511, 1}, {76, 48}}

--- iOS4
2011-11-17 09:16:06.751 UniversalExample[9033:b303] view frame: {{181, 1}, {76, 48}}
2011-11-17 09:16:06.754 UniversalExample[9033:b303] view frame: {{291, 1}, {76, 48}}
2011-11-17 09:16:06.754 UniversalExample[9033:b303] view frame: {{401, 1}, {76, 48}}
2011-11-17 09:16:06.755 UniversalExample[9033:b303] view frame: {{511, 1}, {76, 48}}

I now confirmed that there indeed was a difference in how the center position was being calculated. In iOS5 the first subview is the entire frame of the view. However, I didn’t know which view(sub or super) that it might be referring to.

The best uiview debugging tool for situations like this is the recursiveDescription method. If you haven’t heard or used it before I suggest/encourage you read Oliver Drobnik’s blog on it. In summary it is a private method to iterate over a views hierarchy.

After calling this method on the subviews in both iOS versions I saw the culprit. The first subview of a UITabBarController is an UITabBarBackgroundView, which obviously is going to extended the entire length of the tab bar.

iOS4

iOS5

Why did Apple add this subview you might ask? They added it because of welcome changes to UIKit. Since iOS 2.0 developers have needed the ability to customize many of the UI elements, but unfortunately there wasn’t a straight forward way to accomplish this. The categories and hacks that people came up with are pretty cool, but not maintainable. With iOS5, and updates to UIKit, you can customize background images, tint color, selected images, etc. to your heart’s desire.

In order to fix the iOS5 bug in my template I just had to add a quick runtime check on the OS version and increment the index by 1.

All is right with the world.

UIWebViews Are Horrible

After spending the last few days trying to improve performance with a local html document I am convinced that UIWebView is a horrible component that should be limited to only showing remote web page documents. What brought upon this rant, and proposed solution for it, had to do with the half a second flicker between clicking on a TabBarItem in a TabBarController and the rendering of the local html file itself. If a user navigates away from the TabBarItem and then back again it will load instantly since the contents are now cached in memory. However, I just couldn’t get past that initial flicker.

There are three different methods of loading an html document into.

- (void)loadRequest:(NSURLRequest *)request
- (void)loadData:(NSData *)data MIMEType:(NSString *)MIMEType textEncodingName:(NSString *)encodingName baseURL:(NSURL *)baseURL
- (void)loadHTMLString:(NSString *)string baseURL:(NSURL *)baseURL

Source: http://cwil.es/uJznzA

The first thing I tried was timing each method to see which was faster.

The difference is negligible. The second route that I attempted was to use Grand Central Dispatch (GCD). Using GCD to manage process intensive logic that would normally block the main thread makes EVERYTHING so much better.

Unfortunately, this is not the case with loading the local html file. In fact, the speeds were essentially the same as using the single threaded paradigm. I am assuming this to be the case because of the slight overhead of managing the threads. At the end of the day I am not gaining any performance.

Due to this frustration I am going to look into using @Cocoanetics NSAttributedString-Additions-for-HTML and save UIWebViews for remote display only.

Test Results (iOS5 Simulator. Ran code 4 separate times in - (void)viewDidLoad)

Main Thread Only

loadRequest

0.024776 seconds
0.025888 seconds
0.023575 seconds
0.026948 seconds

loadHTMLString

0.020844 seconds
0.063485 seconds
0.026386 seconds
0.057899 seconds

loadHTMLData

0.027755 seconds
0.036826 seconds
0.026125 seconds
0.025203 seconds

Grand Central Dispatch

loadRequest

0.029463 seconds
0.025321 seconds
0.026808 seconds
0.033857 seconds

loadHTMLString

0.025549 seconds
0.030053 seconds
0.027117 seconds
0.022959 seconds

loadHTMLData

0.035207 seconds
0.034666 seconds
0.029324 seconds
0.023069 seconds

Dynamic Model Update and Using Google App Engine

A few weeks ago I presented at the Mobile Technology for Teaching & Learning. My talk was how to leverage service-oriented-architecture and objective-c runtime functionality in your enterprise applications. In my demo app I had two json files to use as my datasource so that I wouldn't have to rely on an internet connection.

*I was glad that I did because I had lots of problems with it.

After that talk was over I pushed all the code to github with the goal of modifying the project to use an actual web service to show a real world example of how it works. In order to do that I needed to first create the web service. Up to this point in my development career I have been using PHP for the service layer with CouchDB as the datastore. With this particular project I wanted to take things in a different direction. A few years ago I started tinkering with Python and have really grown to love the language. Unfortunately, I haven't been in a situation to were I could use it in any project beyond a few helper scripts (mainly because of me being a novice and my tight time constraints). I wasn't bound by these restrictions for this project. What I wasn't looking forward to was doing the server side, "infrastructure" setup to get things going. I wanted to focus on the development.

Enter Google App Engine to the rescue! Setting up an account and application with GAE is out of the scope of this post, but Google makes it very easy to do…especially if you are focusing on Python. I knew that scaling wouldn't be an issue nor would the data integrity. Google is somewhat good at both of those.

I was able to hack out a basic web service within a day.

The endpoints are:

- http://kwylezds.appspot.com/events

- http://kwylezds.appspot.com/employees

My biggest roadblock came with handling class instances into JSON. The awesome community of stackoverflow.com had it covered.

The demo app is modified to use these endpoints for data via GCD. I also added a class extension for NSDictionary that simplifies the JSON parsing logic, thus removing a lot of the boilerplate code that I had in there originally.

I will be adding in some more enhancements and abstracting the code over the next few weeks. As always, if you have any comments or suggestions please don't hesitate to contact me.

The Sign Language of the Times

Apple has been leading the way for mobile devices since the first iPhone was announced. Except it! However, this doesn’t mean that other competitors can’t be innovative when in this area. Unfortunately, up to this point Google, Microsoft, Blackberry, etc. haven’t been thinking outside of the box. In essence they are just copying, and doing a poor job at that, with what is on iOS devices.

Over the past few years, and most recently with iOS5, there is a lot of discussion about cloud technologies and having seamless access to one’s content. In addition, there is always talk about form factor, physical hard drive capacity, the number of mega-pixels that this camera has vs that one, and the potential of voice interaction with these devices.

All of these features are great, but aside from maybe voice interaction, they are all nice to haves. They are a convienence. What is missing is the answer to: “Where is the next evolution in device interaction?”. Even voice at this juncture is in it’s infancy. Siri is great and I use it all the time, but there are still too many dependencies that any company has to deal with. The computer processing has to be done in the cloud, all the different languages, accents, dialects, connotations, etc.

If I were Apple, Google, Microsoft, Blackberry, Nokia, etc. I would be focusing on multi-touch gestures. Even after 3 short years engineers have started to rest on their laurels with pushing this technology forward and the possibilities. The first place I would start in my research and work on this is by becoming an expert in sign-language. I would hire a team of the best historians and contemporaries in the world on the subject and have them work side-by-side with my UX and engineering teams.

When you look at sign language as a communication tool it beholds such beauty and elegance. It is the calligraphy of gesture communication. The potential for this, as an example, is develop a new gesture “language”. Much like we have spoken and computer languages, but with consistency. The power of this would be unlimited and easy to use and interpret. With the current usage of multi-gesture it is always a singular expression, but when examining sign language a single gesture has to the ability to express an emotion or entire concept. It is multi-dimensional. Unlike, the computational and processing dependencies listed above that come with using voice interaction, all legacy and future multi-touch hardware have the ability to implement a new era of gestures.

Gestures. They are a sign of the times to come.

The Case of Lost Text Messages and AT&T

Oct 14th, like millions of other people, got my iPhone 4S. Because I develop iOS apps I had been using iOS5 for a few months so to have both the new hardware and software together was nothing less than spectacular. Many unfortunate users that day experience failed activations because of Apple's server overload. Since I didn't get mine until later that night I was able to shut off my iPhone4, activate the new phone and restore from iCloud without incident. My new phone was BLAZING fast. Siri, lived up to the height. Cutting the cord with iTunes has made life so much more enjoyable. Finally, I was able to communicate with the other iDevice people with iMessage in mass. It was a good day.

The next night at dinner a friend commented that she had texted me but I never responded. I didn't think too much of it because these things happen especially on iPhone launch/activation days due to network congestion. Over the next few days I noticed that I wasn't receiving text messages from anyone other than iMessages and people weren't receiving mine. Didn't matter the carrier.

After consulting the Google, I learned that many other people were experiencing similar issues, regardless of carrier or device version (4 and 4s). It turns that with iOS5 there is a cap in the number of text messages that the messages app will support. After many frustrating hours with both Apple and AT&T, their "fix" was to wait for update from Apple, start deleting messages from your phone to free up space or wipe your phone and set it up as a new phone. By doing the last suggestion all your text messages would be gone. I wasn't about to start all over with wiping my phone. I deleted any of the old text messages that I could, but still no luck. I waited a few days to see if either one of the carriers or Apple would break their radio silence on this issue and offer some support. Well….NOTHING. While reading on the AT&T community support forum I notice that a moderator had reminded anyone having issues to check their account to make sure that they hadn't gone over their allotted number of texts.

When I checked my account I found something that will make me LOATH AT&T for quite sometime to come.

During the checkout process when I bought my phone I had to choose my data plan. I still don't know why this was part of the process. Maybe because I am one of the remaining few who have unlimited data and AT&T hates me. Regardless, I had to verify that I wanted to keep this plan. Apparently, AT&T found it fairly funny to remove my text messaging that I originally had. I mean why would someone assume that their current service features would still be there just because they are ordering a new phone.

Needless to say as soon as I added the text messaging back everything started working.

Carriers wonder why consumers and companies like Apple (iMessage), Google (GoogleVoice) and even Blackberry are doing whatever they can to bypass their unreliable networks, constantly changing terms/conditions and overall uncoolness towards consumers (hidden fees, data throttling, "upgrade fees", etc).

If you are having text messages check your account before you waste days and hours with wiping and restoring your phone.

Fun Times with Objective-C Runtime

Good reusable solutions to problems usually come out of necessity rather than luck or leisure. Over the past 18 months I was always faced with problems of how to simplify the data integrity, security, and aggregation of multiple datasources, usually in big organizations that would then be fed in an iOS app.

The first solution to this issue is to create the "man in the middle". Essentially a proxy/web services layer that acts as the gateway to all the necessary model objects. Utilizing tools such as Gearman, CouchDB, MySQL, PHP, Python, Memcache and a little bit of Apache to make everything complete I was able to deploy a lightweight standard response to the iOS app. As requests from various business groups would come in I found myself taking more and more out of the app and putting that logic into the services layer. Instead of having to rely on the app to handle security, data integrity, computational analysis, image manipulation, etc. I pushed that onto a light weight stack server side that could be scaled easily and mainainted without having to revision the app for basic model object changes. The added benefits that I saw later on were that I could adapt the response based upon the request…(Android, iOS, mobile web, desktop browser, etc.)

Now that I was concentrating more on feature sets, bug fixes and UI enhancements I was still stuck with having to manage various model classes within the app. This was a serious pain point to me. I wanted to be able to consolidate the various list and detail controllers that I had down to one instance of each.

Objective-C runtime to the rescue.

Let me be very clear that objective-c runtime is VERY powerful and not for the faint of heart, but when used effectively can be quite useful. The approach that I took was as follow.

Use the existing "standard" json format from a web services layer and restructure the "responseObject" so that it gave meta information about the model name, properties for class, which properties should be displayed on the list page and detail page. By structuring the json (xml, plist, etc) play load this way I now can push changes to model object and display anytime I want.

In the github project I use an "Employee" object as the example. The app should be able to show a list of employees (title and subtitle) and then a detail page for each employee that would display up to 3 properties.

I have included another json file for "Events" just so that I can show the flexibility. The same list/detail view controllers are now reused for two different objects.

The (current) limitations.

There is an assumption that your list view will use the subtitle uitableviewcell enum type.
Parsing and creation of classes doesn't take into account types other than nsstrings. I have an idea to remedy this, just haven't implemented yet.
What about images? Most list views have an image associated with it. What I am thinking about it having a flag set in the properties specifically for images so that they can be loaded asynchronously.

I will be adding in these features. I am still getting my head wrapped around some of the advanced features of runtime. Below are the links to resources that I used in my research.

Posted on slideshare is the presentation that I gave. It details the philosophical and practical importance of having an SOA that is similar to what I described. It was heavily inspired by the presentation written on the subject of NPR's SOA.

Download Github Example Project

View Presentation

Presentation References:

Cory Wiles Blog

If You Blog It They Will Come