why phonegap is a winner

by Caner Turkmen
on Thu 11 December 2014

I've spent some time in the last few months playing with iOS. (I'm in the process of launching a toy project as well.) I've shared some observations on iOS, Cocoa, Objective-C, and Swift before. However, after discovering Phonegap/Cordova, I found that I was able to develop a 2x better application in 1/3 the time. I highly recommend the framework for rapid prototyping of small-scale projects. Here's why:

Pros:

I have some experience as a full-stack web developer. This is why, doing the design and layouting in HTML5/CSS is just more intutitive to me.
Again, for interactions, I'm used to Javascript. I've played with both node.js, and front-end frameworks before. I strongly believe Javascript is superior to both Swift and Objective-C. It's higher level, and much more powerful.
PhoneGap plays really well with the iOS developer tools (simulator, XCode, etc.)
Your app and dev environment is "code-only" and fully transparent. No point-and-click as in the case where you would hook up an IBOutlet in XCode development.

Cons:

Granted that you lose a lot of control over the actual computing (no access to threads was extremely annoying to me - and no remedies too as Javascript is single-threaded)
There is a decisive performance overhead;

Losing that control over the machinery is annoying at times, but not nearly as annoying as having to take a substring like this:

var endix = min((self.titleText as NSString).length,100) as Int
var activityItems: [String] = [self.titleText.substringToIndex(advance(self.titleText.startIndex, endix)) + "...", self.url];

as opposed to this:

var message = active.text().substring(0,115) + " " + active.jqmData("url");

dividing sparse matrices in scipy

by Caner Turkmen
on Fri 14 November 2014

Category: numpy scipy python division sparse

I spent quite some time in the last week trying to stabilize an implementation of Nonnegative Spectral Clustering I was working on. This is when I came across a peculiar situation I just wanted to put out there.

For an application I'm working on, I must be able to calculate a nearest-neighbor affinity matrix for about 120K records and work rigorously on this matrix. So sparse was the natural way to go.

However, it was not long after that I ran into a numerical stability problem. The problem was interesting in that I was already controlling for Div/0 errors. After getting down to where the problem occured with ipdb I was able to pinpoint the issue to where the numpy.divide function was being used with two sparse matrices as numerator and denominator.

Peculiarly, where A and B are two CSR matrices of the same shape, A contains zeros and B does not contain zeros.

np.seterr(all='raise')
div = np.divide(A, B)
np.any(np.isnan(div)) # returns True, does not raise exceptions

# likewise
div = np.divide(B,A) # raises exception for division by 0

The solution in my case was to offset the numerator as well, with epsilon. So using np.finfo:

Eps = csr_matrix(np.ones(A.shape) * np.finfo(float).eps)
div = np.divide(A+Eps, B)

I was able to remedy the problem. The code, for all interested.

minix dev environment setup on virtualbox and mac host

by Caner Turkmen
on Mon 10 November 2014

Category: minix virtualbox unix

I've spent few hours today trying to figure out how to best set up a Minix v3.3.0 development machine on my Mac OS X 10.9.4 host with VirtualBox. After what is not less than 5 times of tearing down and setting back up the machine, I believe I have what is a good working version.

After you follow the guide and getting a minix machine to boot, first we install some goodies on our system which will be required later.

$ pkgin update
$ pkgin in vim openssh bash git-base clang binutils

Might take some time to install though.

Next, I need the source. I know I have build 588a35b, which apparently is also a git commit hash. Although in multiple versions of the docs, Minix says the source is under /usr/src by default, I just couldn't find it. So, below are the commands for obtaining the Minix3 source, putting them in the right place in the filesystem, and 'checking out' the build I'm currently running on. Again, the git clone may take some time to complete.

$ cd /usr
$ git clone git://github.com/minix3/minix src
$ git checkout 588a35b

The tty terminal environment was not really appealing to me, as I preferred my Mac's bash, so ideally I had to set up a way to ssh into my Minix virtual machine. To do that properly, here's what I did.

On the VirtualBox guest go to Settings > Network > Port Forwarding. Set any rule name, the host port to 2222 (or any port number > 1023 really), and the guest port to 22 (Do not touch the other fields, you want TCP, and the IP fields should be blank). Now in my configuration, the host to guest connection does not initially work because the Minix user root has no password set. To set a password, in the guest terminal (Minix) type:

$ passwd

Then go to the host shell. Now we will get a brand new RSA key, and load it onto the Minix box:

$ ssh-keygen -b 4096 -t rsa # when prompted for path to save, enter ~/.ssh/id_rsa_minix
$ ssh-add ~/.ssh/id_rsa_minix
$ scp -P 2222 ~/.ssh/id_rsa_minix.pub root@127.0.0.1/root/.ssh/id_rsa.pub

Here, you will need to enter the password you set in the previous step, when prompted.

Now go to the Minix guest terminal:

$ cd /root/.ssh
$ touch authorized_keys
$ cat id_rsa.pub >> authorized_keys

Now through the host, the following command should work, and you should get a nice SSH connection to your local Minix virtual box:

$ ssh -vp 2222 root@127.0.0.1

If all goes well, you have a Minix VirtualBox v3.3.0, which includes the source code, checked out to reflect current build. Furthermore, you can SSH into that virtual machine from the comfort of your host OS. Enjoy!

musings on swift

by Caner Turkmen
on Sun 09 November 2014

Category: Software Programming iOS Swift

I think it was a few months back when I took up iOS programming, and not too long after I decided to start out did Apple launch two new versions of the operating system, along with Swift, a brand new language.

I had the unique experience of starting out as a beginner on two programming languages against the same platform and framework: Objective-C and Swift.

So I can offer some critique for the languages, from the standpoint of a beginner. Caveat emptor. I'm not a programming languages expert, and certainly not a mobile development professional. I'm just someone who has had some experience on other platforms, and decided to try out a new language.

Here we go:

Objective-C syntax had archaic looks, which was quite well justified. The language had broken away from other languages at the advent of object-oriented programming, and all of the other languages came from the other arm: C++. The square bracket syntax as well as still calling 'alloc' was quite a turn-off for the modern programmer.

I was already predicting a new programming language for Apple platforms, and had big hopes in store for Swift. However, I was somewhat disappointed by what it brought. Here are some of the reasons:

Optionals, in my opinion, just don't work. For strictly typed languages, the Java convention of non-nullable primitives and nullable Object types should have been the norm. In the little iOS experience I've had, I can't see the reason why you would need to have the optionals construct.
I have my doubts, furthermore, about implicit and explicit unwrapping. The whole business with question and exclamation marks just adds to the whole confusion. Having used Ruby, my instinct of question marks is boolean-valued functions, and exclamation marks: mutating methods.
The Cocoa API has not changed, so that's good. However, Swift has brought along its new types so now we have just more types to choose from: Array or NSArray? String or NSString? One programming language, in my opinion, has one string type, not a handful.
Syntax has not gotten any better with closures, that is "blocks" of code executed as callbacks etc.
Mobile development, by limitations of the platform, is inherently low level: as memory real estate is limited and a tighter grasp on the memory footprint is needed. It's hard not to feel we've given up some control.
I understand that not having try-catch type structure is a design choice Apple team made, I just can't understand why. It's a complex API, and a complex language, exceptions will happen. I'm yet to find out how exceptions are handled in Swift.

But then, I'm just a newbie. More to come on this later.