My Blog Has Moved!

Due to limitations in blogger, I decided to move my blog to Wordpress. I ported all the posts over there and this blog is now obsolete. Here is my new blog.

Initial Reaction to React Native

This is to summarize my first experience with React Native. My overall impression is positive and setup went rather smoothly. I hit a few issues with build process and the environment that are worth listing here.

After generating the initial app (using 'react-native init') and trying to run it.

1. "Missing Iconions" icon error. It turns out I was missing a 'linking' step. Although I had added the icons using npm, they still needed to be linked to the native project, in this case the Xcode project file. 'react-native link' does the trick.
2. "Could not load app due to a missing require()". Weird error! A quick search suggested that there is a name mismatch. Once you generate the app, you cannot change the name that gets registered with the AppRegistry. 
3. Changing the name back did not fix the issue. I got another strange error but this time React gave me three suggestions to fix it. Apparently there is some caching of app content going on that needs to be cleaned up, either by deleting the npm_modules folder and clearing npm cache or simply by killing the packager process and restarting it. This error occurred a few times since then and every time killing the packager solved it.

As soon as I started writing some UI components, I realized that I must be repeating work that someone else has done. Then, I landed on Native Base, which looked promising. I hit a few issues with it as well which added to my initial list:

• The 'List' component skipped the first item. No matter how many items are included in the model, it always skipped the first one when rendering. It turns out that the List component has to be wrapped in a ScrollView.
• I was not able to mix usage of Native Base 'ListItem' with React Native TouchableHighlight to produce a selectable list item that shows a visual queue when tapped. It turns out that there is already a 'button' property that can be set on the 'List Item' to produce such a visual effect. It is not perfect, but it is OK for a toy app.

Got the basic Native Base components to work in my app. Back to using just React Native components. 

1. I attempted to add some navigation using NavigatorIOS. At first, it did not show any content I added to it. As I found out, some styles must be set on the navigator in order for it to work. Most notably, you have to use the {{flex:1}} style.
2. Specifying property types using PropTypes is neat. Just avoid the pitfall: PropTypes.something.isRequired is correct, whereas PropTypes.something.required produces mysterious error messages!
3. Also be careful when passing functions as properties. I mistakenly passed one of the properties as {someFunc(someParams)} instead of just {someFunc} which, of course, lead to invoking the function rather than passing a callback.

I am aggregating more notes that I will share later.

Compiling Java in Gulp - Fun Experience Chasing Plugin Issues

Working on the Qortoba project, I needed to add a Java tool for generating JavaScript classes from a Java interface. The need to compile a couple of java source files did not warrant  incorporating Maven or Ant into the project, and a shell script sounded too archaic.

Since I was already using gulp, it made sense to search for a plugin that does the job. Indeed, I found 'gulp-javac' which seemed simple enough to adopt. However, after installing with npm and trying to run it I found that it uses some ES6 features, in particular the dot dot dot 'spread' operator. I looked up the version of node vs. ES6 support and decided to install the latest (why not!).

Installing the new version with nvm with a breeze. However, after I switched the node version gulp was not found, even after I reinstalled it globally. I figured maybe there is some difference in how the new npm organizes packages, so I deleted the npm_modules folder and reinstalled.

Now, I was getting an error where the 'addLibraries' method was not found on the javac gulp task. It was not clear why, so I tried a few things. I replaced the short version of the command (which combines 'javac' and 'jar') with the long form, and this solved the 'addLibraries' problem.

But then I started getting a "file not found" error which seems to have resulted from failure to create a tmp directory. Looked inside the code and found that it uses a some package called 'tmp' for managing temporary files and folders. The 'dirSync' method that was used was being fed a parameter that allows it to delete the tmp folder on exit even if it was not empty. I removed that parameter in the source and got the same error. I created a two-liner test for the directory creation and I got the same result. It turns out that the 'tmp' package deletes folders upon exist anyways, unless you pass it a 'keep' parameter. Setting this parameter to true made the two-liner work but not the javac task!

At this point I decided to file an issue on github, and to my delight I got a quick reply indicating that the author found a bug with having the jar file created under any path that contains slashes! The bug fix solved my issue, and I then filed another issue for the 'addLibraries' mis-documentation and that too was fixed within hours. 

Presentation @ Google Developer Group LA

Finally got to put all bits of information together and present on connecting JavaScript to Java on Android. It so happens that the hosts, Victorious, talked to me afterwards and indicated that they had to solve the same problem! They added a very valuable hint: when deploying to the Google Play store you have to exclude any HTML/JS files from being processed (obfuscated) by ProGuard, otherwise any reference to injected Java objects will be rewritten causing obvious problems.

Slides can be found here.

Integrating Third-Party JavaScript Widgets with Angular

I was working on integrating a gauge widget into an Angular directive. The widget requires to be setup in JavaScript that looks something as follows:

       var gauge = new JustGage({             
              id : "the-element-id",             
              min: 0,             
              max: 100,             
              title: "Gauge Title"        
       });

It makes sense to represent the gauge widget as its own directive, so I created a directive for that purpose. Then, this imperative/procedural code that manipulates the DOM should go in the 'link' function of the directive. So far so good.

The gauge requires an element to render itself. The 'id' attribute of the gauge configuration specifies the ID of that element, which. In my case, I needed to specify the ID dynamically. The ID of the gauge element was supplied to the directive via an attribute:

      <my-gauge id="{{model.id}}" ... ></my-gauge>

On first attempt to run, I got an error saying that the element was not found. After debugging a bit and Googling a lot, I found out that the problem is mentioned in this post:

     "Notice that when the link function runs, any attributes that contain {{}}'s are 
     not evaluated yet (so if you try to examine the attributes, you'll get undefined)."

Then came $observe to the rescue. $observe, as opposed to $watch, allows you to listen on changes in the attributes of the directive element via the 'attrs' object passed to the link function. So in my link function I did: 

      attrs.$observe('id', function(newValue) {

           // initialize the gauge here

      };

   
And that solved the problem.

Conditional Injection - Testing Angular App on Android

I was developing an Angular-based app inside an Android app. That is, my Android Activity incorporated a WebView that loaded HTML/JavaScript pages that use Angular.

The Java code in the Andorid app communicated to with the JavaScript running inside the WebView by injecting a Java object into JavaScript as discussed in a previous post.

Cool! So?

The problem arose when I wanted to test the app. One could simply test the app within the Android IDE (Eclipse in this case), but the development cycle is much faster if I could just use a browser as one would normally do when building a web app. But the JavaScript depends on the injected Java objects in order to function, so how can I test my app in the browser in absence of the Java code that injects these objects. 

I embarked on the following solution:

• Encapsulate each of the injected Java objects in a "wrapper" Angular service.
• Also create Angular services that mock each of the wrapper services.
• When running in Android instantiate the wrapper services, otherwise instantiate the mock ones.

The trick here is that both services, the wrapper and corresponding mock, have the same name and only one of them can be instantiated in a particular run. Since there is no built in conditional compilation/inclusion directive akin to that of C, I came up with my own using a directive that looks like this:

       <script conditional-load
                  if-defined="<JavaObjectReference>" 
                  else-load="<MockService.js>" 
                  then-load="<WrapperService.js>">

The core code for the directive checks whether the specified reference is defined and loads either script accordingly.

.directive("conditionalLoad", function($window, ScriptLoaderService) {

    return {
        link: function(scope, element, attrs, controller) {

            ...

            if ($window[attrs.ifDefined]) {
                ScriptLoaderService.loadScript(attrs.thenLoad);
            }
            else if (attrs.elseLoad) {
                ScriptLoaderService.loadScript(attrs.elseLoad);
            }
        }
    }
})

With an instance of this directive for every injected Java object we are left with writing the ScriptLoaderService, which turns out to be non-trivial, and I will leave this to a separate post.

Twitter Broken GeoCode API

Using the GeoCode parameter with the Twitter Search API did not work for me (returned no results). A quick search turned up this post on the Twitter Community pages. It seems that issues were reported with this API over a year ago. Since then, it has not been fixed and now they say it may never be fixed!  I am not going to rant about how shameful it is to have a public API that you know does not work and just leave out there, but ...