Building quality software for Android - A post-mortem for Tweagle | Hacker News Reader

Building quality software for Android - A post-mortem for Tweagle

walkbase.com

32 pointsnjern27 comments

One of our two full-time developers on the project spent most of his time fighting a never-ending war against our XML views to make the app look just as good on a ZTE Blade as it does on a brand new Galaxy Nexus.

I'm definitely feeling that way these days.

When I first started learning Android, I was very excited about the resource system. The ecosystem is highly fragmented, but they give you the tools to deal with it!, I thought. How cool, I want to learn from their wisdom!

Six months later, I'm actively shopping for a different UI framework. This one makes easy things hard and hard things impossible. I was recently trying to make a button 40% of the screen width and aligned kinda leftish and kept from getting too big on wide screens. In CSS, that's couple lines, maybe thirty seconds. But in Android . . . well, I spent six hours thrashing through documentation and recommendations and opinions on StackOverflow before I gave up and wrote my own Button class.

It's really that bad.

Ever try to change the color of a button on Android? It requires something like fifteen 9-patch PNGs, two or three arcane XML files, and navigating a style hierarchy for your activities. No siree, no .myButton { border: 3 px solid #C88; } here.

Ever try to wrap text on a canvas? Or scale a font based on the display size? Or add an element to a view when a screen is a little bit larger? The hoops you have to jump through are simply amazing.

I'm convinced it's not fragmentation, per se, that's the problem. The web has even more fragmentation, and it handles it well enough. No, it's that the tools you're given for dealing with the fragmentation are so very poor.

I think your post highlights one of the problems with the Android framework when it comes to UI: it's difficult to know what the right approach is for a particular task, esp. when thinking about many OS versions and device types.

There are simple solutions to the problems you mention, but it is very difficult to know they even exist without several hours of research and trial and error. For framework environments like Android this slows app development and leads to apps that don't behave well. The tools and docs available right now need to be better if Android wants to be the dominate mobile platform for devs.

Yes, it's learning *nix all over again. There is an easy way to do things, which you cannot on any account find, until you complain and look like an idiot when someone tells you how to do it in twenty-eight characters.

I'm moving to the "It's not me, it's you" phase of the relationship faster this time, though.

Not really looking forward to apologizing for the weak graphical tools, however-

"Ever try to change the color of a button on Android?"

android:background="#ffee11" on the button?

Or use something like CSS variables (you know -- where CSS wants to go) and pull from a resource. Are you talking about something much more complex?

You can define a shape resource and then reference that in your button. Your button can be almost any shape and look you desire, via a shared resource file so it's easy to replicate elsewhere in your app. Making a shape occupy n-percent of the screen, or having padding left, right, above, below is trivial, and I'm predominately an NDK developer and don't even deal with this normally (but just quickly validated it. Rounded corner buttons with gradient backgrounds and specific padding occupying n-percentage of the screen).

I am left completely baffled by your complaints. There is nothing arcane about a shape XML file (it is profoundly clear, and is obviously intended to be reused so you aren't scattering disparate presentation code -- such as a button style -- throughout your code), and if you don't like that you can declare the color directly. Of course that is only one of your complaints, but if you don't understand something so elegant and trivial, I have to question the rest of your concerns.

Android shapes are great for getting basic shapes done and is a good alternative for 9-patch images. I also like the xml layout files and it's quite easy to understand. But if you are making a more complex design you will need to get those .9.png files involved.

android:background="#ffee11" on the button?

That . . . actually works! Well, now I'm embarrassed. I had previously had to go this route http://www.androidworks.com/changing-the-android-edittext-ui... with text fields, and assumed it was the only way.

How do you get buttons to occupy a percentage of the screen (I hesitate to ask, as it's probably something similarly trivial . . . but it's been such a painful subject for me that I really want to know!)?

How do you get buttons to occupy a percentage of the screen

In a LinearLayout take a look at layout_weight -- it is the key (or rather, one of the keys) to goodness. It really is an onion, and if you can get through the crying the flexibility comes in the layers.

Oh dear! That's really not the answer I was hoping for. That's about where I gave up last time -- somewhere around a dozen nested vertical and horizontal linear layouts with various weights, gravities and alignments that still wasn't giving me quite what I wanted.

Still, for correcting me on the button background thing alone, I suspect you've added years to my life. Dunno what the online equivalent of buying you a drink is, but . . . thank you.

Did you ever try to give a button one or two rounded corners (not the whole object, just some of the corners)?

Turns out that for pre-3.1 it will round the wrong corners (http://code.google.com/p/android/issues/detail?id=9161), meaning you need "corrected" drawables for API level 12+

That's one example of things that definitely made the whole XML experience a little trippy for us.

Where's the actual post mortem analysis? I wanted to see a discussion about actual problems encountered developing Tweagle specifically. Instead, this is just another exposition of the generic flames we've all seen done to death. Android apps have lots of screen resolutions (so what did you have to do to manage that?). "Some devices lack WiFi, A-GPS or both." (what is your fallback if the user doesn't have local internet or location services?).

I especially like this one: "Some people “hack” their Google Play app to be able to download incompatible apps and then write 1-star reviews when the app doesn’t work" ... with no discussion about why their app was marked incompatible in the first place.

You have a valid argument. My only excuse is that this is the first (publicly available) post-mortem I've written.

The answers to the questions you bring up have enough material for a blog post of their own, so I'll most likely end up doing a follow-up to this blog post where I address them.

Thanks for the feedback!

I was surprised by the word "post-mortem" in the title. Looks like the project is alive and well.

From Wikipedia: "A project post-mortem is a process, usually performed at the conclusion of a project, to determine and analyze elements of the project that were successful or unsuccessful. Project post-mortems are intended to inform process improvements which mitigate future risks and to promote iterative best practices."

I was trying to learn android recently and had to take a break due to increasing headaches. Theres so many issues. Some of which they mentioned in this article. One of which is that supposedly alot or some of the performance issues are fixed in 4.0. Of of curse nobody uses 4.0. So you have to just pretend those fixes dont exist.

And the xml views are a nightmare. I actually decided to learn ios instead maybe sometime next year.

An application I'm currently building (shameless plug - www.quaffic.com) will definitely benefit from this post. If nothing else, I feel affirmed that we're on the right track. From what you're saying, the value of ACRA is huge. That's my big takeaway at this point. I wondered what the normal process was for getting crash reports from users.

Also, what did you end up doing to alleviate your JSONObject problem? JSONSmart?

The article said they use Jackson for parsing json. A great library by the way.

Missed that. Thanks. I'll tuck that little tidbit away for later use.

Parsing json is a pain in android. You have to do so may things to do what should be a simple straightforward thing. One article mentioned that library but i never looked at it.

I can definitely recommend Jackson. Performance is stellar and it's easy to use.

It does make your APK a bit bigger (in our case, about 20% of the APK size comes from Jackson) so if you're only going to be parsing very little JSON, you can still make do with JSONObject.

Another neat library is JSON Simple if you looked for SAX style json parsing.

http://code.google.com/p/json-simple/

I use Google GSON on Android. Also a great library for serializing and deserializing JSON. I did not yet try the streaming api.

"..we’re using MS Paint while we would like to use Photoshop."

Ok i feel some of the authors pain on getting complex layouts right, managing lots of different density image resources, etc. But what in android dictates what graphic editing software you have to use? I'm happily using Photoshop for my resources.

What I meant was that the current system feels like Paint while I would like something with the power of Photoshop.

Sorry for the confusion, English isn't my native language :)

Ha! I don't think its your English that failed, its mine...

Now that I agree with. The ui tools are not first class though they're improving slowly. There's a ton of power in there but one hits a lot of speedbumps on the way to building good layouts that work on many devices.

Post-mortem? I don't get it.

"I’m a firm believer that premature optimization is the root of all evil."

Gah. Premature optimization is not a willful blindness to all matters performance related. It is a perversion of computer science that it keeps being twisted to such grossly different meanings that Knuth's original statement.

Don't celebrate and herald your invalid philosophical righteousness with the Knuth quote (as we've seen over, and over, and over, and over again here on HN. Yes, we all know the quote, and yes it always appears when someone discovers how crap their implementation is). You failed and wasted your time because you didn't consider performance a basic tenet of development, especially on limited mobile devices.

Don't take me wrong, I'm not a proponent of being "dumb" with performance either. However, I'm perfectly okay with using the standard libraries like JSONObject & AsyncTask until they prove to be too slow.

Thanks to a decent architecture the usage of JSONObject was confined to just a few methods that had to be refactored. There were a few more AsyncTasks we had to switch out, but in the end the refactoring took about two days which I would find to be perfectly acceptable "losses" compared to researching the performance of every single class before using it.

I definitely see where you're coming from though and I'll be sure to not throw that quote around without a better explanation in the future (at least on HN).

Reply on news.ycombinator.com