Wednesday, November 25, 2015

Gradle Performance Tips from Android Dev Summit

Posted by Drew Hannay on November 25, 2015 No comments
Yesterday was the last day of the Android Dev Summit conference by Google. The conference was packed with deep-dive tech talks on some of the most relevant topics to Android devs today, including Gradle build optimizations, performance testing tools, and data binding tips. I left the conference with a long list of things I can change to improve the performance of my build and tests, and be ready for new tooling features from Google like Instant Run.

 The full list of improvements will take a while to implement, but there were three simple fixes I was able to make in a day that had a dramatic impact on the performance of my builds.

 ------------------------------------------------------------------------

#1 Don't add timestamps, etc to BuildConfig fields in debug builds

It's very convenient to add extra fields to your BuildConfig file. My app adds the git commit sha and the build time so we can display this information inside the app on the developer settings screen. Unfortunately, both of these things change very frequently; timestamp obviously changes every time you build. This means the BuildConfig file needs to be regenerated on every build, and any files that depend on it (probably many files in the app) need to be recompiled as well, essentially killing the benefits of incremental builds.

The fix for this one is to only generate these fields in release builds. In debug fields, we just set them to the constant string "DEBUG".

For example:
buildTypes {
  debug {
    buildConfigField 'String', 'GIT_SHA', "\"DEBUG\""
  }
  release {
    buildConfigField 'String', 'GIT_SHA', "\"${gitSha()}\""
  }
}

------------------------------------------------------------------------

#2 Use the official api for generating code

Version 0.7.0 of the Android Gradle plugin added two api methods: registerJavaGeneratingTask() and registerResGeneratingTask(). These let you register a custom gradle task that will generate code, and will automatically set up the code to be generated at the proper time and indexed & recognized as source by the IDE.

My app used a custom task to generate model files from API data schemas, but we were manually inserting it into the build pipeline without using the official api. The downside of this is that the task was never recognized as "up to date", which forced all models (and all classes that depended on those models) to be recompiled on every build, removing the benefits of incremental builds. Before switching to the official API, running "gradle assembleGoogleDebug" consistently took ~2 minutes, even if there were no code changes. After switching, running the task with no code changes takes ~16 seconds.

------------------------------------------------------------------------

#3 Fix dependency conflicts between different configurations

This one isn't a performance optimization, but rather a way to increase the maintainability of your build script. I ran into this issue a long time ago on my app; Android testing works by loading your app apk and your test apk in the same classpath. If the two configurations depend on different versions of the same library, you'll run into crashes at runtime due to the conflict. The most frequent offender is the support annotations jar, since almost every modern Android library depends on it. Gradle already forces your build to use the same version of shared transitive dependencies, but this only works within the same configuration.

I fixed this previously using Gradle's resolutionStrategy option to force all configurations to resolve to the same version of a dependency, but this isn't a great solution because it requires you to hardcode the version of your dependency in a second location and there's a strong chance you'll forget to update it. The better option is to just add the dependency normally, but add it to both your prod and test configurations.

For example:
compile "com.android.support:support-annotations:${supportLibVersion}"
androidTestCompile "com.android.support:support-annotations:${supportLibVersion}"

------------------------------------------------------------------------

This is just the low hanging fruit that I was able to optimize in a day. Going forward, I'll be following up on several other lessons learned from the conference and hopefully improving build times and incremental compilation times even further. Be sure to check out the session videos from the Android Dev Summit conference if you want to learn more!

Sunday, October 11, 2015

How to ask a good question

Posted by Drew Hannay on October 11, 2015 No comments
In any technical job, you will inevitably run into situations where you simply get stuck. You’ve tried to find the next step, but aren’t making any progress. Rather than waste time spinning your wheels, you decide to ask a co-worker for help. However, you don’t want to waste anyone’s time or be the annoying employee that can’t figure out anything on his own. Here are some tips for how to ask a question while still leaving a good impression on your teammates.

Before asking, be able to answer a few questions…

  • What have you tried so far? Make sure you do some initial investigation on your own before asking for help. You may find that five extra minutes of searching will get you the answer just as fast as asking someone else. When you do need to ask for help, make sure to share what you’ve already tried, so your co-worker doesn’t repeat the same steps.
  • What information do you have about the problem? Keep track of anything you think might be relevant so you can share it with the person you ask for help. Did the problem start happening right after upgrading a related piece of software? That’s probably worth sharing.
  • What are the next things you think you should try? It should be pretty rare that you get to the point where you’re truly out of ideas for how to investigate your problem. If your next step is “google the error message,” then you probably haven’t spent enough time investigating on your own. On the other hand, if your next step is “debug the source code of the OS kernel,” you’ve probably waited a bit too long to ask for advice. The trick is to make sure you’ve done your due diligence, but haven’t wasted too much time on a problem that your co-worker might be able to solve in five minutes.

Still stuck? Time to bring in the cavalry…

  • Start with a wide audience and narrow if necessary. Does your team have a group chat room? That’s a good first place to ask, since many people will see your question and be able to either provide an answer or point you in the right direction.
  • Don’t interrupt, if possible. If you need to target your question at a specific person, start with an instant message. This lets them respond to you when it’s convenient for them. Dropping by their desk unannounced may break their concentration or take them away from another important task.
  • Start with a short summary of the problem. There’s always a chance your co-worker has seen this before, so give them enough information to recognize the problem, but there’s no need to give a multi-page summary right off the bat.

Once you've asked...

  • Be respectful of the person's time. If you're asking about something that might take a while, like building the codebase or running tests, rather than standing at their desk while you wait to see if the fix works, you can collect a series of things to try. Write them down so you don't forget, and then go back and try them on your own. Feel free to send a message to the person who helped you to let them when and how things got resolved.
  • Pay attention to how the person helping you investigates the problem. What do they do that you didn’t? Sometimes they will just “know” the answer; if this is the case, ask them how you could have figured out the solution on your own. Could documentation be improved for the process? Volunteer to record your findings so others can benefit from your solution.
  • For coding-related questions, make sure your IDE/keyboard/etc are using standard shortcuts and settings. If the person helping you needs to use your computer, be ready to turn off vim mode and your Dvorak keyboard settings to make it easy for her to navigate and type.
  • Pay back the favor! If someone else runs into the same problem in the future, help him or her find the answer and share your findings. This is the best way to scale knowledge and improve the skills of your entire company.

Friday, January 9, 2015

How intuitive is Android Lollipop?

Posted by Drew Hannay on January 09, 2015 , No comments
My wife and I spent Christmas with my in-laws, as we've done for the past several years. When we arrived, they made their request for this year's "tech support project": they wanted to move their emails and calendar from MSN to Gmail. If you're the go-to IT person for your family, you're probably thinking "Just what I wanted to do on my Christmas break...migrate thousands of emails and contacts to Gmail"... but I was thrilled. See, what my in-laws didn't know is that we were already planning on giving them each a Nexus 4 for Christmas and having their email, contacts, and calendar already in Google's realm would make it much easier to start using their new Android devices.

Out with the old...

Their old phones were tiny-screened, heavily OEM-modified, extremely buggy Android phones running the four-year-old Gingerbread operating system. They used them for phone calls, texting, and the occasional Hangouts group chat, but they weren't really capable of much else. Doing something like email on their phone was unimaginable, because email was something that was (in their minds) tied to their laptop.
For years they have paid for MSN Premium and have continued to pay for the antiquated service because, let's be honest: it worked well enough for them and why put the energy into learning something new? But they had finally gotten fed up with waiting multiple minutes for their email to load and were ready to make the switch.

...and in with the new

I migrated all their data to Gmail, Google Calendar, and Google Contacts and then it was time to show them how to use their new tools. For the most part, the web features were pretty simple; maybe a few buttons have moved around, but an email program is an email program. I explained Gmail's label system and how it was different from folders, showed them how to send email from their MSN addresses through Gmail, and how the Social and Promotions tabs keep their inbox clear of clutter. Making new Calendar appointments was similarly simple. The most interesting insights came when it was time to set up their Nexus 4 devices.

Give me tricks that work everywhere

As I walked them through how to use the device, it was fascinating to see them interact with each new app. One of the Android Design Principles is "Give me tricks that work everywhere" and I can definitely confirm that worked for my in-laws. My father-in-law quickly latched on to the swipe-to-dismiss gesture and merrily tried to swipe away anything he didn't want to see. The cool part? It usually worked! The pattern is now pervasive enough that swiping away a card, an email, or a notification will usually do what you wanted. Another trick I showed them was tapping on a contact picture to enter multiple select mode, which worked across many of the apps they used.

Web & App consistency is a win/win

Question: Where do you find your labels in Gmail? Answer: On the left side. Wait, was I talking about the Android app, the iOS app, or the web? It doesn't matter! How do you write a new email? There's a big red Compose button. Sure, it's a pencil icon on mobile, but it's still a (relatively) big red button. This consistency helped my in-laws tremendously. All the knowledge they gained learning the web interface of Gmail, Calendar, Photos, and more could be directly applied to the app's Android counterpart. They didn't need to learn an entirely new interface, which greatly increases the chances they'll actually use these things on their phones. From the app developer's perspective, it's also great...you don't need to come up with both a web interface and a mobile interface; you can just make minor tweaks and have your app scale across screens. Interface consistency is a win for everyone.

Visual cues work

Android Lollipop's Material Design encourages using motion to gives users a hint about how to use your app. When you tap a notification on the lock screen, you get a subtle message telling you to double tap to open. When you take a photo with the camera app, the picture peeks out from the right side of the screen, letting you know you can swipe right to view your photos. When you tap the status bar (something my in-laws do frequently as they forget they should pull it down), your notifications peek out and encourage you to swipe down to open the notification drawer. These cues, at least in my limited set of users, work amazingly well. As I watched them interact with their phones, they would try to perform some action, get a visual hint from the system about what to do, and then reach success relatively quickly.

What didn't work

There are still some areas in Android that aren't very intuitive and caused some struggles for my in-laws. The thing we spent the most time reviewing was home screen widgets, which heavily utilize one of the most undiscoverable features of Android - the long press. To add a new widget, you must long press on the background, click "Widgets", then long press again on the widget you want and (don't lift your finger or you'll mess it up!) drag it to the right location. To resize a widget, you must long press it yet again and then lift your finger (which is okay this time, it won't mess things up) and the resizing grabbers will appear. To remove a widget, you must long press it and drag it to the top of my screen. I spent several painful minutes watching my father-in-law try to swipe-to-dismiss a widget before finally intervening and reminding him what to do.

Take aways for Third Party developers

On their old phones, my in-laws used the text-messaging app, Hangouts, and the Dialer. Needless to say they aren't big app users. Now they're using several more of the stock Google apps, but don't really have much interest in installing third-party apps (maybe that will be next year's Christmas project). The one exception is that my mother-in-law uses Facebook, so she installed both Facebook and Facebook Messenger. Immediately she was confused by the non-standard behavior in those apps. "Why can't I swipe away my conversation with Suzanne? Why aren't my groups on the left side like on my computer? Why doesn't it select more than one conversation when I click on a person's picture?" I'd argue that these aren't features of Google apps, they're features of Android apps. Supporting swipe-to-dismiss or a floating action button probably won't cause any damage to your brand and it can really help decrease the amount of new stuff a user has to learn to use your app. And the easier your app is to use, the more likely they are to use it, right?
Subscribe to: Posts ( Atom )