[jonasmaturana]

When they say "all Android developers" you can be pretty sure they mean "most Android developers".

Working on several high profile apps here in Denmark I have felt the pain of not having enough real devices to test on. A bank just wont accept that you don't support one of the most common phone on the market because it has some problems with the camera.

Another project ran into the BigInteger.modPow is not thread safe issue way too late in the development process because we didn't test intensively on pre 2.3 devices.

After facing all these issues, my current project run automated test on 20 physical devices daily. It also helps us keep the UI looking sharp on all devices.

[estel]

Isn't "not supporting one of the most common phones" a different order of magnitude from what they're discussing here?

Hell yes that app should be supporting the Galaxy S2s and HTC Desires of this world, but the majority of those 400+ devices will probably be used by < 0.0001% of your users, particularly if the app is targeting a Western audience.

[gurkendoktor]

I maintain a business app for iOS and my client fully expects me to reproduce and fix bugs for obscure versions of iOS that only 0.0001% of people might still use, even if those people can (and should) just upgrade their OS. If it was my own app I wouldn't care.

[jonasmaturana]

I totally agree that 400+ devices is pretty crazy.

By looking at device usage stats I would say that 20 devices will get you above 90% in one region but you have to use closer to 40 if you aim for 90% on the global market. (Of course it depends on your user base)

However, having 400 devices on the shelf for debugging when you get some strange error report might be very valuable. I would love to have a bucket of CDMA phones available since I can't just go buy one in the store. (I live in Denmark)