View Single Post
Posts: 1,418 | Thanked: 1,541 times | Joined on Feb 2008
#53
Originally Posted by Benson View Post
I doubt my position is typical for much of anyone, but my biggest obstacle was the difficulty in establishing an on-device build environment.
This isn't really typical. A typical mobile app developer will cross-compile and package applicaiton on a PC (usually under Windows, usually with MSVS or Eclipse as IDE) and send it to the device for testing. Some may run it on the emulator first.

The main problem with Maemo SDK is that it is Linux-specific, so most of these traditional developers are initially at loss: they are forced to install and learn a new OS. The VMware "solution" is actually adding to complexity, not making things simpler.

The second problem is, of course, Autoconf. Maemo SDK documentation does not make it clear that you can develop without Autoconf and developing with Autoconf is just annoying.

The third problem is Debian packaging. In a traditional SDK, you provide a single text file that tells the SDK which files you are going to include into the package, at what locations, and how you identify the package (i.e. package name, id, etc). In Maemo SDK, you have to create a separate directory for packaging and fiddle with multiple config files to get things right. The documentation is non-existant or misleading, so most work is done by following examples from existing packages.

So, if I were going to speed up application development, I would start by creating a simpler packaging system (or adopting one from something like FreeBSD). Secondly, I would base all SDK examples and tutorials around simpler build process that does not involve Autoconf. Most Maemo applications are specific to the Maemo platform and thus do not really need Autoconf. Thirdly, I would actually invest time and money into creating a traditional SDK running in Windows. Maybe using coLinux as base.

Last edited by fms; 2009-07-01 at 14:19.
 

The Following 3 Users Say Thank You to fms For This Useful Post: