It depends on your goals.

Do you just want to get this game made? Easy -- use what you already know and go with the most easily targeted platform.
From what you've said in your specific case that means using C++ and SDL to target Windows and Linux.
If the game is fun and polished you'll also have the option of trying to make some money and perhaps porting to other platforms -- either by doing a substantial re-write of the game, or by utilising the fact that SDL can be ported to a selection of other platforms.

Is your goal to learn new technologies? Again, this is a pretty simple case of choosing a language and/or libraries that would be suitable for your idea and setting about learning it. It's always great to learn new technologies and get access to more potential target platforms, and if this isn't something you've done before it'll be a great and interesting experience with plenty of things to learn.

Do you want the game to be successful and try to earn some money? This is more complicated...
In this case, you would want to start thinking now about how you might monetize the game and what method or methods of distribution would be most suitable.
Is the game-play well suited to the small touch-screens found on mobiles? Would it be well suited to browser-based arcades such as Kongregate or Newgrounds?
Can you target the platforms you're interested in with what you already know, or would it be necessary or easier to learn a new technology?


To me this sounds like the type of game that probably would be well suited to either mobile devices or in-browser. Perhaps you could try whipping up a prototype using the free version of Construct, or a similar tool. Once you're able to play around with a basic version of the game you might have a better idea of how suitable it is for certain platforms, and you could then proceed to either develop in an appropriate technology or simply continue with a licensed version of Construct.



Hope that's helpful!

Mostly if you want to focus on "getting it made," using an engine is the way to go.
On college I learned python using Panda3D (I was used to code with c++ and openGL only, so it was a HUGE gain), it was a great experience to learn something new and getting the game done. Also nice to learn how to deal with collisions, lights and other tools (such as blender and inkscape).

Also, have you considering using a c/c++ lib that compiles for android/ios (such as orx)?

If you don't have -- and aren't planning to get, or at least in some way gain access to -- at least one mobile device for each platform you probably don't want to develop for them. Whilst some of the development tools provide an emulator for testing, there can be problems you'll only find on the real device, and even things like balancing the game when using the proper input can make a huge difference in the final product.

Also I'm used to code in Java (using Eclipse) at work (business applications), so for writing an Android game I would use Java (using Eclipse).

Then you can also take a look at playn:
http://code.google.com/p/playn/

It uses java and you can import your game to multiple plataforms. From its site:

PlayN is a cross-platform game abstraction library for writing games that compile to:
Desktop Java
HTML5 Browsers
Android
iOS
Flash
More?

But more and more people I start talking to, say that this type of game will do much better as mobile app or web app. For example a web app has the advantage of easely distributing updates or store/compare high scores with other players.

They are right, the web is a great distribution channel: it addresses virtually everyone and it has an "implied" networking model, developing for the web is the easiest way of developing almost anything IMHO. Performances can be a problem but there are ways to address and solve it (just remember someone successfully put a playable Quake IV inside Chrome through WebGL and completely JavaScript-based physics).

The dilemma is: Would you take on this opportunity to learn something new (eg Android, HTML5,..) OR would you stick to what you know and focus on the game itself at first?
[/quote]

I would normally suggest to focus on the game itself, but learning to develop for the web is such a right investment that you should definitely do as soon as you can IMHO.

In order to speed up development you should consider an engine, and I can suggest Canvace www.canvace.com (though shamelessly self-promoting ): it provides great support for 2.5D isometric games.