switch ( Platform.get() )
{
    case WIN:
        bpp = 32;
        break;
    case MAC:
        bpp = 32;
        break;
    case NIX:
        bpp = 24;
        break;
    default:
        bpp = 24;
        break;
}

And that's about it in my case (bpp is bits per pixel btw, turns out Windows/OSX defaults to 32 whereas *nix platforms default to 24).

What kind of boilerplate?

outside of video/audio and sometimes networking, and sometimes threading

Even those are all easy enough to make portable. Use OpenGL, an audio library like FMOD or Wwise, and Boost/STL threading or a library like Intel's Threading Building Blocks. IO can be done via PhysicsFS, Boost/STL, and so on as well. All the windowing and input is handled by all the common window-system libraries like SDL.

Does anyone use vim/emacs as their "IDE"?

When I used Linux as a desktop/development OS (which I haven't for some years) I used Vim exclusively. These days a lot of people swear by QtCreator which also runs on Windows and OSX so it's a decent choice for a cross-platform IDE. I can't really stand any IDE I've put time into aside from Visual Studio anymore, though, though Sublime Text (which runs on Linux) make a good run.

How much boilerplate did you have to write?

The boilerplate is no more or less on Linux than it is elsewhere when you use a window-system toolkit like SDL2, GLFW, SFML, etc. and a build system tool like CMake. If you're writing raw OS API code (Win32, X11, etc.) there's tons of boilerplate to write but there's really no good reason to ever do that these days.

Since I started out with programming on UNIX I tend to use that same development environment even when on Windows.

In a typical Linux/UNIX project I generally find all of these tools useful.

Editor: Vim - Including the NETRW and Buffergator plugins. Helps navigating scattered source bases quickly. More importantly, it does not require a "project file" for source aware navigation to work. My biggest annoyance with Visual Studio or other IDEs.

Code completion / Navigation
- exuberant ctags (Vim supports by default now).
- cscope (performs some tasks ctags cannot)
- doxygen (With C the automatic flow chart generator is fantastic for development and learning codebases)

Compiler:
- gcc for main releases
- Any compiler I can get my hands on since they all potentially pick up on different issues.

Build System:
- cmake
- simple Makefiles for release builds and test harness

Continuous Integration: Jenkins on a bunch of VMs targeting many platforms

Misc: tmux for multiple terminals

Debuggers / Profilers:
- Valgrind
- gdb
- ElectricFence
- Intel VTune
- AddressSanitizer
- splint

Arguably GLX requires less boilerplate code than Windows alternatives when setting up an OpenGL rendering context.

http://www.opengl.org/wiki/Programming_OpenGL_in_Linux:_GLX_and_Xlib

You may notice it has a line count close to when using Glut.

Honestly, my development experience on Linux is mostly as a kind of playground. That is, the kind of coding I do there is mostly recreational -- messing around with Arduinos or other embedded stuff who's tool chain runs best on *nix, exercising my more 'academic' language pursuits like Haskell, or playing with neat linux stuff like Docker. Because I'm not so concerned with efficiency then, that informs some of my other choices. I use Vim not because I find it to be instantly productive, but because leaning Vim is a goal in itself -- its simply useful to know at least one lightweight text editor that you can use even from a remote text console -- and because the long walk leads to becoming possibly more efficient than WYSIWYG editors and fancy IDEs.

In general I like Linux because its a more 'orthogonal' environment in some sense of the word. The trend on Windows (and old-world Macs, the trend has reversed since OS X) has been towards more monolithic, swiss-army-knife applications, or to largely independent 'kingdoms' of apps that inter-operate nicely. The trend on *nix has always been towards a more inter-operable ecosystem of smaller and more focused tools that do one thing very well. It goes beyond the apps themselves though -- even where windows has functional equivalents (say grep vs. findstr) the command-line apps on Windows all use wildly-different argument formats, and not every app necessarily supports piping in ways it ought to. The *nix world largely has that problem sorted.