Can drawing to an FBO be done purely with a depth input texture, or do I also need a color attachment?

Question

I am trying to implement Exponential Shadow Maps and I've got it almost working. The part I am stuck on is the "optional" separable Gaussian blur of the depth map to give soft-looking shadows.

I am using an OpenGL Framebuffer Object attached to a temporary depth map, and I am trying to use a screen quad (two triangles) with the original depth map as an input texture. Then, I am doing the same for the second stage of the blur, with the input and output reversed. Unfortunately, this results in a depth buffer filled with 1.0 which suggests to me that nothing is being drawn.

In this example code I've found, they are using a color and depth attachment in order to draw to a screen quad with a blur shader. (unfortunately the only OpenGL example I could find)

I want to know if this is necessary or that they are only attaching and blurring the color buffer because they are storing their depth values in it.

In other words, I would like to know if it is actually possible to only use a depth buffer Framebuffer Object attachment in this scenario.

Answer 1

Specifically: blurShadowMap() uses fbo2 which has only one R32F texture set as "color".

I want to know if this is necessary or that they are only attaching and blurring the color buffer because they are storing their depth values in it.

Correct; depth is encoded into a "color" and stored in the color buffer

In other words, I would like to know if it is actually possible to only use a depth buffer Framebuffer Object attachment in this scenario.

Using the hardware depth-buffer is never required and you are free to attach your "depth"-buffer as the only color target. To render correct 3D depths to the depth map, you will need a "depth" buffer (color) to store the depths, as well as the actual hardware depth buffer(float) to ensure proper culling. While blurring in 2D, the hardware depth-buffer is not required.