I haven't done very much with OpenGL, but my understanding is that if SFML 3 were to use only the "modern" shader-based pipeline instead of the fixed functions it currently uses, it would have to do things like apply a drawable's transform matrix inside a vertex shader and a drawable's (interpolated) color(s) inside a fragment shader.
It seems like this would require SFML to have default shaders and shader parameters as implementation details which users like me would have to know about when using our own shader code. Does this seem like a problem? Or is sf::Shader a sufficiently "advanced" feature for that to be a non-issue?