We address the slow generation of crack networks as a problem of pattern formation. Issues of pattern selection and the associated statistical properties were considered by means of a detailed theoretical analysis and simulations of a discrete spring-block model. Developed after observations in desiccation experiments, the model describes the nucleation and propagation of cracks in a layer in contact with a frictional substrate. Competition between stress concentration at crack tips and pinning effect by friction leads to a cellular pattern. We characterized the events prior to cracking by a growth of correlation in the stress field, and those during cracking by progressive damages manifested in the number of broken bonds and energy releases. Qualitatively distinct regimes were shown to correspond to different stages of development. A host of scaling behaviors in measurable quantities were derived and verified. In particular, consistent with experiments, fragment area was found to be quadratic in the layer thickness and be smaller with increasing friction, which explains why morphologically similar patterns may occur over a diverse length scales.