This work establishes a theoretical framework for signal reconstruction from phase of linear measurement or affine linear measurement, and obtains some results on measurement number. We consider d-dimensional complex signals. In the scenario of linear measurement, at least 2d but no more than 4d-2 measurements are required to recover all signals up to a positive scalar. Moreover, we prove that 2d-1 generic measurements can recover generic signals or a specific signal, and 2d-1 is minimal in the sense that 2d-2 measurements can hardly recover signals. In the scenario of affine linear measurement, at least 2d+1 measurements but no more than 3d measurements are required to recover all signals exactly, while no less than 2d generic measurements can handle generic signals. Throughout the paper, we employ our theoretical results to understand existing results and algorithms of phase-only signal reconstruction.