The JPEG algorithm uses a combination of lossy and lossless steps. The algorithm breaks up an image into 8x8 blocks, known as Minimum Coding Units (MCU). If an image has dimensions that are not divisible by 8, MCUs along the right and bottom edges are padded. Each MCU is then processed through FFT/DCT (Fast Fourier Transform/Discrete Cosine Transform) and quantized. Quantization is the main lossy step. Then a series of lossless steps are performed, including lossless compression, are performed.

Lossless operations can be performed on JPEG by undoing the final lossless steps and performing lossless operations on the MCUs, such as rearranging them. If image dimensions are not divisible by 8 (or 16 if chroma subsampling is enabled), artifacts may appear along the edges because padded blocks will be along the wrong edge and the blocks that need padding won't be.

By default, exiftran discards these blocks. However, these blocks can be retained, and artifacts along the edges appear, if exiftran is called with the -np flag.

I suspect most digital cameras and cell phones produce images with dimensions that will not cause this problem because it makes optimal use of available resources to do so. This is the case for devices I have checked, made by Apple, Canon, FujiFilm, Lenovo, and Samsung.