I think what the parent is trying to say is that PDF is not like HTML or text or other formats where the viewer is primarily responsible for a lot of the formatting --- in fact, a PDF page contains not much more than primitive instructions of the form "move to X, Y"; "set font to F"; "draw text 'Some text here'" (some pathological cases issue individual moves and draws for each character) --- so expecting all PDF viewers to be able to somehow "reverse-engineer" or "decompile" that set of low-level drawing instructions into more semantic entities like lines of text or even words in order to reformat the text is a little too much.

Anyone who has tried selecting text from a two-column PDF page will also quickly realise the nature of the problem.