Here is one example: For a true text file, it does not make a big
difference if your line endings are LF or CR or a combination of both
(which ones are used depends on your operating system and text editor
software used). A good text editor can juggle all these and give you what a
human would consider the correct view of the document. In a PDF file, it
does matter quite a bit if you use LF or CR/LF - the first one only adds
one byte to the file, the second one adds two bytes. Because everything in
the PDF file is referenced via the byte offset from the Xref table, having
a line ending be one vs. two bytes makes a big difference. In one case,
your offsets are correct, in the second one they are not, and therefore the
PDF file is corrupt.
Contrary to what lrosenth says, I believe you can create a PDF file in
NotePad++ (or any other text editor), and I do so regularly in my
trainings, but you have to be absolutely certain you understand how byte
offsets work and how to get them correct. Even though you can do this,
normally you would not (unless it's a training exercise). It's much easier
to write out the PDF file in software and keep track of all byte offsets as
the objects are generated so that you can then create a valid Xref table.
If you want to know more about the potential problems with the Xref table,
see this blog post I wrote a few years ago:
http://khkonsulting.com/2013/01/the-trouble-with-the-xref-table/