Despite having used TEXTAREA, I hadn't thought much about this one.
Three things come to mind:
a. honor any line-ends found in the data, using whatever line-end is
native to the server platform (because you can't tell the client
platform). In effect this means the contents of TEXTAREA are
treated as if they were being displayed as <pre>
b. Ditto, sticking strictly to UNIX line-ends (LF alone).
c. Honor only double line-ends (LF-LF or CR/LF-CR/LF etc) as marking
logical line or para boundaries, and fold remaining data as if it
were a wrapped para.
///Peter