I only partially agree, and I was a bit too lazy to point all this out in the beginning. The matter of spotting a URI inside a text is quite difficult. But I think that the fact is that many of those characters are valid for a URI, so it's probably an error putting them right after a URI without separating them from the URI itself with a space.

In particular, according to the standard the full stop is an unreserved character, and the comma and the semicolon are sub-delimiters that play no role in the HTTP scheme. This is why http://www.polettix.it/, and http://www.polettix.it/ciao, are perfectly valid HTTP URIs even with the comma. There is no really correct solution for this: if you keep them you'll annoy your audience most of the time, if you decide to keep them out you're ignoring that they are perfectly valid characters in a URI. I would simply stick to the simplest solution, i.e. the global substitution in this case.

I also didn't elaborate about why the OP was having difficulties with h(er|is) approach, which was basically due to the fact that splitting on spaces... leaves the newlines. And there are other spacing chars to consider, too, so something along these lines:

my @parts = split /(\s+)/, $message; # parents preserve spaces
# operate on @parts as in the OP
my $filtered = join '', @parts;
[download]