o Apache::Test is required early in Makefile.PL, so we can't rely on
  CPAN.pm's prereq_pm to fetch it if not available. Therefore we need
  to use Module::Install, which handles this kind of situations gracefully.

o look for 'XXX' in the source, they mark some bits of code that need work

o multipart_buffer_headers punts on headers more than 5k (which could
  happen with a really long form input name, for example). better than
  the old behavior of locking up, but we could be clever and actually
  handle this.

o should probably handle Apache::Request->param http://blah.com/?foo
  like CGI.pm does (turn it into a param named "keywords")

o mmap upload files ?

o improve/automate build process for dtcl

o $r->status reportedly returns success when a large upload is
  cancelled by the user (clicks stop) before the upload is complete.
  See http://marc.theaimsgroup.com/?l=apache-modperl&m=105561094424410&w=2
  The problem is likely to show up in fill_buffer's return value,
  but since we never check it...?  It would be nice if we had an 
  "interrupted upload" test to work from.

o solve the memory leak for:
  my $r = shift;
  $r = Apache::Request->new($r);
  as discussed at:
  http://marc.theaimsgroup.com/?t=102615343600005&r=1&w=2

o The behavior of our mfd parser is inconsistent wrt empty file uploads:
  An active upload_hook isn't called with a 0-byte file upload.

o Our Makefile.PL is lame.

o Investigate sporadic segfaults related to new upload code in Request.xs
  and older (pre-5.8.0) perls.