1. Better error reporting: report the line number instead of the
   paragraph number.

2. Act upon the =encoding directive.

CURRENT PROBLEMS IN THE MODULE
(this list is not exhaustive).

Cannot specify the author in META elements. Does not specify
pod2html version in META elements.

Index is commented out rather than removed.

Cannot suppress HTML preamble/postamble.

Excessive use of <p></p>.

The POD directory scanning code could be simplified by using to
File::Find for the heavy lifting.

The external cross-referencing code is a baroque construction of
empirical code.

The link resolution code is baffling (IOW producing <a
href="$deep_magic">).

Heuristics for spotting links in pure text contains unreachable
code.

Unwinding of L<C<B<...>>> and variants has some buggy edge cases,
obsolete tag Z<foo> not excised correctly.

Trailing X< in para incorrect error message.

POD not transformed in =for pod2html this I<word>

--htmlroot must not have a trailing /

=for comment warns.

Cannot resolve bug #9385 due to a fundamental design decision to
read the data file in paragraph mode.

Many internal item targets map to the same link (e.g.
	=item do
	=item do {block}
wind up with the same name.

Uses global variables, Getopt uses a separate set of variables that
are then used to overwrite the global variables.

Line 122:

if( ${$dataref}[$i] =~ /^\s+$/m and $dataref->[$i] !~ /^\s/ ){

Can never be true (and mixed dereferencing styles).  Change made
following Jeff Pinyan's bug report in 2001, but the generated HTML
is correct anyway.

Line 147:

$Backlink = _html_escape($Backlink) if defined $Backlink;

The variable is initialised to an empty string, thus the statement
modifier can never be false. Similar cases abound (checking for
defined() and then non-empty, non-false, ...)

Line 151:

unless (@ARGV && $ARGV[0])

Cannot process a file named '0'

Line 330:

_process_head( $1, $2, $Doindex && $index )

At this point, the file has already been analysed once. If there
were =head directives, then $index will be defined.

Here in the second pass, we encounter a =head directive, and will
render it, so logically $index could never be undefined the second
time through.

Line 351:

next if @Begin_Stack && $Begin_Stack[-1] eq 'html'

Lots of twisty conditionals, all different

	(e.g. $Begin_Stack[-1] ne 'html')

to confuse the unwary throughout the code.

Line 380:

Experimental (undocumented) table recognition heuristics: emitted
HTML is not compliant with anything (lacking </tr> and </td>).

Line 481:

$opt_netscape variable is configured. Does nothing, not warned as
deprecated.

Line 1110:

return if $Ignore;

Can't reach this, since the sub is never called if $Ignore is active.

Line 1808:

my $uo = URI::file->new($destination_file,$source)->abs;

dumps core, since the abs() method requires a parameter.

Line 1956:

sub fragment_id_obfuscated()

Routine is no longer be called from module, was never documented.

Test suite has make-work code to deal with its EBCDIC-incompatible
output.