There are still many problems with man2html.
Partly these are caused by the imprecise definition
of the man file format. (And the many buggy man pages.)
Partly by the incomplete implementation of the man/doc macro packages.
Partly by the imperfect emulation of troff.
Partly by the variation between various browsers in the
accepted html.
Partly just because man2html is buggy.

Of course in reality a man2html converter must contain
large parts of the troff source, so that it can be fed
with the defining macro packages and always do the right thing.

On a RedHat 5.0 system:
- /usr/man/mann/DirDlg.n is not formatted correctly.
  (It does not start with .TH)
- <i>bug_readline</i>@<i>prep.ai.mit.edu</i> does not generate
  a mailto: link. It should generate
    <a href="mailto:bug_readline@prep.ai.mit.edu>
      <i>bug_readline</i>@<i>prep.ai.mit.edu</i>
    </a>
  but this requires parsing of the surrounding html markup.
  Easy, but not done yet.
- Some manpages generate bad whatis information
  [where the name in the whatis line is not the filename of the man page].
  (E.g., tc589_cs.4, auto.master.5, pcmcia.5, proc.5 and autofs.8
   generate whatis lines
        3c589_cs (4)         - 3Com 3c589 Etherlink III device driver
	/etc/auto.master (5) - Master Map for automounter
	/etc/init.d/rc.d/autofs (8) - Control Script for automounter
	/etc/pcmcia/config (5) - PCMCIA card configuration database
	/proc (5)            - process information pseudo-filesystem
  )
  As a consequence, the corresponding manwhatis links are bad.
- Some manpages have bad .so information.
  (E.g., TIFFScanlineSize.3t contains .so TIFFsize.3t
   instead of .so man3/TIFFsize.3t .)
  This also confuses whatis, and generates bad links.
  (But right now these happen to work.)


Many browsers do not handle tables well (or at all).
Arena (arena-0.98.beta3-3) fails on many constructions.