[l2h] LaTeX2HTML-2K for Win32: thanks, and a question

[Alan Bridle]

I wrote:
> > I normally process my LaTeX files to .dvi keeping any included
> > .eps graphics in the same input directory as my .tex source.
> > When converting the same .eps graphics into .gif images,
> > latex2html-2K.1beta is looking for them only in the output
> > subdirectory (where it writes images.tex), not in the input
> > directory. This makes it necessary to copy the .eps input files
> > into the output subdirectory before running latex2html, or to
> > run it with $NO_SUBDIR = 1; (both work around the issue, but
> > running with the default settings and all of my .tex and .eps
> > files in the same input directory fails because latex2html
> > cannot find the .eps files during the image conversion step). 

Ross Moore replied:

> Are you using  \includegraphics from the graphics/graphicx
> LaTeX packages ?
> This has code to take into account that extra directory level.

I was previously using epsf and epsffile within a figure environment 
to be processed by LaTeX.  Today, I tried using \includegraphics
with the graphics and graphicx packages.  Both behaved in just
the same way, i.e. the processing of images.tex from the
subdirectory failed at the dvips step because the .eps files 
could not be located in the directory above.

> 
> You can set environment variables within the .latex2html-init
> file for your job. (Certainly this works for Unix; with NT, I don't
> have the relevant experience, sorry.)

I note that this issue is indeed mentioned as a "known
problem" in the documentation that accompanied v99.1 for
Unix, and that setting $TEXINPUTS was mentioned there as a
possible solution.   Unfortunately, it doesn't seem to be
the cure in Win32, at least not in the NT/perl configuration 
I am using.  

I tried specifying $TEXINPUTS explicitly in my initialization 
file, but this did not correct the problem for me.  

The work-arounds of pre-placing copies of the .eps files in the 
sub-directory, or of not using the sub-directory option, are 
still the only routes that work for me thus far.

If I find an elegant solution that works, I'll report again. 
Thanks, Ross, for the suggestions.