[gs-devel] Converting pdf to eps

[Ralph Giles]

On Tue, Dec 11, 2007 at 03:47:14PM -0500, Yang wrote:

> ps2pdf: uses gs to produce postscript-3; results are terrible

I think you mean this produces PDF? We don't have a Postscript level 3 
output device.

> convert: identical to ps2pdf

I believe convert just calls ghostscript, so that is to be expected.

> ps2ps2: uses gs to produce postscript-2; for some files, the result 
> cannot be viewed in evince (errors); so far have always been able to 
> view in gs or display; seems to yield high quality as well

Ghostscript's ps2 output is really a limited PDF output with a PDF 
interpreter written in Postscript prepended. So it will be better or 
worse depending. It's intended for printing, and doesn't currently 
produce a DSC-conformant Postscript stream.

> Does pdftops perform lossless conversion? Does ps2ps2? They both seem to 
> produce better quality, but I'm not sure the conversion is completely 
> lossless, by which I mean I can fearlessly convert between pdf and ps 
> for ad infinitum and not worry about degradation in quality.

Completely lossless conversion between pdf and ps (or vice versa) isn't 
possible. The best you can hope for is that they render identically, but 
even that is difficult in some cases. It's best to pick one format for 
your work and only convert to the other for printing or export.

> Why does pdf2ps yield abysmal results? My eye tells me it's rasterizing 
> the fonts; is that true, and is that the only problem? Is there some way 
> I can tell pdf2ps to produce high-quality output/can I use ghostscript 
> to produce high-quality postscript-3? (I could find no documentation for 
> pdf2ps on the website, and the man page is minimal.)

pdf2ps does rasterize a lot of fonts. Have you tried passing -r300 or 
-r600 and seeing if that helps?

> Is postscript-2 inherently more verbose than postscript-3 (for 
> equivalent constructs)?

If my postscript-3 you mean PDF, then yes, postscript level 2 is more 
verbose for some documents. It offers fewer compression options, and the 
image model is a subset, so some contents must be converted to less 
abstract representation or rasterized.

Hope that's of some help.

 -r