LuaLaTeX parse PDF for unencoded small string DeviceCMYK?

Question

EDIT: It turns out that Lua is not needed. Since I call my TeX from script, the necessary investigation can be done directly, using findstr in Windows batch, or the equivalent in BASH. But that's using script prior to running the TeX file. The answers provided below will do it within the TeX file.

Question:

I use only LuaLaTeX, Windows and Linux. For cross-platform compatibility, and for non-admin users, I cannot invoke sed or most other shell-escape programs from within TeX, even if it is available. I can invoke from outside TeX with a non-admin batch or BASH script.

My document class produces PDF/X-1a. Thus, any included image must be DeviceGray or DeviceCMYK. Currently, I avoid potential problems by severely limiting TeX commands that include images or other PDF. I would like to lift the limitations, if possible.

Thanks to this very helpful answer of a related question, I can block using \includepdf if the included file has fonts in it. Now, I would like to block using \includepdf unless the file has DeviceGray or DeviceCMYK written into it.

In all cases, the included PDF is just one page, one image.

If I open a suitable PDF in a hex editor, I see something like this:

10 0 obj
/DeviceCMYK
endobj

The actual image follows as another object (with FlateDecode).

So, to rephrase my question: Is there any way LuaLaTeX can screen a PDF file, perhaps using methods similar to sed or grep, and provide a yes/no answer as to whether string DeviceCMYK or DeviceGray appears?

I myself can simply use Adobe Acrobat Pro, or perhaps ImageMagick. This is for the benefit of more general users, without having to step outside of LuaLaTeX.

Answer 1

A solution with epdf:

\documentclass{scrartcl}
\usepackage{luacode,pdfpages}
\begin{luacode*}
  function check_colorspaces(allowed_colorspace, name)
  local doc = epdf.open(name);
  if doc == nil then
    tex.sprint(luatexbase.catcodetables['latex-package'],
        "\\errmessage{Could not open " .. name .. "}{}{}\\@gobbletwo")
    return;
  else
    for pageno=1,doc:getNumPages() do
      local xobjs= doc:getCatalog():getPage(pageno):getResourceDict():lookup("XObject");
      if not xobjs:isNull() then
        for i=1,xobjs:dictGetLength() do
          xobjDict = xobjs:dictGetVal(i):streamGetDict()
          if xobjDict:lookup('Subtype'):getName() == 'Image' then
            if not allowed_colorspace[xobjDict:lookup('ColorSpace'):getName()] then
              tex.sprint(luatexbase.catcodetables['latex-package'], '\\@firstoftwo')
              return
            end
          end
        end
      end
    end
  end
  tex.sprint(luatexbase.catcodetables['latex-package'], '\\@secondoftwo')
  return;
end
\end{luacode*}
\newcommand\PDFHasDisallowedColorspaceTF[1]{%
  \directlua{check_colorspaces({DeviceCMYK=true, DeviceGray=true},"\luaescapestring{#1}")}%
}
\begin{document}
\PDFHasDisallowedColorspaceTF{image.pdf}{%
  \errmessage{image.pdf contains invalid colorspace!}%
}{%
  \includepdf[pages=-]{image.pdf}%
}
\end{document}

This does not check if the file contains one of the allowed colorspaces, it instead checks that no image in the PDF has an not allowed colorspace. If you have exactly one image, this should result in the same result, but if multiple images are included, this checks every one. If no image is found, this is not considered an error and colorspaces in the text are ignored.

This answer also can handle compressed PDF files.

Answer 2

Since the color space will be in the file in plain ASCII you can just search for the string. This might come with some performance penalty compared to epdf.

\documentclass{article}
\pagestyle{empty}
\begin{document}

\directlua{
local filename = "test.pdf"
local color_space_found = false
%
for line in io.lines(filename) do
    if string.find(line,"/DeviceGray") or string.find(line,"/DeviceCMYK") then
        color_space_found = true
        break
    end
end
%
if not color_space_found then
    error(filename .. " does not specify the correct color space")
end
}

\end{document}