发现 lua-visual-debug 显示哪个 (La)Tex 元素（打印到标准输出）？

Question

好的，我想我已经找到了某个地方 - 下面的胶水标有“GL”+一个数字（我不知道它是什么意思，但在Lua中显示为任何Lua节点tostring(node)- 也是胶水节点 - 并且似乎是唯一的），它看起来像这样：

测试-unc.png

问题在于将这个数字与 Latex 源中的实际元素联系起来；luatexref-t.pdf文档中没有提到它可能意味着什么，但它提到了（粘合？）节点的一些属性，可以用来获取更好的信息。

很快，我就没有足够的时间来正确写作了——首先，参考：

下面是一个有效的 PDF 流示例：

/F1 24 Tf            % font select; must have! /F1 must be correct ref.; size (24) can be arbitrary 
0 1 0 rg 100 100 Td  % position text - text reacts on fill (rg) only!
( Hello World ) Tj   % output text
%q  1.5 w  0 0 m  30 30 l  S  Q % example line q/Q push/restore graph. stack, width, moveto, lineto
0 1 0 RG  1.5 w  0 0 m  30 30 l  S % example colored line - reacts on stroke (RG) only!

第一名inspect.lua（https://github.com/kikito/inspect.lua) 在 /path/to/texlive/2011/texmf-dist/tex/luatex/lua-visual-debug/ 目录中；然后lua-visual-debug.lua像这样进行破解：

...
module(...,package.seeall)

-- https://stackoverflow.com/questions/9168058/lua-table-dump-to-console
-- https://github.com/kikito/inspect.lua/blob/master/inspect.lua
-- this is enough
local inspectfile = assert(loadfile("/path/to/texlive/2011/texmf-dist/tex/luatex/lua-visual-debug/inspect.lua"))
local inspect = inspectfile()
print("inspect is" .. inspect(inspect))
...
function show_page_elements(parent)
...
  elseif head.id == 10 then -- glue
      local wd = head.spec.width
      ...
      local pdfstring = node.new("whatsit","pdf_literal")
      local wd_bp = math.round(wd / number_sp_in_a_pdf_point,2)
      if parent.id == 0 then --hlist
        pdfstring.data = string.format("q %s [0.2] 0 d  0.5 w 0 0  m %g 0 l s Q",color,wd_bp)
      else -- vlist
        pdfstring.data = string.format("q 0.1 G 0.1 w -0.5 0 m 0.5 0 l -0.5 %g m 0.5 %g l s [0.2] 0 d  0.5 w 0.25 0  m 0.25 %g l s Q",-wd_bp,-wd_bp,-wd_bp)
      end
      -- q: gsave; Q grestore; Tf selectfont; Td Move text position; Tj show text
--~       pdfstring.data = pdfstring.data .. "  q  0 1 0 RG  0 0 m 40 40 l  Q" -- this shows, but not color; '0 0 m' is relative!
      -- from a latex: /F8 15.0 Tf 0 0 Td (abcd)Tj should works; wrong size is rendered, wrong /F8 not - and even with all correct, if a character in string is not in the exported font, it will NOT work!
      -- so, in latex source file, make sure the used letters are output once! e.g.
      -- \node[text=white,anchor=west] at (0,0) {glueGLUE0123456789.};
      headstr = tostring(head)
      headmidnum = string.match(headstr, '.*<%s*(%d+)%s*>')
      pdfstring.data = pdfstring.data .. string.format("  q  /F1 24 Tf  0 0 Td  (GL%s) Tj  Q", tostring(headmidnum))
      parent.list = node.insert_before(parent.list,head,pdfstring)
      -- tostring(head.length):: head.length: Invalid field id length for node type glue (0)
      print("Got glue; wd " .. tostring(wd) .. "; " ..  "; " .. tostring(head))
      print("hd: " .. head.id .. ", " .. head.subtype .. ", " .. tostring(head.attr) .. ", ''" .. tostring(head.spec) .. "'', " .. tostring(head.leader) )
      -- no head.name: invalid field id name for node type glue (0)
      if (head.spec ~= nil) then
        print("hd spec: " .. head.spec.width .. ", " .. head.spec.stretch .. ", " .. head.spec.stretch_order .. ", " .. head.spec.shrink .. ", " .. head.spec.shrink_order .. ", " .. tostring(head.spec.writable))
      end
      if (head.attr ~= nil) then -- always nil
        print("attr: " .. node.fields(head.attr.id, head.attr.subtype ))
      end
      -- extract (apparently) "sequence index": the middle number in headstr='<node 126920 < 107034 > 107021'
      if (headmidnum ~= nil) then
        print("mid: " .. headmidnum)
      else print ("NO headmidnum!") end
      --print(inspect(head.tostring)) -- no __tostring() -- both invalid; no .number or .next.number
      -- print(inspect(head) .. " -- " .. inspect(getmetatable(head)) .. "--" .. inspect(node.fields(head.id, head.subtype))) -- <userdata 1>
      --~ Got glue; wd 0; ; <node 126920 < 107034 > 107021 : glue 2>
      --~ <userdata 1> -- {
      --~   __eq = <function 1>,
      --~   __index = <function 2>,
      --~   __newindex = <function 3>,
      --~   __tostring = <function 4>
      --~ }--{ "id", "subtype", "attr", "spec", "leader",
      --~   [-1] = "prev",
      --~   [0] = "next"

-- PDF font declarations - these seem to change always :(
--~ /Font
--~ <<
--~ /F47 5 0 R
--~ /F20 6 0 R
--~ /F46 7 0 R
--~ /F16 8 0 R
--~ /F45 9 0 R
--~ /F44 10 0 R
--~ /F39 11 0 R
--~ /F38 12 0 R
--~ /F35 13 0 R
--~ /F34 14 0 R
--~ /F33 15 0 R
--~ >>
...

完成后，在终端/控制台（此处使用bash）中执行以下操作：

# note that print() from lua goes to stdout - but not to Latex log!
# so capture entire stdout to another log with redirection: 
lualatex test.tex 2>&1 | tee test-lua.log

# must unpack the produced test.pdf file:
pdftk test.pdf output test-unc.pdf uncompress


# if want to make sure our literal is inserted:
# must use less -L to show entire contents:
less -L test-unc.pdf # or
grep '/F1 ' test-unc.pdf | head -2

# find one actual font reference - grep for 'GLUEgl'
# (note that regardless of \mbox, the text GLUEglue... may 
#  end up split in the .pdf!

grep 'glueGL' test-unc.pdf
# example output:
# /F15 9.96264 Tf 1 0 0 1 274.444 12.896 Tm [(glueGLU)]TJ

# now replace the `/F1 24` we had previously with a 
# proper reference (/F15) and smaller font size (5):
sed -i 's_/F1 24_/F15 5_g' test-unc.pdf

现在可以test-unc.pdf打开（例如，在中evince），并且应该显示带标签的胶水，如上图所示。请注意，即使 PDF 查看器可能不会发出警告消息，PDF 仍会损坏，因为更改后，PDF xref 尾部可能不正确。您可能需要pdftk再次重新运行文件以尝试更正 - 但即使这样，我也无法使用 ImageMagickconvert来渲染test-unc.pdf（但我可以在 GIMP 中渲染它，这就是获得上图的方式）。

Answer 1