ps2ascii.ps

GhostScript的源代码

%    Copyright (C) 1991, 1995, 1996, 1998 Aladdin Enterprises.  All rights reserved.
% 
% This file is part of Aladdin Ghostscript.
% 
% Aladdin Ghostscript is distributed with NO WARRANTY OF ANY KIND.  No author
% or distributor accepts any responsibility for the consequences of using it,
% or for whether it serves any particular purpose or works at all, unless he
% or she says so in writing.  Refer to the Aladdin Ghostscript Free Public
% License (the "License") for full details.
% 
% Every copy of Aladdin Ghostscript must include a copy of the License,
% normally in a plain ASCII text file named PUBLIC.  The License grants you
% the right to copy, modify and redistribute Aladdin Ghostscript, but only
% under certain conditions described in the License.  Among other things, the
% License requires that the copyright notice and this notice be preserved on
% all copies.

% $Id: ps2ascii.ps $
% Extract the ASCII text from a PostScript file.  Nothing is displayed.
% Instead, ASCII information is written to stdout.  The idea is similar to
% Glenn Reid's `distillery', only a lot more simple-minded, and less robust.

% If SIMPLE is defined, just the text is written, with a guess at line
% breaks and word spacing.  If SIMPLE is not defined, lines are written
% to stdout as follows:
%
%	F <height> <width> (<fontname>)
%		Indicate the font height and the width of a space.
%
%	P
%		Indicate the end of the page.
% 
%	S <x> <y> (<string>) <width>
%		Display a string.
%
% <width> and <height> are integer dimensions in units of 1/720".
% <x> and <y> are integer coordinates, in units of 1/720", with the origin
%   at the lower left.
% <string> and <fontname> are strings represented with the standard
%   PostScript escape conventions.

% If COMPLEX is defined, the following additional types of lines are
% written to stdout.
%
%	C <r> <g> <b>
%		Indicate the current color.
%
%	I <x> <y> <width> <height>
%		Note the presence of an image.
%
%	R <x> <y> <width> <height>
%		Fill a rectangle.
%
% <r>, <g>, and <b> are RGB values expressed as integers between 0 and 1000.
%
% Note that future versions of this program (in COMPLEX mode) may add
%   other output elements, so programs parsing the output should be
%   prepared to ignore elements that they do not recognize.

% Note that this code will only work in all cases if systemdict is writable
% and if `binding' the definitions of operators defined as procedures
% is deferred.  For this reason, it is normally invoked with
%	gs -q -dNODISPLAY -dNOBIND -dWRITESYSTEMDICT ps2ascii.ps

% Thanks to:
%    J Greely <[e-mail]> for improvements to this code;
%    Jerry Whelan <[e-mail]> for motivating other improvements;
%    David M. Jones <[e-mail]> for improvements noted below.

%%  Additional modifications by David M. Jones
%%  ([e-mail]), December 23, 1997
%%  
%%  (a) Rewrote forall loop at the end of .show.write.  This fixes a
%%      stack leakage problem, but the changes are more significant
%%      than that.  
%% 
%%      .char.map includes the names of all characters in the
%%      StandardEncoding, ISOLatin1Encoding, OT1Encoding and
%%      T1Encoding vectors.  Thus, if the Encoding vector for the
%%      current font contains a name that is not in .char.map, it's
%%      redundant to check if the Encoding vector is equal to one of
%%      the known vectors.  Previous versions of ps2ascii would give
%%      up at this point, and substitute an asterisk (*) for the
%%      character.  I've taken the liberty of instead using the
%%      OT1Encoding vector to translate the character, on the grounds
%%      that in the cases I'm most interested in, a font without a
%%      useful Encoding vector was most likely created by a DVI to PS
%%      converter such as dvips or DVILASER (and OT1Encoding is
%%      largely compatible with StandardEncoding anyway).  [Note that
%%      this does not make my earlier changes to support dvips (see
%%      fix (a) under my 1996 changes) completely obsolete, since
%%      there's additional useful information I can extract in that
%%      case.]
%%      
%%      Overall, this should provide better support for some documents
%%      (e.g, DVILASER documents will no longer be translated into a
%%      series of *'s) without breaking any other documents any worse
%%      than they already were broken.
%% 
%%  (b) Fixed two bugs in dvips.df-tail: (1) changed "dup 127" to "dup
%%      128" to fix fencepost error, and (2) gave each font it's own
%%      FontName rather than having all fonts share the same name.
%%  
%%  (c) Added one further refinement to the heuristic for detecting
%%      paragraph breaks: do not ever start a new paragraph after a
%%      line ending in a hyphen.
%% 
%%  (d) Added a bunch of missing letters from the T1Encoding,
%%      OT1Encoding and ISOLatin1Encoding vectors to .letter.chars to
%%      improve hyphen-elimination algorithm.  This still won't help
%%      if there's no useful Encoding vector.
%% 
%%  NOTE: A better solution to the problem of missing Encoding vectors
%%  might be to redefine definefont to check whether the Encoding
%%  vector is sensible and, if not, replace it by a default.  This
%%  would alleviate the need for constant tests in the .show.write
%%  loop, as well as automatically solving the problem noted in fix
%%  (d) above, and the similar problem with .break.chars.  This should
%%  be investigated.  Also, the hyphen-elimination algorithm really
%%  needs to be looked at carefully and rethought.

%%* Modifications to ps2ascii.ps by David M. Jones
%%* ([e-mail]), June 25-July 8, 1996

%%* Modifications:
%%* 
%%* (a) added code to give better support for dvips files by providing
%%*     FontBBox's, FontName's and Encoding vectors for downloaded
%%*     bitmap fonts.  This is done by using dvips's start-hook to
%%*     overwrite the df-tail and D procedures that dvips uses to
%%*     define its Type 3 bitmap fonts.  Thus, this change should
%%*     provide better support for dvips-generated PS files without
%%*     affecting the handling of other documents.
%%* 
%%* (b) Fixed two bugs that could potentially affect any PS file, not
%%*     just those created by dvips: (1) added missing "get" operator
%%*     in .show.write and (2) fixed bug that caused a hyphen at the
%%*     end of a line to be replaced by a space rather than begin
%%*     deleted.  Note that the first bug was a source of stack
%%*     leakage, causing ps2ascii to run out of operand stack space
%%*     occasionally.
%%* 
%%*     Search for "%%* BF" to find these modifications.
%%*     
%%* (c) Improved the heuristic for determining whether a line break
%%*     has occurred and whether a line break represents a paragraph
%%*     break.  Previously, any change in the vertical position caused
%%*     a line break; now a line break is only registered if the
%%*     change is larger than the height of the current font.  This
%%*     means that superscripts, subscripts, and such things as
%%*     shifted accents generated by TeX won't cause line breaks.
%%*     Paragraph-recognition is now done by comparing the indentation
%%*     of the new line to the indentation of the previous line and by
%%*     comparing the vertical distance between the new line and the
%%*     previous line to the vertical distance between the previous
%%*     line and its predecessor.
%%*     
%%* (d) Added a hook for renaming the files where stdout and stderr
%%*     go.
%%* 
%%* In general, my additions or changes to the code are described in
%%* comments beginning with "%%*".  However, there are numerous other
%%* places where I have either re-formatted code or added comments to
%%* the code while I was trying to understand it.  These are usually
%%* not specially marked.
%%* 

/QUIET true def
systemdict wcheck { systemdict } { userdict } ifelse begin
/.max where { pop } { /.max { 2 copy lt { exch } if pop } bind def } ifelse
/COMPLEX dup where { pop true } { false } ifelse def
/SIMPLE dup where { pop true } { false } ifelse def
/setglobal where
 { pop currentglobal /setglobal load true setglobal }
 { { } }
ifelse

% Define a way to store and retrieve integers that survives save/restore.
/.i.string0 (0               ) def
/.i.string .i.string0 length string def
/.iget { cvi } bind def
/.iput { exch //.i.string exch copy cvs pop } bind def
/.inew { //.i.string0 dup length string copy } bind def

% We only want to redefine operators if they are defined already.

/codef { 1 index where { pop def } { pop pop } ifelse } def

% Redefine the end-of-page operators.

/erasepage { } codef
/copypage { SIMPLE { (\014) } { (P\n) } ifelse //print } codef
/showpage { copypage erasepage initgraphics } codef

% Redefine the fill operators to detect rectangles.

/.orderrect	% <llx> <lly> <urx> <ury> .orderrect <llx> <lly> <w> <h>
 {	% Ensure llx <= urx, lly <= ury.
   1 index 4 index lt { 4 2 roll } if
   dup 3 index lt { 3 1 roll exch } if
   exch 3 index sub exch 2 index sub
 } odef
/.fillcomplex
 {	% Do a first pass to see if the path is all rectangles in
	% the output coordinate system.  We don't worry about overlapping
	% rectangles that might be partially not filled.
	% Stack: mark llx0 lly0 urx0 ury0 ... true mark x0 y0 ...
   mark true mark
	% Add a final moveto so we pick up any trailing unclosed subpath.
   0 0 itransform moveto
    { .coord counttomark 2 gt
       { counttomark 4 gt { .fillcheckrect } { 4 2 roll pop pop }  ifelse }
      if
    }
    { .coord }
    { cleartomark not mark exit }
    { counttomark -2 roll 2 copy counttomark 2 roll .fillcheckrect }
   pathforall cleartomark
    { .showcolor counttomark 4 idiv
       { counttomark -4 roll .orderrect
	 (R ) //print .show==4
       }
      repeat pop
    }
    { cleartomark
    }
   ifelse
 } odef
/.fillcheckrect
 {	% Check whether the current subpath is a rectangle.
	% If it is, add it to the list of rectangles being accumulated;
	% if not exit the .fillcomplex loop.
	% The subpath has not been closed.
	% Stack: as in .fillcomplex, + newx newy
   counttomark 10 eq { 9 index 9 index 4 2 roll } if
   counttomark 12 ne { cleartomark not mark exit } if
   12 2 roll
	% Check for the two possible forms of rectangles:
	%	x0 y0  x0 y1  x1 y1  x1 y0  x0 y0
	%	x0 y0  x1 y0  x1 y1  x0 y1  x0 y0
   9 index 2 index eq 9 index 2 index eq and
   10 index 9 index eq
    {	% Check for first form.
      7 index 6 index eq and 6 index 5 index eq and 3 index 2 index eq and
    }
    {	% Check for second form.
      9 index 8 index eq and
      8 index 7 index eq and 5 index 4 index eq and 4 index 3 index eq and
    }
   ifelse not { cleartomark not mark exit } if
	% We have a rectangle.
   pop pop pop pop 4 2 roll pop pop 8 4 roll
 } odef
/eofill { COMPLEX { .fillcomplex } if newpath } codef
/fill { COMPLEX { .fillcomplex } if newpath } codef
/rectfill { gsave newpath .rectappend fill grestore } codef
/ueofill { gsave newpath uappend eofill grestore } codef
/ufill { gsave newpath uappend fill grestore } codef

% Redefine the stroke operators to detect rectangles.

/rectstroke
 { gsave newpath
   dup type dup /arraytype eq exch /packedarraytype eq or
    { dup length 6 eq { exch .rectappend concat } { .rectappend } ifelse }
    { .rectappend }
   ifelse stroke grestore
 } codef
/.strokeline	% <fromx> <fromy> <tox> <toy> .strokeline <tox> <toy>
		% Note: fromx and fromy are in output coordinates;
		% tox and toy are in user coordinates.
 { .coord 2 copy 6 2 roll .orderrect
	% Add in the line width.  Assume square or round caps.
   currentlinewidth 2 div dup .dcoord add abs 1 max 5 1 roll
   4 index add 4 1 roll 4 index add 4 1 roll
   4 index sub 4 1 roll 5 -1 roll sub 4 1 roll
   (R ) //print .show==4
 } odef
/.strokecomplex
 {	% Do a first pass to see if the path is all horizontal and vertical
	% lines in the output coordinate system.
	% Stack: true mark origx origy curx cury
   true mark null null null null
    { .coord 6 2 roll pop pop pop pop 2 copy }
    { .coord 1 index 4 index eq 1 index 4 index eq or
       { 4 2 roll pop pop }
       { cleartomark not mark exit }
      ifelse
    }
    { cleartomark not mark exit }
    { counttomark -2 roll 2 copy counttomark 2 roll
      1 index 4 index eq 1 index 4 index eq or
       { pop pop 2 copy }
       { cleartomark not mark exit }
      ifelse
    }
   pathforall cleartomark
   0 currentlinewidth .dcoord 0 eq exch 0 eq or and
	% Do the second pass to write out the rectangles.
	% Stack: origx origy curx cury
    { .showcolor null null null null
       { 6 2 roll pop pop pop pop 2 copy .coord }
       { .strokeline }
       { }
       { 3 index 3 index .strokeline }
      pathforall pop pop pop pop
    }
   if
 } odef
/stroke { COMPLEX { .strokecomplex } if newpath } codef
/ustroke
 { gsave newpath
   dup length 6 eq { exch uappend concat } { uappend } ifelse
   stroke grestore
 } codef

% The image operators must read the input and note the dimensions.
% Eventually we should redefine these to detect 1-bit-high all-black images,
% since this is how dvips does underlining (!).

/.noteimagerect		% <width> <height> <matrix> .noteimagerect -
 { COMPLEX
    { gsave setmatrix itransform 0 0 itransform
      grestore .coord 4 2 roll .coord .orderrect
      (I ) //print .show==4
    }
    { pop pop pop
    }
   ifelse
 } odef
/colorimage where
 { pop /colorimage
    { 1 index
       { dup 6 add index 1 index 6 add index 2 index 5 add index }
       { 6 index 6 index 5 index }
      ifelse .noteimagerect gsave nulldevice //colorimage grestore
    } codef
 } if
/.noteimage		% Arguments as for image[mask]
 { dup type /dicttype eq
    { dup /Width get 1 index /Height get 2 index /ImageMatrix get }
    { 4 index 4 index 3 index }
   ifelse .noteimagerect
 } odef
/image { .noteimage gsave nulldevice //image grestore } codef
/imagemask { .noteimage gsave nulldevice //imagemask grestore } codef

% Output the current color if necessary.
/.color.r .inew def
  .color.r -1 .iput		% make sure we write the color at the beginning
/.color.g .inew def
/.color.b .inew def
/.showcolor
 { COMPLEX
    { currentrgbcolor
      1000 mul round cvi
      3 1 roll 1000 mul round cvi
      exch 1000 mul round cvi
		% Stack: b g r
      dup //.color.r .iget eq
      2 index //.color.g .iget eq and
      3 index //.color.b .iget eq and
       { pop pop pop
       }
       { (C ) //print
	 dup //.color.r exch .iput .show==only
         ( ) //print dup //.color.g exch .iput .show==only
         ( ) //print dup //.color.b exch .iput .show==only
	 (\n) //print
       }
      ifelse
    }
   if
 } bind def

% Redefine `show'.

% Set things up so our output will be in tenths of a point, with origin at
% lower left.  This isolates us from the peculiarities of individual devices.

/.show.ident.matrix matrix def
/.show.ident {		% - .show.ident <scale> <matrix>
%   //.show.ident.matrix defaultmatrix
%               % Assume the original transformation is well-behaved.
%   0.1 0 2 index dtransform abs exch abs .max /.show.scale exch def
%   0.1 dup 3 -1 roll scale
  gsave initmatrix
		% Assume the original transformation is well-behaved.
  0.1 0 dtransform abs exch abs .max
  0.1 dup scale .show.ident.matrix currentmatrix
  grestore
} bind def

/.coord {		% <x> <y> .coord <x'> <y'>
  transform .show.ident exch pop itransform
  exch round cvi exch round cvi
} odef

/.dcoord {		% <dx> <dy> .coord <dx'> <dy'>
		% Transforming distances is trickier, because
		% the coordinate system might be rotated.
   .show.ident pop 3 1 roll
   exch 0 dtransform
    dup mul exch dup mul add sqrt
     2 index div round cvi
   exch 0 exch dtransform
    dup mul exch dup mul add sqrt
     3 -1 roll div round cvi
} odef

% Remember the current X, Y, and height.
/.show.x .inew def
/.show.y .inew def
/.show.height .inew def

% Remember the last character of the previous string; if it was a
% hyphen preceded by a letter, we didn't output the hyphen.

/.show.last (\000) def

% Remember the current font.
/.font.name 130 string def
/.font.name.length .inew def
/.font.height .inew def
/.font.width .inew def

%%* Also remember indentation of current line and previous vertical
%%* skip

/.show.indent .inew def
/.show.dy     .inew def

% We have to redirect stdout somehow....

/.show.stdout { (%stdout) (w) file } bind def

% Make sure writing will work even if a program uses =string.
/.show.string =string length string def
/.show.=string =string length string def
/.show==only
 { //=string //.show.=string copy pop
   dup type /stringtype eq
    { dup length //.show.string length le
       { dup rcheck { //.show.string copy } if
       } if
    } if
   .show.stdout exch write==only
   //.show.=string //=string copy pop
 } odef
/.show==4
 { 4 -1 roll .show==only ( ) //print
   3 -1 roll .show==only ( ) //print
   exch .show==only ( ) //print
   .show==only (\n) //print
 } odef

/.showwidth	% Same as stringwidth, but disable COMPLEX so that
		% we don't try to detect rectangles during BuildChar.
 { COMPLEX
    { /COMPLEX false def stringwidth /COMPLEX true def }
    { stringwidth }
   ifelse
 } odef

/.showfont	% <string> .showfont <string>
 { gsave
	% Try getting the height and width of the font from the FontBBox.
     currentfont /FontBBox .knownget not { {0 0 0 0} } if
     aload pop      % llx lly urx ury
     exch 4 -1 roll % lly ury urx llx
     sub            % lly ury dx
     3 1 roll exch  % dx ury lly
     sub            % dx dy
     2 copy .max 0 ne
      { currentfont /FontMatrix get dtransform
      }
      {	pop pop
	% Fonts produced by dvips, among other applications, have
	% BuildChar procedures that bomb out when given unexpected
	% characters, and there is no way to determine whether a given
	% character will do this.  So for Type 1 fonts, we measure a
	% typical character ('X'); for others, we punt.
	currentfont /FontType get 1 eq
	 { (X) .showwidth pop dup 1.3 mul
	 }
	 {	% No safe way to get the character size.  Punt.
	   0 0
	 }
	ifelse
      }
     ifelse .dcoord exch
     currentfont /FontName .knownget not { () } if
     dup type /stringtype ne { //.show.string cvs } if
   grestore
	% Stack: height width fontname
   SIMPLE
    { pop pop //.show.height exch .iput }
    { 2 index //.font.height .iget eq
      2 index //.font.width .iget eq and
      1 index //.font.name 0 //.font.name.length .iget getinterval eq and
       { pop pop pop
       }