path: root/dviware/screenview/vms/crudetype.web

% ADAPTED FROM DVITYPE, VERSION 2.6.

% REVISIONS:
% 9/86: clarify names of global variables, supply hooks for attempted
%           Hewlett-Packard Laserjet version.
% 1/88: Several bugfixes. Chiefly the noscheme bug (TFM files without coding
%       schemes)
%       Also, added some MATH EXTENSION character codes.
% 4/88: Unix change file by P. King.
%
% 6/88: Changes to produce a "normal" VMS text file by Andrew Trevorrow.
%       All such changes are flagged by AKT comments.

% Here is TeX material that gets inserted after \input webmac
\def\hang{\hangindent 3em\indent\ignorespaces}
\font\ninerm=cmr9
\let\mc=\ninerm % medium caps for names like PASCAL
\def\PASCAL{{\mc PASCAL}}

\def\(#1){} % this is used to make section names sort themselves better
\def\9#1{} % this is used for sort keys in the index

\def\title{Crudetype}
\def\contentspagenumber{1}
\def\topofcontents{\null
  \def\titlepage{F} % include headline on the contents page
  \def\rheader{\mainfont\hfil \contentspagenumber}
  \vfill
  \centerline{\titlefont Crudetype}
  \vskip 50pt
  \centerline{An adaptable device driver (Version 1, 1988)}
  \vskip 50pt
  \centerline{R.M.Damerell,} \vskip 30pt
  \centerline{Mathematics Dept.,} \vskip 20pt
  \centerline{Royal Holloway and Bedford College,} \vskip 20pt
  \centerline{Egham, Surrey, U.K.} \vskip 20pt
\vfill}
\pageno=\contentspagenumber \advance\pageno by 1


% These macros for verbatim scanning are copied from MANMAC.TEX. But we cant
% use the vertical bar for a temporary escape character as WEAVE catches it.
% So we will use ! instead and hope for the best
\chardef\other=12
\def\ttverbatim{\begingroup
  \catcode`\\=\other  \catcode`\{=\other  \catcode`\}=\other  \catcode`\$=\other
  \catcode`\&=\other  \catcode`\#=\other  \catcode`\%=\other  \catcode`\~=\other
  \catcode`\_=\other  \catcode`\^=\other
  \obeyspaces \obeylines \tt}

\outer\def\begintt{$$\let\par=\endgraf \ttverbatim \parskip=0pt
  \catcode`\!=0 \rightskip-5pc \ttfinish}
{\catcode`\!=0 !catcode`!\=\other   % ! is temporary escape character
  !obeylines !obeyspaces    % end of line is active
  !gdef!ttfinish#1^^M#2\endtt{#1!vbox{#2}!endgroup$$}}

\def\up{\hbox{\tt{\char'013}}}
\def\markarrow#1{\vtop{\hbox{#1}\up}}

@* Introduction.

COPYRIGHT ( C ) R.M.Damerell, 1988.

Permission is given to any person to make and distribute copies of this
software, subject to the following conditions:

1. All copies of the software must carry an exact copy of this notice.

2. This software is distributed free of charge, "AS IS" with absolutely no
guarantee of performance. Any persons receiving or using this software must do
so entirely at their own risk. Neither the authors nor their institutions
accept any liability for any defects of this software, or for any consequential
loss or damage however caused.

3. Any person who changes this software must clearly mark it as modified and
add a note describing the changes made.

This is an experimental version and no guarantee of performance is given.
I would like to receive bug reports, same address or electronic mail to
DAMERELL at ARPA.UCL.CS.NSS. \par\vskip 0.5in

This program was originally based on D.E.Knuth's program \.{DVItype}, but so
many changes were needed for various reasons that there is hardly any of the
original code left. The purpose of this program is to provide a framework for
users to write \TeX\ device drivers for a variety of `crude' devices. Roughly
speaking, `crude' means any printer that cannot print the fonts that Metafont
generates. This would include daisy-wheels and most impact dot-matrix
printers. Considered as output printers for \TeX, such devices usually have
some of the following misfeatures: \item
1. Coarse resolution.\item
2. Restricted character set. \item
3. Some printers cannot do reverse line feeds, some can, and tear the paper.
\item
4. Slow interface between CPU and printer.\par

Although such printers cannot do justice to \TeX\ output, drivers for them
are still needed. Some users cannot afford high quality printers. Some can
only afford to use them for final output; so they need to make proofs on a
cheaper printer. Also, anybody who has a high quality printer may well need
to refer to various \.{WEB} files while writing a driver for it. These can
become illegible in critical places. Here is a sample from \.{DVItype}:
\begintt

A |fix_word| whose respective bytes are $(a,b,c,d)$ represents the number
$$x=\left\{\vcenter{\halign{$#$,\hfil\qquad&if $#$\hfil\cr
b\cdot2^{-4}+c\cdot2^{-12}+d\cdot2^{-20}&a=0;\cr
-16+b\cdot2^{-4}+c\cdot2^{-12}+d\cdot2^{-20}&a=255.\cr}}\right.$$
\endtt

Using the basic  (line printer) version of \.{Crudetype}, we can get a copy of
these formulae which is at least legible, even though the result may not be at
all pleasant to look at. A further difficulty with conventional drivers is
that most of these use the algorithm `paint a page of pixels, send it down the
line'. This places a heavy load on both the host computer and the link to the
printer. Of course, one can try to reduce this load by various optimisations,
(e.g. by writing critical bits of code in machine language) but this makes the
program non-portable, and often introduces bugs. \.{Crudetype} is written
entirely in \PASCAL, without any attempt at optimisation. When compiled on a
VAX 780 with the NO-OPTIMISE, CHECK and DEBUG qualifiers it runs at about 2--3
seconds a page. These times are highly variable, and the VMS optimiser reduces
them by about 10-15\%.

@ Printers vary enormously both in their capabilities and in the commands that
drive them. The behaviour of \.{Crudetype} is controlled by a large number of
constants, which supposedly describe how the target printer does things. This
does have the disadvantage that the user must compile a separate copy of the
program for each different printer, and also devise some way to ensure that he
uses the right version for the intended printer. But the only alternative
seemed to be that \.{Crudetype} should read and parse a file describing the
printer and this appeared to be unbearably messy. Ideally, these constants
should be so designed that: \item
1. Any decent printer can be driven by assigning the right values to these
constants and recompiling. \item
2. If the printer is properly documented, it should be immediately obvious
what are the correct values for all these constants.

At present I do not have enough experience of different printers to come near
this ideal. In particular, some printers can download characters. The
problems of writing a program to support this facility in proper generality
are horrible and ghastly. I have not made any serious attempt yet to tackle
them. There are just a few places where a hook appears, and I hope eventually
to attach actual routines for downloading.

Some of the more obvious problems of downloading are: when can you download?
At any time? start of page? or only at start of document? Can you load one
character, or must you load a whole font at a time? How much memory does the
printer provide for down loading? How efficiently does it use its memory? What
does it do when it runs out? Can you clear out old fonts to make more space?
What is the format of a down-load command? What parameters does it need, in
what order, with what punctuation? In what order must pixels be sent? Should
they be compressed, and how?

@* Implementation.

The original version of \.{Crudetype} was aimed at a line printer, (because
everybody has these) and was written on the VAX-VMS operating system. The
intention is that this program should be easily adaptable both to other
systems and to other printers. So most of it is written in Standard \PASCAL.
(It is not possible to tell exactly how much of it is Standard, as we do not
have a certified compiler.) But in some places, it is necessary to use
extensions. In particular, \.{Crudetype} must read the font files, whose names
are dynamically specified. That would be impossible in pure \PASCAL.
\.{Crudetype} also uses non-Standard code in order to talk to the user's
terminal. It asks for the name of the \.{DVI} file, and for the first page and
the number of pages to print. If an operating system forbids terminal
interaction, the installer will have to find another way to give the program
this information. As file handling is inevitably system-dependent, I have here
allowed myself a lot of latitude in using VMS-specific procedures. If
\.{Crudetype} cannot find a file, it will ask the user for another name. On
the other hand, all files are read and written sequentially, and I have got
rid of all uses of the default |case| statement. The intention is that all the
system-dependent stuff goes near the top of the file, and all
printer-dependent stuff at the end. Then with any luck you can merely
concatenate Change files for the local system and the local printer, instead
of having to merge them. All the code that is known to be non-Standard has
been carefully segregated from the rest of the program. It amounts to about 20
lines out of 750.
@.System dependencies@>

It is clearly impossible to predict what difficulties will appear in trying
to install \.{Crudetype} on other systems, it would seem to be advisable to
get the line printer version working before trying to adapt it for any other
printer. To try to ease the process, I propose to distribute several test
files with the program. These are of the form SAMPLE.TEX, SAMPLE.DVI and
SAMPLE.PRI (the line printer output).

Although `crude' printers differ very much in their capacities, one thing
they nearly all have in common is that they cannot feed the paper backwards.
Some printers cant |Backfeed| at all; some tear the paper, and others let the
paper slip and so lose position. Therefore it seems to be essential to process
each page as follows: first copy the page into a suitable structure, then sort
it by vertical and horizontal position, then print it.

@* Main Program.

@d banner=='This is Crudetype, Version 1, copyright, experimental'
{printed when the program starts}

@p program crudetype
  @<Declarations@>
  begin
    @<Initialize@>;
    read_BOP;          {AKT: was at start of For each page...}
    repeat
      @<For each page of \.{DVI}, print it if desired@>
    until time_to_stop ;
    @<Clean up afterwards@>;
    666:               {AKT: come here instead of crashing!}
  end.

@ Now here are some of the messy things we must do to satisfy the rules of
\PASCAL.

@<Declarations @>=
  (@<Files @>) ;
  label 666;      {AKT}
  const @< Constants in the outer block @>
  type @< Types in the outer block @>
  var @< Globals in the outer block @>
  @<Forward declarations @>
  @<Lowest level procedures @>
  @<Medium level procedures @>
  @<Top level procedures@>

@ @<Initialize@>=
  @<Set initial values@>
  @<Open terminal channels@>
  @<Assign character codes@>
  @<Dialog with user@>
  @<Open \.{DVI} file@>
  @<Open |printfile|@>
  @<Read \.{DVI} preamble @>

@ @<Glob...@>=
  in_i, in_j :integer; {loop index for initializations}

@ Next, here are some macros for common programming idioms.

@d incr(#) == #:=#+1 {increase a variable by unity}
@d decr(#) == #:=#-1 {decrease a variable by unity}
@d do_nothing == {empty statement}
@d exit == 732
@d return == goto exit
 {Go here when a loop ends abnormally}

@ The next two procedures are very primitive debugging aids.  All internally
detected errors call |error|. Then they can be caught (in VMS) by the debugger
command\begintt
  DBG>set break error \endtt
If a fatal error occurs, then the program will force a crash. With the VMS
debugger, you can then interrogate variables, etc. I chose the square root of
$-1$ as this does not figure prominently in \TeX-related programs.
@^square root@>

@<Forward...@>=
procedure error   ; begin end;
@#
procedure crash;
var u: real;
begin
  goto 666;       {AKT: was
  u := -1 ;
  error;
  u := sqrt(u) ;
                  }
end;

@* Interface to Operating System, 1: Material specific to VAX/VMS.

The purpose of these sections is to try to give a reasonable interface between
the operating system and the rest of the program, which is supposed to be
Standard \PASCAL. Nearly all the non-Standard code is concerned with file
handling and the lowest level of I/O. This is an area where Standard \PASCAL\
seems to be particularly weak. This particular section contains all of the
most system dependent code, and it will probably have to be entirely rewritten
for any other machine. It is hoped that most of the later sections will work
on a wide range of machines. Everything here is system dependent, so there is
no point in indexing each module separately.
@.System dependencies@>

@ The character set. I have here deleted all the code from \.{DVItype} that
translates from characters to small integers and back.  This is because we have
to do a quite different translation anyway. If it is necessary to put that
code back in, then it will probably be necessary to insert \begintt
define zchr(#) == xchr[#] \endtt
because of the different brackets.

@d zchr == chr
@d zord == ord

@ Here are some other system-dependent types. We use double length arithmetic.
The VMS-specific function |dble| converts its argument to double precision.
Integers are 32 bits in VMS. Normally, I use |integer| whenever the bit length
is unimportant, but I use subranges in the |page_record| type, as this allows
packing and may improve the program's performance.

@d real_num == double
@d make_double == dble {convert a |real| to double length}
@d max_half = 32767

@<Types...@>=
  byte = 0..255 ;
  i_word = -max_half-1 .. max_half ;

@ Here we consider the lowest level of file handling. The main input file is
the |dvi_file|. Output for printing goes to |printfile| and diagnostic output
to |term_out|. The terms |display| and |print| are used instead of |write| so
that output may be redirected if desired. Throughout the program, it is
assumed that |@!write| appends its argument(s) to the current record of the
selected file, and |@!write_ln| ends that record and sends it off; this
behaviour is specified in the \PASCAL\ Standard. If these assumptions are
false, it will require major restructuring of the program.

These macros describe how we use the terminal. VMS actually opens the terminal
channels for us, but we get a better style of output by re-opening it, and
using these funny-looking macros to write to it. We can then print a stream of
progress reports without falling foul of a finite record length.

@d term_in==input {the terminal, considered as an input file}
@d term_out==output {and output}
@d i_reset_terminal == do_nothing    {Switch terminal to input}
@d o_rewrite_terminal == do_nothing
  {and back to output. VMS does all this automatically}

@d display(#)==write_ln(term_out, #)
@d display_ln(#)==write_ln(term_out, #, chr(13), chr(10))
@d c_con == @=carriage_control@>

@d warn (#)==begin display_ln('Error: ', #); error; end
@d abort(#)==begin display_ln('Fatal: ', #); crash; end
@d bad_dvi(#)==abort('Bad DVI file: ',# )

@<Files @>= term_in, term_out

@ @<Open terminal channels@>=
  open(term_out, 'SYS$OUTPUT', c_con := none) ;

@ @<Const...@>= can_interact = true ;

@ The printed output goes to |@!printfile|.

@d print(#)==write(printfile, #)
@d print_ln ==write_ln(printfile )

@<Files@>= , printfile

@ @<Glob...@>= printfile : text ;
  print_name: var_string ;

@ In this section we generate a name for the printed file. Essentially, this
involves deleting the `.DVI' at the end of the \.{DVI} filename and adding
`.PRI' instead. But: the \.{DVI} file might be in a funny place, and it might
have a funny extension. Of course, this code all depends crucially on the VMS
file name format, and it will probably be a lot more complicated on systems
that do not allow the elaborate facilities of the VMS |open| command.

@<Open |printfile|@>=
  print_name := dvi_name ;
  chop_tail(print_name, ':') ;
  chop_tail(print_name, ']') ;
  {Chop off directory, disc, and perhaps a logical name}
  chop_top(print_name, '.') ;
  chop_top(print_name, ';') ;
  {and extension and version number}
  append(print_name, print_end) ;     {In VMS, usually  `.PRI'}
  display ('Output is ');             {AKT: was PRINTFILE IS}
  string_show(print_name);
  display_ln(' ') ;
    @.Printfile is...@>

@ VMS \PASCAL\ allows 3 types of carriage control, called |list|, |fortran|,
or |none|. No doubt other systems will have other peculiar types of carriage
control. In VMS, |none| is to be used if at all possible, but some printers
insist on a line feed after every carriage return. Roughly speaking, |@!list|
directs the operating system to put a CR--LF at the end of each record when the
file is printed. |@!fortran| means that a Fortran-type carriage control
character must be put at the start of each record, and \.{Crudetype} assumes
that this must be inserted explicitly. One type of run-time error that causes
a lot of trouble occurs if you try to write too many characters onto one
record of the |printfile|. I have tried to defeat this by declaring a very
long record length.

@ @d r_len == @=record_length@>

@<Open |printfile|@>=
  if fortran then
  open (printfile, print_name.data, c_con := fortran, r_len := 30000 )
  else if list then
  open(printfile, print_name.data, c_con := list, r_len := 30000 )
  else
  open(printfile, print_name.data, c_con := none, r_len := 30000 ) ;
  rewrite(printfile) ;

@ Now here is the lowest-level procedure for opening binary files. This will
have to be rewritten to run on any other system. The VMS |open| procedure
tries to open the file with the given name; if bits of the name are missing,
it can obtain them from the |default| parameter. It generates a non-zero
|status| if it fails.

@d close_binary(#)==
    close(# , @=error := continue@> )

@<Lowest...@>=
  function open_binary
  (var f_f: byte_file; name: var_string; other_name: def_name_type): boolean;
  var s: integer;
  begin
    close_binary(f_f );
      {in case the file was left open}
    open(f_f, name.data , readonly, ,, fixed, default := other_name,
      @=error := continue@> );
    s := status(f_f) ;
    if s <> 0 then open_binary := false
    else begin
      reset(f_f , @=error := continue@> );
      s := status(f_f) ;
      open_binary := (s = 0 );
    end;
  end;

@ Here we define system-dependent properties of these files. The easiest way
to tell VMS where to look for files is by giving them default names. These
should all be the same length if possible.

@d block_length = 512

@<Const...@>=
  dvi_def =   '          *.DVI' ;
  tfm_def =   'TEX$FONTS:*.TFM' ;
  pxl_def =   'TEX$PIXLDIR:*.*' ;

@ @<Types...@>= def_name_type = packed array [1..15] of char ;

@ Here are macros for the adaptable merge sort. See the section on sorting for
explanation.

@d image(#) == pool[#]
@d create == incr(cell)
@d link_type == page_i
@d first_cell == cell := 0
@d wipe_out(#) ==
@d declare_pool ==  pool: array [page_i] of page_record;
@d garbage == cell := zzz ;

@ These upper bounds are put in to catch runaway arguments.

@<Const...@>=
  page_max = 10000 ;
  max_line_size = 1000;

@* Interface to Operating System, 2: Terminal input and output.

When \.{DVItype} begins, it engages the user in a brief dialog so that the
options will be specified. This version of \.{Crudetype} does the same. This
requires nonstandard \PASCAL\ constructions to handle the online interaction.
So it may be necessary on some systems to omit the dialog. If so, the
installer must find some way to get the \.{DVI} file name into the |@!buffer|.

@<Dialog...@>=
  {AKT: removed display_ln(banner, ' --- ', device_ID) ;}
  buffer := blank ;
  repeat
    ask_prompt('DVI file name? ');
    dvi_name := buffer ;
  until dvi_name.len > 0 ;
  ask_prompt('First page? (default = 0) ' );
  first_page := get_number(0 ) ;
  ask_prompt('maximum no. of pages? (default = 1000000) ' ) ;
  max_pages := get_number(1000000) ;
    @.DVI file name?@>@.First page?@>@.max. no. of pages?@>

@ Most characters in \TeX\ fonts are narrower than line-printer characters. So
we must spread them out to make them fit. Originally, this was done by
multiplying \.{DVI} distances by a constant factor |h_fudge|. This is all
right for one size of type but it tends to fail for other sizes because if the
predominant type size is larger than expected, then rounding with a constant
factor makes everything\qquad\ very\qquad\ badly\qquad\ spread\qquad\ out. It
seemed that the least bad way to tackle this is to allow the user to specify
an extra magnification factor.

@<Dialog...@>=
  {AKT: removed
  if can_interact then display_ln(
    'What magnification?  This must be an integer, and is a percentage');
  }
  ask_prompt ( 'Default = 100% = DVI file magnification ') ;
  extra_mag := get_number(100)/100.0 ;
    @.What magnification?@>

@ Since the terminal is being used for both input and output, some systems
need a special routine to make sure that the user can see a prompt message
before waiting for input based on that message. (Otherwise the message may
just be sitting in a hidden buffer somewhere, and the user will have no idea
what the program is waiting for.) Here, we assume that the system-dependent
macro |@!i_reset_terminal| (defined above) will do whatever is necessary to
switch the terminal from output to input. Likewise, |@!o_rewrite_terminal|
must switch it from input to output. Note that the program assumes that the
terminal is normally in output mode, and explicitly calls these macros when it
wants input. If the system does not allow this, then |@!can_interact| should
be set false.

Here is how the program prompts for input: the argument of |ask_prompt| is the
prompt text. Because of the anomalous behaviour of |write|, this ought to work
with arguments of any length, even on versions of \PASCAL\ that only allow
fixed length strings.

@d ask_prompt(#) == if can_interact then begin
  {AKT: removed display_ln(#) ;} read_terminal ; end;

@<Lowest...@>=
  procedure read_terminal;
  var k: byte ;
  begin i_reset_terminal;
    buffer := blank ;
    if not eof(term_in ) then begin
      if eoln(term_in) then read_ln(term_in);
      k:=0;
      while not eoln(term_in) do
      begin incr(k); buffer.data[k]:=term_in^; get(term_in);
      end;
      buffer.len := k ;
      upcase(buffer) ;
    end;
    o_rewrite_terminal ;
  end;

@ @<Glob...@>=
  @!buffer: var_string ;  {for terminal input}
  @!extra_mag:real_num ;

@ The next function reads an integer from the |buffer|. It assumes a previous
call of |ask_prompt| and returns the default if the input is unrecognisable.
BODGE: this cant handle negative numbers.

@<Lowest...@>=
  function get_number(default: integer): integer;
  var k, m : integer; c:byte ;
  begin
    k := 0 ;
    repeat
      incr(k) ;
      c := zord(buffer.data[k] ) ;
    until (k > buffer.len)
        or ((c <> " " ) and (c <> "+" ) ) ;
    if (k > buffer.len) or (c < "0") or (c > "9" )
    then get_number := default
    else
    begin
      m:=0;
      while (c >="0") and (c <="9") do begin
        m:=10*m+ c -"0"; incr(k);
        c := zord(buffer.data[k] ) ;
      end;
      get_number := m ;
    end;
  end;

@ If the printer is actually a VDU, then possibly the user will want to pause
at intervals.

@<Check pause@>=
  if can_interact and do_pause and (PR_v >= next_pause) then begin
    display_ln(pause_ask);
    i_reset_terminal;
    read_ln (term_in );
    o_rewrite_terminal ;
    string_show(pause_after);
    next_pause := next_pause + pause_steps ;
  end;

@ @<Pause reset@>=
  if do_pause then
  begin next_pause := pause_steps; @<Check pause@> end;

@ @<Const...@>=
 @<Pause constants, to be defined by the printer change file@>

@ @<Glob...@>=
  next_pause: integer;
  pause_after: var_string ;

@* Interface to Operating System, 3: Input from binary files.

The main input file is the \.{DVI} file. Logically, this is just a stream of
8-bit bytes, with no record or block structure. However VMS \PASCAL\
apparently cannot handle files of this type; so I have adopted the blocking
scheme (due to D.R.Fuchs) from the VMS \.{DVItype} change file. But a lot of
the code has been rewritten. Some other operating systems use similar
blocking schemes; so this code may possibly work without much change. The
program deals with two binary file variables: |@!dvi_file| is the main input
file that we are printing, and |@!tfm_file| the current font metric file from
which character-width information is being read. Each of these has a name and
a counter, declared here; also a default name (system dependent, and so
declared previously). As an initial attempt at downloading, we declare a
|@!pxl_file|.
@^Fuchs, D.R.@>

@<Files...@>= , dvi_file, tfm_file, pxl_file

@ @<Glob...@>=
  dvi_file, tfm_file, pxl_file: byte_file ;
  dvi_name, tfm_name, pxl_name: var_string ;
  dvi_indx, tfm_indx, pxl_indx: integer ;
  font_ok: boolean ;

@ @<Types...@>=
  @!byte_block=packed array [0..block_length-1] of byte ;
  @!byte_file= packed file of byte_block;

@ This code opens the \.{DVI} file; clearly, a failure is fatal.

@<Open \.{DVI} file@>=
  {AKT: removed display_ln ('Opening DVI file ' );}
  if not open_and_ask(dvi_file, dvi_indx, dvi_name, dvi_def)
  then abort('Could not open DVI file!');
    @.Fatal: couldnt open@>@.Opening DVI file@>

@ But when we come to open a font file, we merely report a failure:

@<Open font file@>=
  font_ok := open_and_ask (tfm_file, tfm_indx, tfm_name, tfm_def) ;

@ Here is the procedure that actually opens files. It searches for a file
called |name|, supplying missing bits from the default file-specification in
|other_name|. |f_f| is the file being opened, and |f_c| is its counter.

@<Medium...@>=
  function open_and_ask
  (var f_f: byte_file; var f_c: integer; var name: var_string;
    other_name: def_name_type ) : boolean ;
  label exit ;
  var success : boolean;
  begin
    success := false;
    repeat
      success := open_binary(f_f, name, other_name) ;
      if success then f_c := 0
      else
      @<Try to get an alternative name, |return| if this fails@>
    until success ;
    exit: open_and_ask:= success ;
  end;

@ If this fails, then ask the user for another name. If the operating system
forbids this, or if the user refuses, then return |false| to indicate failure.

@<Try to get...@>=
  begin
    return;    {AKT: don't ask user for another name}
    warn ('Couldnt open file, search name was, ' );
    string_show(name) ;
    display_ln (' ') ;
    display_ln  ('default name was  ' , other_name );
    if can_interact then begin
      ask_prompt('Please type a replacement or NO to abandon search' ) ;
      name := buffer ;
      if (name.len = 2) and
      (name.data[1] = 'N') and (name.data[2] = 'O')
      then return;
    end else return;
  end ;
    @.error: couldnt open@>@.Please type...@>

@ \.{DVItype} has seven functions for reading integers from the \.{DVI} file
and two more for the \.{TFM} file. I have condensed these. In order for
these procedures to work, they must all have as parameters, both the file and
its attached counter. These macros generate the procedure calls.

@d read_end(#) == # @=)@>
@d skip(#) == skip_bytes @=(@> # @& file, # @& indx, read_end
@d get_integer(#) == read_integer @=(@> # @& file, # @& indx, read_end
@d get_byte(#) == read_byte(# @& file, # @& indx)
@d get_real(#) == read_real(# @& file, # @& indx)

@<Lowest...@>=
  function read_byte(var f_file: byte_file; var f_indx: integer) : byte;
  begin
    if eof(f_file) then
    warn('fallen off end of file' )
      @.error: fallen off...@>
    else begin
      read_byte := f_file^[f_indx] ;
      incr(f_indx);
      if f_indx =block_length then begin
        get(f_file );
        f_indx:=0;
      end;
    end;
  end ;

@#
  procedure skip_bytes(var f_file: byte_file; var f_indx: integer; n:integer);
  {discard n bytes from |f_file|}
  begin
    if n < 0 then abort('skip_bytes called with negative number');
    f_indx := f_indx + n;
    while f_indx >= block_length do
    begin
      if eof(f_file) then
      warn('fallen off end of file' )
      else get(f_file );
      f_indx := f_indx - block_length ;
    end ;
  end;
    @.error: fallen off...@> @.Fatal: skip_bytes called...@>

@ The next function reads an integer from a file. |k| specifies the type.
|abs(k)| is the number of bytes, and the integer will be signed if |k<0|.

@<Lowest...@>=
  function read_integer
    (var f_file: byte_file; var f_indx: integer; k: integer): integer;
    var a, i : byte; n: integer;
    begin n := get_byte(f );
      if (k < 0) and (n > 127) then n := n-256 ;
      for i := 1 to abs(k) - 1 do
      begin
        a := get_byte(f ) ;
        n := n*256 + a ;
      end ;
      read_integer := n ;
    end;

@ A real number is stored in the file as 2 integers, numerator first.

@<Medium...@>=
  function read_real(var f_file: byte_file; var f_indx: integer ): real_num;
  var a, b: integer;
  begin a := get_integer(f )(-4);
    b :=  get_integer(f )(-4);
    if b <= 0 then
    begin
      warn('denominator <= 0! '); read_real:= 1;
    end
    else read_real:= make_double(a)/make_double(b) ;
  end;
    @.error: denominator...@>

@* Page selection.

We have now disposed of all the code that is known to be system-dependent, so
we can resume a proper top-down description of the program. The basic method
for processing each page is that all printable characters are written onto a
structure called a `page image'. This is a list of things called `page
records'. Each page record represents one printable character, and contains
two fields giving the intended position on the page. Eventually the image will
be sorted and then copied to the |printfile|. This means that \.{Crudetype}
has to remember three sets of coordinates. In order to help to keep track of
many global variables, we use prefixes. \.{DVI} variables are prefixed with
|D_|, page image variables with |IM_|, and the printer's variables with |PR_|.
When this module starts, the \.{DVI} file should be positioned at or before a
BOP.

@<For each page...@>=
  {AKT: moved first read_BOP before repeat loop}
  if (counter[0] >= first_page) then start := true ;
     {AKT: what if counter[0] is < 0???}
  if start and (max_pages > 0 )
  then begin
    decr(max_pages);
    display('[', counter[0]:1, ']' ); {Progress report}
    Read_one_page ;
    @<Sort the page@>
    Send_page ;
    {AKT: removed @<Formfeed@>;}
  end
  else if max_pages > 0 then Skip_page
  else time_to_stop := true;
  {AKT: only call Formfeed BETWEEN pages; this requires some hackery}
  read_BOP;                           {sets max_pages to -1 if no more}
  time_to_stop := max_pages <= 0;
  if (not time_to_stop) and start then begin
     @<Formfeed@>;
  end;

@ This program only gives a small subset of the page-selection facilities of
\.{DVItype}. The most you can do is to specify the starting page and the
maximum number of pages to print. This will be controlled by these variables:

@<Glob...@>=
  start, time_to_stop: boolean;
  first_page,  max_pages: integer;
  counter: array[0..9] of integer ;

@ @<Set init...@>=
  start := false ; time_to_stop := false;
  for in_i := 0 to 9 do counter[ in_i ] := 0 ;

@ |@!D_com| is the \.{DVI} command byte, |@!D_par| its first parameter.

@<Top level...@>=
  procedure Read_one_page ;
  var D_com: byte; D_par: integer; end_page: boolean ;
  begin end_page := false;
    @<Set up an empty page image@>
    repeat
      @<Get \.{DVI} command |D_com|, and do it@>
    until end_page;
  end ;

@#procedure Skip_page ;
  var D_com: byte; D_par: integer; end_page: boolean ;
  begin
    end_page := false;
    repeat
      @<Skip \.{DVI} command, but we must process any |font_def|@>
    until end_page;
  end ;

@* Translating the device-independent file, 1: The big switch.

Refer to \.{DVItype} or to \.{TUG}boat (Vol.3, No.2) for a description of the
\.{DVI} file format. As in \.{DVItype}, we process each \.{DVI} command via a
big |case| statement. But 192 of the cases are very similar, so lets dispose
of them first.
 @.TUGboat@>

@d id_byte=2 {identifies the kind of \.{DVI} files described here}
@d move_right ==
    D_h := D_h + D_dis ;
    IM_h := IM_h + IM_dis

@<Get \.{DVI} command...@>=
  D_com := get_byte(dvi);
  if D_com < 128 then begin
    set_character(D_com); move_right ;
  end
  else if (D_com >= 171) and (D_com <= 234) then
    change_font(D_com - 171)
  else

@ @<Skip \.{DVI} command...@>=
  D_com := get_byte(dvi);
  if (D_com < 128)
  or ((D_com <= 234) and (D_com >= 171))
  then do_nothing
  else

@ Now we come to the |case| statement proper. This section of the program is
long and complicated, and I have tried to clean it up. Some commands want an
unsigned parameter, called |D_par|, to be read from the file. We use
|four_cases| for those. Others want a signed parameter; they are all
movements. We use |move_cases| for those.

@d four_case_end(#) == # ; end
@d four_cases(#)==
  #,#+1,#+2,#+3: begin D_par := get_integer(dvi)( D_com - # + 1 );
    four_case_end
@d move_cases(#)==
  #,#+1,#+2,#+3: begin D_par := get_integer(dvi)( # - D_com - 1 );
    four_case_end
@#

@<Get \.{DVI} command...@>=
case D_com of
  four_cases(128)    (set_character(D_par); move_right );
  132:                begin set_rule; move_right ; end;
  four_cases(133)    (set_character(D_par) );
  137:                set_rule ;
  138:                do_nothing ;
@#
  140:                end_page := true ;
  141:                push;
  142:                pop;
  move_cases(143)    (D_h := D_h+D_par);
  147:{W0}            D_h := D_h+D_w ;
  move_cases(148)    (D_w := D_par; D_h := D_h+D_w );
  152:{X0}            D_h := D_h+D_x ;
  move_cases(153)    (D_x := D_par; D_h := D_h+D_x );
  move_cases(157)    (move_down(D_par));
  161:{Y0}            move_down(D_y);
  move_cases(162)    (D_y := D_par; move_down(D_y) );
  166:{Z0}            move_down(D_z);
  move_cases(167)    (D_z := D_par; move_down(D_z) );
@#
  four_cases(235)    (change_font(D_par) );
  four_cases(243)    (define_font(D_par) );
@#
  @<Fourteen illegal cases: print suitable error messages@>
end ;

@ When skipping a page, we must throw away parameters instead of using them.

@d four_throw(#) ==
  #,#+1,#+2,#+3: skip(dvi)(D_com - # + 1 )
@#

@<Skip \.{DVI} command...@>=
case D_com of
  four_throw(128);
  132, 137: skip(dvi)(8); {sizes of a rule}
  four_throw(133);
  138:                ;
  140:                end_page := true ;
  141,142:            ;
  four_throw(143);
  147:                ;
  four_throw(148);
  152:                ;
  four_throw(153);
  four_throw(157);
  161:                ;
  four_throw(162);
  166:                ;
  four_throw(167);
@#
  four_throw(235);
  four_cases(243)    (define_font(D_par) );
@#
@<Fourteen illegal...@>
end ;

@ Finally, there are 14 illegal values of |D_com| that generate various errors.

@<Fourteen illegal...@>=
  four_cases(239)
    ({AKT: removed warn('ignoring \special') ;} skip(dvi)(D_par) );
  139, 247, 248, 249:
    bad_dvi('byte: ', D_com:1 , ' out of context inside page' ) ;
  250,251,252,253,254,255:
    bad_dvi('Illegal command byte, ', D_com ) ;
      @.error: cant do xxx@>
      @.Fatal: Bad DVI file@>

@* Translating the device-independent file, 2: Paging and the stack.

The definition of \.{DVI} files refers to six registers, (|D_h, D_v, D_w,
D_x, D_y, D_z|), which hold integer values in \.{DVI} units. We shall need
additional registers in order to calculate a rounded position. From time to
time, we save the current values of these on a stack, represented by the
following arrays.

@d max_stack = 100 {\.{DVI} files shouldn't |push| beyond this depth}

@<Glob...@>=
  D_h,D_v,D_w,D_x,D_y,D_z : integer;            {current \.{DVI} state values}
  D_h_stack, D_v_stack, D_w_stack, D_x_stack, D_y_stack, D_z_stack:
    array [0..max_stack+2] of integer; {pushed down values }
  @!stack_ht: 0..max_stack;    {current stack depth}
  just_pushed: boolean;

@ @<Set up an empty page image@>=
  D_h := 0 ; D_v := 0 ;
  D_w := 0 ; D_x := 0 ;
  D_y := 0 ; D_z := 0 ;
  stack_ht := 0 ;
  rail_base := 0 ;
  just_pushed := false ;

@ Here is how \.{DVI}type manipulates the stack: The first |push| on a page
fills the zeroth place on the stack and sets |stack_ht| = 1. So the used
places are numbered |0..stack_ht- 1|. Now |push| and |pop| do the obvious
things.

@<Lowest...@>=
  procedure push;
  var x: real_num ;
  begin if stack_ht=max_stack then
    warn('Capacity exceeded (stack size=', max_stack:1,')')
    else begin
      D_h_stack[stack_ht]:=D_h; D_v_stack[stack_ht]:=D_v;
      D_w_stack[stack_ht]:=D_w; D_x_stack[stack_ht]:=D_x;
      D_y_stack[stack_ht]:=D_y; D_z_stack[stack_ht]:=D_z;
      @<Some adjustments are needed here for rounding@>
      incr(stack_ht); just_pushed := true ;
    end;
  end;
    @.error: Capacity exceeded @>

@# procedure pop;
  begin if stack_ht=0 then warn('POP illegal at level zero')
    else  begin
      decr(stack_ht);
      D_h:=D_h_stack[stack_ht]; D_v:=D_v_stack[stack_ht];
      D_w:=D_w_stack[stack_ht]; D_x:=D_x_stack[stack_ht];
      D_y:=D_y_stack[stack_ht]; D_z:=D_z_stack[stack_ht];
      IM_h := IM_h_stack[stack_ht];IM_v := IM_v_stack[stack_ht];
      @<Set |rail_base|@>
    end;
  end;
    @.error: POP illegal...@>

@ This procedure gets called when we expect to read a new page. It looks for
the next |BOP|; if it finds the postamble instead, it sets |max_pages < 0| as
a signal.

@d POST = 248
@d NOP = 138
@d BOP = 139

@<Top level...@>=
  procedure read_BOP;
  var k: byte ; D_par:integer ;
  begin
    repeat k:= get_byte(dvi);
      if (k>= 243)and(k <= 246 ) then {a |font_def|}
      begin D_par:=get_integer(dvi) (k-242 ); define_font(D_par); k:=NOP;
      end;
    until k<>NOP;
    if k=POST then
    max_pages := -1
    else if k<>BOP then bad_dvi('byte is not BOP')
      @.Fatal: Bad DVI file@>
    else begin
      for k:=0 to 9 do counter[k]:= get_integer(dvi)(-4);
      skip(dvi)(4);
    end;
  end;

@ A \.{DVI}-reading program that reads the postamble first need not look at the
preamble; but \.{Crudetype} reads the \.{DVI} file sequentially.

@d PRE=247 {preamble}

@<Read \.{DVI} preamble@>=
  bbb:= get_byte(dvi); {fetch the first byte}
  if bbb<>PRE then bad_dvi('First byte isn''t start of preamble!');
    @.Fatal: Bad DVI file@>
  bbb:= get_byte(dvi); {fetch the identification byte}
  if bbb<>id_byte then
  warn('identification byte should be ',id_byte:1,', it is actually', bbb:1 );
    @.error: identification...@>
  @<Compute the conversion factors@>;
  bbb:= get_byte(dvi); {fetch the length of the introductory comment}
  {AKT: removed display(' ');}
  for nnn := 1 to bbb do
      {AKT: was display(zchr(get_byte(dvi)));}
      bbb := get_byte(dvi);
  {AKT: removed display_ln(' ');}

@ The conversion factor |h_conv| is figured as follows: There are exactly
|n/d| decimicrons per \.{DVI} unit and 254000 decimicrons per inch, and
|h_resolution| |h_steps| per inch.

@<Glob...@>=
  dvi_factor, h_conv, v_conv, magnification : real_num;
  nnn:integer; {general purpose register}
  bbb: byte ;

@ @<Compute the conversion factors@>=
  dvi_factor := get_real(dvi)/254000.0 ;
  magnification :=  get_integer(dvi)(4) / 1000 ;
  dvi_factor := dvi_factor * magnification ;
    {This converts \.{DVI} units to inches (on an ideal device) }
  h_conv:= dvi_factor * h_resolution * h_fudge * extra_mag;
  v_conv:= dvi_factor * v_resolution * v_fudge * extra_mag ;

@* Translating the device-independent file, 3: Setting a Rule.

|D_p| is the height and |D_q| is the width. A rule has to be assembled from
the available characters. First: is the rule to be set at all? Second: is it
horizontal or vertical? (Because of the limited name lengths, we call them
|Post| and |Rail|.) The test applied here is quite arbitrary.

@<Medium...@>=
  procedure set_rule;
  var D_p,D_q: integer;
  begin
    D_p:=get_integer(dvi) (-4);
    D_q:=get_integer(dvi)(-4);
    if (D_p<=0)or(D_q<=0) then
      {an invisible rule! Dont ask me why \TeX\ wants to do this}
    else if D_p*v_conv <= post_height/2
    then do_rail(D_p, D_q)
    else do_post(D_p, D_q);
  end;

@ Setting a vertical rule is simple: we just fill all the space with the
relevant character.

@<Lowest...@>=
  procedure do_post(D_rul_ht, D_rul_width: integer);
  var vn, vi, hn, hi, post_v, rule_hp : integer;
  rule_cod: code_object ;
  begin
    @<|Post| set sizes@>;
    for vi := vn - 1 downto 0 do
    begin
      post_v := IM_v - vi * post_height ;
      for hi := 1 to hn do
      begin
        rule_hp := IM_h + (hi - 1) * post_width ;
        do_set_char(post_v, rule_hp, rule_cod);
      end;
    end;
  end;

@ Note that whereas \.{DVItype} rounds all sizes up, \.{Crudetype} rounds to
nearest integer.  This seems more likely to work on a crude resolution.
But we force the rounded size to be |>= 1| .

@<|Post| set...@>=
  round_IM_h ( 0);
  hn := round(D_rul_width * h_conv / post_width );
  vn := round(D_rul_ht * v_conv / post_height);
  if hn <= 0 then hn := 1;
  if vn <= 0 then vn := 1;
  rule_cod := post_char;

@ A horizontal rule is more complicated, as there is then a selection of
characters. This matters if the printer has only a very coarse vertical
positioning. For example, a line printer has only minus and underscore, but a
VT-100 has 5 bars at different heights. |@!rail_types| should be set to the
number of different horizontal bars that the printer can draw within one
|v_step|. We measure the vertical position of a rule in |rail_steps|,
which are smaller than |v_steps| in the same ratio.

@<Glob...@>=
  rail_chars : packed array [1..rail_types] of code_object ;
    {Number from bottom of page up; so no. 1 might be an underscore}
  rail_base : integer ;
    {Position of bottom edge of a  rule in |rail_steps|}
  post_char : code_object ;

@ @<Const...@>=
  @<Rule setting constants@>
  {Printer-dependent, so they must go at the end of the file}

@ @<Lowest...@>=
  procedure do_rail(D_rul_ht, D_rul_width: integer);
  var vn, vi, hn, hi,
  rail_v,  {Current position in |rail_steps|}
  char_vp,  {Position in |v_steps| where a rule char will be set}
  rule_hp: integer;
  rule_cod: code_object ;
  char_i : 1..rail_types ; {indicates which character to be used}
  begin
    @<|Rail| set sizes@>
    for vi := vn-1 downto  0 do begin
      rail_v := rail_base - vi ;

@ Now to assign |char_i| and |char_vp|. The easiest way is to consider a simple
example. Suppose |rail_types = 5| and |rail_v = 50|. This addresses the
underscore at the bottom edge of a text character at |10 v_steps|.
So |char_i| wants to be 1 and |char_vp| 10. So...

@<Lowest...@>=
  char_vp := ((rail_v - 1) div rail_types ) + 1 ;
  char_i :=  rail_types - ((rail_v - 1) mod rail_types ) ;
  rule_cod := rail_chars [ char_i] ;
  for hi := 1 to hn do begin
    rule_hp := IM_h + (hi-1) * rail_width ;
    do_set_char(char_vp, rule_hp, rule_cod) ;
  end;
end;
end;

@ @<|Rail| set...@>=
  round_IM_h ( 0);
  hn := round(D_rul_width * h_conv/ rail_width);
  vn := round(D_rul_ht * v_conv * rail_types/ rail_height );
  if hn <= 0 then hn := 1;
  if vn <= 0 then vn := 1;

@ Now consider how to set |rail_base|. Horizontal rules are mostly used for
underlining text, and then they should be aligned with the underscore
character on the same line of text. So normally, we just do the following. The
exception occurs when the \.{DVI} file does an explicit vertical move.

@<Set |rail_base|@>=
  rail_base := IM_v * rail_types ;

@* Translating the device-independent file, 4: Changing and defining Fonts.

The following tables describe all the \TeX\ fonts that \.{Crudetype}
currently knows about.

@<Glob...@>=
  nf: D_font_ptr ;
    {The number of fonts so far defined. These will be numbered |0..nf-1| }
  @!font_num,         {external font numbers}
  @!font_space,       {boundary between ``small'' and ``large'' spaces}
  @!scheme,           {pointer to coding scheme}
  @!first_ch,         {First character in the font}
  @!last_ch:          {and last}
    array [D_font_ptr] of integer;
  D_width: array[D_font_ptr, D_char_ptr ] of integer ;
    {character widths, as given in \.{TFM} file, should be in \.{DVI} units}
  @!D_check,     {the font checksum must be global for HPGF}
  thin_space, D_font, cur_scheme: integer ;     {The current values}

@ @<Type...@>=
  D_font_ptr = 0..max_D_fonts;
  D_char_ptr = 0..max_D_char;

@ The size of the tables can be altered at compile time to extend or reduce
\.{Crudetype}'s capacity.

@<Constants...@>=
  @!max_D_fonts=100; {maximum number of distinct fonts per \.{DVI} file}
  @!max_D_char =255; {AKT: was 127 but we want to handle PostScript fonts}

@ Initially, all these tables are blank.

@<Set init...@>=
  nf:=0;
  for in_i := 0 to max_D_fonts do
  begin
    font_num[in_i ] := 0 ;
    scheme[in_i ] := 0 ;
    first_ch[in_i ] := 0 ;
    last_ch[in_i ] := 0 ;
    font_space[in_i]:= 0 ;
  end;

@ @<Set up an empty page image@>=
  D_font := nf  ;
  cur_scheme := 0 ;

@ @<Medium...@>=
  procedure change_font (D_new: integer);
  begin
    D_font := 0 ;
    font_num[nf]:=D_new;
    while font_num[D_font]<>D_new do incr(D_font);
    if D_font = nf then
    warn('Undefined font called for, number ', D_new:1 );
      @.error: Undefined font@>
    cur_scheme := scheme[D_font] ;
    thin_space := font_space[D_font] ;
  end;

@ The following procedure is called whenever we read a |font_def| command from
the \.{DVI} file. In general, any error while defining a font causes a jump to
label |bad_font|, leaving the new font undefined.

@d bad_font = 9999
@d good_font = 9998
@d font_error(#) == begin
  warn(#); display_ln('font number ', D_new:1, ' cannot be loaded') ;
  goto bad_font ;
end

@<Medium...@>=
  procedure define_font (D_new:integer );
  label bad_font , good_font ;
  var @<|font_def| vars@>
  begin
    @<Read the font parameters from the \.{DVI} file,
      calculate scaling factors@>;
    @<Try to load the new font, unless there are problems@>;
    good_font:
    @<Final checks; various mild errors which often are symptoms of bugs@>
    incr(nf) ; {the new font is officially present}
    bad_font: if font_ok then close_binary(tfm_file);
  end;

@ First we read the parameters from the \.{DVI} file. Whatever errors are
found, we must try to do this, or we lose place in the file.

@<|font_def| vars@>=
  scale_size, design_size, k, f : integer;
  dir_len,      {length of the area/directory spec}
  nam_len:byte; {length of the font name proper}
  font_mag: real_num;

@ @<Read the font parameters...@>=
  @!D_check := get_integer(dvi)(-4) ;
  scale_size:= get_integer(dvi)( -4) ;
  design_size:= get_integer(dvi)(-4) ;
  dir_len:= get_integer(dvi)(1) ;
  nam_len:= get_integer(dvi)(1) ;
  nam_len := nam_len + dir_len ;
  if nam_len = 0 then
  font_error('null font name! ')
    @.error: null font name@>
  else if nam_len >= string_length then
  font_error('too-long font name! length =  ', nam_len:1 ) ;
    @.error: too-long font name@>
  tfm_name := blank ;
  for k:=1 to nam_len do begin
    tfm_name.data[k] := zchr(get_byte(dvi)) ;
  end;
  tfm_name.len := nam_len ;
  upcase(tfm_name) ;
  {AKT: removed
  display_ln(' ');
  string_show(tfm_name);
  display(' ');
  }

@ Next, check that the sizes are reasonable:

@<Read the font parameters...@>=
  if (scale_size<=0)or(scale_size>=@'1000000000) then
  font_error('--- bad scale (',scale_size:1,')!')
    @.error: bad scale@>
  else if (design_size<=0)or(design_size>=@'1000000000) then
  font_error('--- bad design size (',design_size:1,')!') ;
    @.error: bad design size@>
  font_mag := scale_size/design_size ;
  if (font_mag > 1000) or (font_mag < 0.001) then
  warn('thats a very unusual font magnification!!! ', font_mag) ;
    @.error: unusual font mag...@>
  if nf=max_D_fonts then
  abort('Crudetype capacity exceeded (max fonts=', max_D_fonts:1,')!');
    @.Fatal: Capacity exceeded... @>
  font_num[nf]:=D_new; f:=0;
  while font_num[f]<>D_new do incr(f);
  if f<nf then font_error('---this font was already defined!');
    @.error: font already defined@>
  font_space[nf] := scale_size div 6 ; {a `thin space' }

@* Loading the font file.

See \.{TFTOPL} or \TeX 82 for details of the \.{TFM} file format. The
description given in \.{TUGboat} (Vol.2, no. 1) is apparently no longer
accurate. The only difference that I have seen is that all words of the font
header array after the first 2 are now apparently regarded as optional.
@.TFTOPL@> @.TeX82@> @.TUGboat@>

@<Try to load...@>=
  @<Open font file@>
  if not font_ok then
  font_error('---TFM file can''t be opened!');
    @.error: TFM file cant be opened@>
  @<Read past the header data, leave the file pointer just after the header@>
  @<Read the character-width indices@>
  @<Read the widths, copy them into the font array@>

@ @<|font_def| vars@>=
  @!TFM_check,
  @!lh, {length of the header data, in four-byte words}
  @!nw:integer; {number of words in the width table}

@ @<Read past the header...@>=
  skip(tfm)(2);                     lh:= get_integer(tfm)(2);
  first_ch[nf]:=get_integer(tfm)(2);   last_ch[nf]:=get_integer(tfm)(2);
  if (last_ch[nf]<first_ch[nf]) or (last_ch[nf] > max_D_char) then
  font_error(
    'Illegal values for first_char and/or last_char, first_char = ',
      first_ch[nf]:1 , ' last_char = ', last_ch[nf]:1 );
      @.error: Illegal value@>
  nw:=get_integer(tfm)(2);
  if (nw=0)or(nw>256) then
  font_error('Illegal value for nw, nw= ', nw );
    @.error: Illegal value@>
  skip(tfm)(14);
  TFM_check := get_integer(tfm)(-4);
  skip(tfm)(4);
  @<Get coding scheme and re-align file, then see if the printer knows it@>

@ The header contains |4*lh| bytes, of which 8 have been read so far. If it
conforms to the \.{TUGboat} format, then the next byte (|@!ck|, say) is the
number of bytes in the coding scheme name. So, first we must try to see if a
scheme is present; if so, then we will read |ck+1| bytes and chuck the rest.
If no coding scheme is present, we simply skip the rest of the header.
Internally, scheme names are represented by |var_string|s.

@<Get cod...@>=
  TFM_scheme := blank ;
  if lh < 2 then font_error( ' Header must have at least 2 words')
  else if lh = 2 then do_nothing
  else begin
    ck := get_byte(tfm);
    if ( ck >= 40 ) or ( ck > 4*lh - 9) then
    skip(tfm)(4*lh - 9)
    {there is something here, but not a coding scheme}
    else begin
      TFM_scheme.len := ck ;
      for j := 1 to ck do
      TFM_scheme.data[j] := zchr(get_byte(tfm)) ;
      skip(tfm)(4*lh - ck - 9);
      upcase(TFM_scheme) ;
    end;
  end;

@ @<|font_def| vars@>=
  j , ck : byte ;
  @!coding_scheme, TFM_scheme: var_string ; {coding scheme of current font}

@ Now we can start reading the character widths.

@<|font_def| vars@>=
  @!in_width:array[byte] of integer; {\.{TFM} width data in \.{DVI} units}
  @!wid_ptr: array[byte] of byte ; {pointers into |in_width|}
  b3,b2,b1,b0: byte;       {bytes from \.{TFM} file}
  @!alpha,@!beta, @!z :integer;

@ @< Read the character-width indices...@>=
  for k:=first_ch[nf] to last_ch[nf] do
  begin wid_ptr[k] := get_byte(tfm); skip(tfm)(3);
    if wid_ptr[k] > nw then font_error('impossible width ' , wid_ptr[k]);
  end;
    @.error: impossible width @>

@ Here is the width computation. This code is copied from \.{DVItype}. See
that program for an explanation.

@<Read the font parameters...@>=
  z := scale_size ;
  alpha:=16*z; beta:=16;
  while z>=@'40000000 do
  begin z:=z div 2; beta:=beta div 2;
  end;

@ @<Read the widths...@>=
  for k:=0 to nw-1 do
  begin
    b0 := get_byte(tfm); b1 := get_byte(tfm);
    b2 := get_byte(tfm); b3 := get_byte(tfm);
    in_width[k]:=
    (((((b3*z)div@'400)+(b2*z))div@'400)+(b1*z))div beta;
    if  b0 = 255 then in_width[k]:=in_width[k]-alpha
    else if b0 <> 0 then
    font_error('Out-of-bounds value for b0') ;
      @.error: font: Out-of-bounds |b0|@>
  end ;

@ Rounding widths. This bit of \.{DVItype} is changed, because \.{Crudetype}
has to calculate rounded positions by a completely different method.

@<Read the widths...@>=
  if in_width[0]<>0 then font_error('the first width should be zero ');
    @.error: first width...@>
  for k:= first_ch[nf] to last_ch[nf] do
  D_width[nf, k] := in_width[ wid_ptr[k]] ;

@ Then there are various erroneous states that do not necessarily show that
the font is corrupt, but may indicate bugs in the program. In principle, a
character might have negative width, but I do not believe it.

@d bad_char = -32766  {Indicates an unprintable character}
@d foot == 50000000    {about a foot}

@<Final checks...@>=
  for k:= first_ch[nf] to last_ch[nf] do
  if (D_width[nf, k] < 0) or (D_width[nf, k] > foot) then begin
    warn('Way-out width = ', D_width[nf,k]:1,
      'DVI units, character number ', k:1 );
    codes[ scheme[nf], k].breadth := bad_char ;
  end;
  if (D_check<>0)and(TFM_check<>0)and(D_check<>TFM_check) then
  begin warn('check sums do not agree!');
      @.error: check sums...@>
    display_ln('DVI check was: ', D_check, ' TFM check was: ', TFM_check);
    display('   ');
  end;
  {AKT: removed display_ln('---loaded at size ',scale_size:1,' DVI units');}
  font_mag := (font_mag -1) * 100.0 ;
  {AKT: removed
  if abs(font_mag) > 1 then
  begin display_ln(' ');
    display_ln(' (this font is magnified ', round(font_mag):1,'%)');
  end;
  }
    @.this font is magnified@>@.error: Way-out width@>

@* Coding schemes.

In this section we describe the mapping from characters in \TeX\ fonts to
characters in the printer's fonts (which are presumably much fewer). All
characters on a crude printer are the same size. We therefore need one piece
of data, not for each \TeX\ font, but for each coding scheme. The mapping is
defined in an array called |codes|. For each character |c| in a \TeX\ font
whose coding scheme has internal number |s|, |codes[s,c]| describes the
corresponding printer character. Also |known_schemes[s]| is a character
string which usually contains the coding scheme of that \TeX\ font.
|max_codes| is the number of coding schemes the program knows about. First,
define that structure:

@ @<Glob...@>=
  @!known_schemes: array[code_ptr] of var_string ;
  @!codes: array[code_ptr, D_char_ptr] of code_object;
  no_char: code_object ;

@ @<Types...@>=
  code_object = packed record
    breadth: i_word ;
    case boolean of
      true: (IM_font: byte ; IM_char: byte );
        {Printers font and character}
      false: (multi: i_word) ;
  end;
  @! code_ptr = 0..max_codes;
    {0 is a coding scheme the printer doesnt know about}

@ Initially, all these tables are blank. If |c| is a |code_object|, then
|c.breadth| will usually be its printed width in |h_steps|. |c.breadth =
bad_char| indicates that the character is unprintable. |bad_char| can be any
large negative value. Other negative values of |@!breadth| indicate other
types of peculiar characters.

@d down_loaded = -32765

@<Set init...@>=
  no_char.breadth := bad_char ;
  no_char.IM_font := 0 ;
  no_char.IM_char := 0 ;
  for in_i := 0 to max_codes do
  for in_j := 0 to max_D_char do begin
    codes[in_i, in_j] := no_char ;
  end;

@ So when a font is read in, we try to assign the right value to its |scheme|.
If the printer is not absolutely crude, then it might have italic or bold
fonts. Then we might want a coding scheme to correspond to a single \TeX\
font. So first we look at the actual font name and see if that matches any
of the |known_schemes|. But if the printer is |fixed_width|, then all fonts
of the same face are the same size, so we drop the font size digits off the
end of the name.

@<Get cod...@>=
  k := tfm_name.len ;
  if fixed_width then
  while (zord(tfm_name.data[k]) >= "0" ) and
    (zord(tfm_name.data[k]) <= "9" ) do
  decr(k) ;
  coding_scheme:= tfm_name;
  chop_length(coding_scheme, k) ;
  j := max_codes ;
  while (j > 0) and (coding_scheme.data <> known_schemes[j].data ) do decr(j);
  scheme[nf] := j ;
  if j = 0 then

@ If the font name is not in |known_schemes|, then we try again with the
scheme given in the \.{TFM} file. If that fails, then try if we can download
the font. If that fails, then the font is deemed to be unprintable, so we do
not load it.

@<Get cod...@>=
  begin
    j := max_codes ;
    while (j > 0) and (TFM_scheme.data <> known_schemes[j].data ) do decr(j);
    scheme[nf] := j ;
  end;
  if (j = 0) and can_dl_font then
  @<Download a whole font@>
  else if j = 0 then begin
    scheme[nf] := 9 ;               {AKT: handle PostScript font }
    {AKT: was
    display ('Scheme is: ') ;
    string_show(TFM_scheme) ;
    font_error(' That coding scheme is unknown' );
    AKT}
  end;
    @.error: unknown coding scheme @>

@ This procedure sets a character. The character to be set is number |@!c_num|
in the current font. I have deleted the bit of \.{DVItype} that deals with
oriental fonts, as I dont believe that crude printers can support them.

@<Medium...@>=
  procedure set_character(c_num: integer );
  var cod: code_object;
  d_i, d_j : integer; {Used for downloading}
  begin
    if cur_scheme = 0 then
    else if (c_num < first_ch[D_font] ) or (c_num > last_ch[D_font] )
    then begin
      warn('character ',c_num:1,' invalid in font number ',
        font_num[ D_font]:1 );
        @.error: character invalid...@>
    end
    else begin
      cod := codes[ cur_scheme, c_num];
      if cod.breadth <> bad_char then begin
        round_IM_h( c_num) ;
        if cod.breadth = down_loaded then
        @<Enter a download request for |cod| and adjust its |breadth|@> ;
        do_set_char(IM_v, IM_h, cod ) ;
        @<Do messy things to adjust the positions |D_h|, |IM_h|, etc@>;
      end;
    end;
  end;

@ @<Forw...@>=
  procedure do_set_char(Set_v, Set_h: i_word; cod: code_object ); forward;

@ @<Lowest...@>=
  procedure do_set_char ;
  var k_i, k_k, temp_v, temp_h: i_word ;
  m_c: code_object ;
  k_ptr: 1..max_ligs;
  begin
    if cod.breadth >= 0 then begin
      @<Check the position@>
      @<Add the record to the page image@>
    end
    else if cod.breadth = bad_char then do_nothing
    else @<Set multi-character command@> ;
  end;

@* Multiple-character commands.

Several crude printers (e.g. daisy-wheels) have only a limited set of
characters, which cannot be extended. Sometimes you can generate more
characters by overstriking. \.{Crudetype} can be programmed to do this, by
placing suitable entries into a table called |ligatures|. The name is chosen
by analogy with the |lig_kern| programs in \.{TFM} files, but the data is
completely different. When one \TeX\ character maps onto several printer
characters, we call the image a `multi-character' command.

@<Const...@>=
  max_ligs = 10000 ;

@ @<Glob...@>=
  ligatures : array[1..max_ligs] of lig_thing;
  top_of_ligs: 0..max_ligs ; {highest used point in |ligatures|}

@ @<Types...@>=
  trio = 1..3 ;
  lig_thing = packed record
    case trio of
      1: (v_move: i_word ;
        h_move: i_word) ;
      2: (code: code_object) ;
      3: (num : i_word ;
        guard : i_word) ;
  end;

@ @<Set init...@>=
  top_of_ligs := 0;
  for in_i := 1 to max_ligs do ligatures[ in_i].code := no_char ;

@ The |code_object| addresses a multiple character when its |breadth| is
negative, and not one of the special classes defined above. It must then be
the |false| variant, and its |multi| field (which must be |>0|) points to the
corresponding entry in |ligatures|. Suppose that field is |c| . Then
|ligatures[c]| is the last entry of a string of items that defines the
replacement text of the |code|. It should be of the third variant; The |num|
field of this entry counts the number of characters that |code| expands into.
The |guard| field is an arbitrary impossible value called |sentry| to give a
check on the data in |ligatures| .

@d sentry = -32767

@<Set multi...@>=
  begin
    if (cod.multi <= 0) or (cod.multi > top_of_ligs) then
    warn('Illegal value of char in multi-character command')
      @.error: Illegal value@>
    else begin
      k_ptr := cod.multi ;
      if ligatures[k_ptr].guard <> sentry then
      warn('Sentry not found in Kerns ' ) ;
        @.error: Sentry ...@>
      k_i := ligatures[k_ptr].num ;
      k_ptr := k_ptr - 2*k_i ;
      if (k_i <= 0) or (k_ptr < 0 ) then
      warn('Illegal value of k_i in multi_character command');
        @.error: Illegal value@>
      for  k_k := 1 to k_i do
      @<Get that character and write it @>;
    end;
  end

@ Each character of a multi-character command needs 2 entries in |ligatures|.
The first defines the position, the second defines the character. |v_move| and
|h_move| are relative to the current (rounded) position |Set_v, Set_h| and use
the same units. A multi-character command can call another one recursively.

@<Get that character ...@>=
  begin
    temp_v := Set_v + ligatures[k_ptr].v_move ;
    temp_h := Set_h + ligatures[k_ptr].h_move ;
    incr(k_ptr);
    m_c := ligatures[k_ptr].code ;
    do_set_char(temp_v, temp_h, m_c ) ;
    incr(k_ptr);
  end;

@* Getting data into the |codes| array.

This is clearly a very long and error-prone job, so the next procedures are put
in to reduce this. First suppose that: in the \TeX\ coding scheme with number
|s|, a run of |length| characters starting from |start| maps onto a run of
consecutive characters in printer font |PR_font|, starting at |PR_first|. This
procedure will enter the whole run at one go.

@<Lowest...@>=
  procedure alphabet
  (start, length: byte; s: code_ptr ; PR_font, PR_first : byte );
  var i:integer; ccc:code_object;
  begin @<Check alphabet data@>;
    ccc.IM_font := PR_font ;
    ccc.breadth := char_width ;
    for i := 0 to length-1 do begin
      ccc.IM_char := PR_first +i;
      codes[s, start+i] := ccc ;
    end; end;

@ @<Check alph...@>=
  if (s < 1) then abort('alphabet: scheme < 1 ')
  else if (s > max_codes) then abort('alphabet: scheme too large')
  else if (PR_first < 0) then abort('alphabet: negative first')
  else if (start < 0) then abort('alphabet:  negative start')
  else if (length < 0) then abort('alphabet: negative length')
  else if (start + length -1 > max_D_char) then abort('alphabet: overflow')
    @.Fatal: alphabet...@>

@ Clearly, |alphabet| will only cover a very small part of the problem.  The
next procedure enters data into a subset of the |codes| array corresponding to
a single row of a \TeX\ font. In the standard font tables, row number |m| is
the subrange |8*m..8*m+7| of a font. It is hoped that when the calls of
procedure |row| are written out in a program, the result will be (just about)
legible, whereas a flood of statements like  \begintt
         codes[i,j].IM_font := 121; \endtt
is certainly not legible.

The parameters are as follows. |@!row_spec| specifies what characters are to
go into the row. |@!scheme | is the number assigned to the \TeX\ coding scheme
within the program. |@!row_num | is the number of the row in that scheme
(starting from 0). |@!first_font|  is the initial printer font.

@<Top...@>=
  procedure row
    (row_spec: row_str; scheme, row_num: integer; first_font: i_word );
  var n :integer;  codd: code_object;
  begin
    incr(row_count);
    row_pt := 1 ;
    row_font := first_font ;
    row_string := row_spec;
    for n := 8*row_num to  8*row_num + 7 do begin
      row_char (0 ,codd);
      if codd.breadth = bad_char then do_nothing
      else codes[ scheme, n ] := codd ;
    end;
  end;

@ \.{TANGLE} imposes a limit of 69 on the length of quoted strings. This is a
considerable nuisance, as we could make the |row_spec| strings look much better
if they could be longer.

@<Const...@>=
  row_length = 69 ;

@ @<Types...@>=
  row_str = packed array [1..row_length] of char ;

@ @<Glob...@>=
  row_pt: integer;  {Points to next char from |row_spec|}
  row_font: i_word; {printer font being addressed during the |row| procedure}
  row_string: row_str ;
  row_count: integer ;

@ @<Set init...@>=
  row_count:= 0 ;

@ In order to help debugging, error messages will print |row_string| and a
pointer. The diagnostics of |row| are known to be very poor; I have not
bothered to fix them because up to now they have been adequate, and they are
really meant for the installer rather than the end user.

@d row_warn(#) == begin
  display_ln(row_string ) ;
  display_ln('^' : row_pt-1 ) ;
  warn('Row: ', #);
  return;
end

@ The overall format of the |row_spec| is a set of 8 character specifiers
separated by one or more spaces. The procedure |row_char| reads one character
specifier from the |row_string|, and constructs the specified |code_object|.
Logically, |row_char| should be a function and return that |code_object| as
its value. \PASCAL\ does not permit this. So we assemble the result in the
variable parameter |value|.

@<Medium...@>=
  procedure row_char(context: integer; var value: code_object);
  label exit ;
  const @<|Row_char| constants@>
  var c :byte; @<Row locals@>
  begin
    value.breadth := char_width ; {default}
    value.IM_font := row_font ;    {default font}
    c := row_get ;
    if ( context = 0) and ( c <> " " ) and ( c <> "Z") then
      row_warn('Character specifiers must start with at least one space') ;
    while ( c = " " ) do c := row_get ;
    @<Escape sequences in the |row_spec| @>
    else value.IM_char := c ;
  exit: end;
    @.error: Row: Character spec...@>

@ There are several escape sequences that need to go into the |rowstring|.
Since all the PLAIN.TEX coding schemes (except the math extension one) have
the upper case Roman characters in their ASCII positions, these characters
will surely be inserted into |codes| by the |alphabet| procedure. So they are
available as flag characters. But the brackets are also used as flags, as they
are so much more perspicuous than anything else. Here is a list of the
characters currently used as escapes: \begintt

A C D E F K L M N S Q U W Z \endtt

This list should be updated if other escapes are added .
@.Escape sequences@>@.ASCII@>

@ Some characters, called `bad', have most undesirable effects when used in
\.{WEB} strings. So the following upper-case letters stand for them. The
actual characters may not be used, so they generate errors.

@<Esc...@>=
  if c = "A" then value.IM_char := 64   {at sign}
  else if c = "S" then value.IM_char := 32   {a space}
  else if c = "Q" then value.IM_char := 39   { a single quote char}
  else if c = "W" then value.IM_char := 34   { a double quote char }
  else if c = "E" then value.IM_char := 127  { a delete char }
  else if (c = "'") or (c = """") or (c = "@@") or (c = 127)
    then row_warn( 'Bad character---Rejected' )
  else if (c = " ")
    then row_warn('space found out of context')
      @.Error: Row: Bad character@>@.error: Row: space found...@>

@ Then the  `Z' escape is provided to generate a do-nothing  code. This would
be used if a previous call (say, of |alphabet|) had left a row partly
incorrect. Then you might issue a call of |row| to change that row. Typing `Z'
at the positions occupied by correct characters would leave them alone.

@<Esc...@>=
  else if c = "Z" then value.breadth :=  bad_char

@ Since many letters and brackets are used as escapes, the `L' escape is
needed to enable them to be used Literally. `LL' generates `L'.

@<Esc...@>=
  else if c = "L" then value.IM_char := row_get

@ In order to address printer characters in the range 0..32, where ASCII has
no graphics, here is a Control escape. This simply reads the next character
from the |row_spec| and reduces it modulo 32. It is best to use the lower case
alphabet (the range 95..126) as this avoids all the `bad' characters (and
their escapes). So control-A should be typed `Ca' , not `CA' .

Then the Meta escape addresses meta-characters, i.e. those in the range
128..255. We cannot just read a character and add 128, as we might want to
Mutate the ASCII controls, or the `bad' characters. So `M' must read a complete
|code_object| (respecting the escapes given above) and add 128 to its |IM_char|
field. So we must say `MS' for `meta-space' = 160 , and  `MLS' for `meta-S'
= 211 .
@.ASCII@>

@d M_con == 250
    {Context while reading a Meta character}

@<Esc...@>=
  else if c = "C" then
  value.IM_char := row_get mod 32
  else if context >= M_con then value.IM_char := c
    {During a Meta, forbid any of the later escapes}
  else if c = "M" then
  begin
    row_char(M_con , value ) ;
    value.IM_char := value.IM_char + 128 ;
  end

@ A |narrow| character is one with zero width. To generate one, precede it
with an `N' . To mark a character  as down-loadable, precede it with  `D'. A
character cannot be both narrow and down-loadable.

@d N_con == 230
    {Context while reading a Narrow or |down_loaded| character}

@<Esc...@>=
  else if ((c = "N" ) or (c = "D" )) and (context >= N_con) then
  row_warn('Narrow or Down escape out of context')
    @.error: Row: Narrow escape...@>
  else if c = "N" then
  begin
    row_char(N_con, value ) ;
    value.breadth := 0 ;
  end
  else if c = "D" then
  begin
    row_char(N_con, value ) ;
    value.breadth := down_loaded ;
  end

@ Changing printer fonts in the middle of a |row| is done by inserting an `F'
character, followed by an integer. This is the printer font to be used, from
now on till the next `F' . Note that the initial font was passed as the 3rd
parameter to |row|.

@<Esc...@>=
  else if c = "F" then begin
    row_font := row_integer ;
    if context = 0 then row_char(1, value)
    else  row_char(context, value);
  end

@* Assembling a multi-character in |row|.

Now we come to the difficult part, which is assembling a multiple-character
command into the |ligatures| array. For this purpose, we use brackets. Curly
brackets mean that the characters inside are to be overstruck, square brackets
mean they are to be typed horizontally, and angle brackets mean that they are
to be typed vertically above each other. Finally the `U' escape (which must
come immediately after a |<| ) means to raise the (logical) cursor before
starting the vertical list.

    Warning!! I use the numerical (\.{ASCII}) values of these chars
@.System dependencies@>@.ASCII@>

@<|Row_char| const...@>=
  o_bra = "{" ;   o_ket = "}" ;
  h_bra = "[" ;   h_ket = "]" ;
  v_bra = "<" ;   v_ket = ">" ;
    {`o' means overstrike, `h' means horizontal, and `v' vertical}

@ So if we want to generate a Macsyma style summation sign, which looks like
this: \begintt
.                   ====
.                   \
.                    >
.                   /
.                   ====
\endtt
we have to insert the following mess into the |row_spec| string: \begintt
            <S[====]\[SL>]/[====]>
\endtt
The `S' is needed to get correct vertical alignment. The  `L'  is needed to
prevent the following |>| being taken as a |ket|. See the lineprinter change
file for examples.

@ In order to keep some control over all these escape sequences, I have made a
special rule of syntax. The escape sequences in |row_char| may only be nested
in a definite order. That order is: (bad characters or Control or Literal)
inside Meta inside (Narrow or Down-loadable) inside Font inside over-lists
inside |h_list|s inside |v_list|s. The parameter |context| keeps track of
this. The innermost constructions have the highest values of |context|. If
these rules are broken the user should get an error message saying `Row' and
some diagnostics.

@<Esc...@>=
  else if (c = o_bra) or (c = h_bra) or (c = v_bra)
  then begin
    if context >=  c then
    row_warn('Illegal nesting of brackets in row_spec');
      @.error: Row: Illegal nesting @>
    @<Assemble characters into |lig_buff| until we read the matching |ket|@>;
    @<Copy |lig_buff| into |ligatures| and return a pointer to it@>;
  end

@ |hoister| and |ender| are arbitrarily selected impossible classes for a
character, indicating respectively that a |v_list| has to be raised one
|char_ht| or that a |ket| has been read.

@d hoister = -32764
@d ender   = -32763

@<Assemble...@>=
  for i := 1 to max_buf do lig_buff[ i].code := no_char ;
  buf_len := 0; delta_h := 0; delta_v := 0;
  repeat
    row_char(c ,row_cod ) ;
    @<Do suitable action if |row_cod| is peculiar@>
    else begin
      incr(buf_len);
      lig_buff[buf_len].v_move := delta_v ;
      lig_buff[buf_len].h_move := delta_h ;
      incr(buf_len);
      lig_buff[buf_len].code := row_cod ;
      if c = v_bra then delta_v := delta_v + char_ht;
      if c = h_bra then delta_h := delta_h + char_width ;
    end;
  until row_cod.breadth = ender;

@ @<Row loc...@>=
  lig_buff: array[1..max_buf] of lig_thing ;
  buf_num: 0..max_buf ;
      {Number of characters (or multi-characters) in current list}
  buf_len: 0..max_buf ;
      {Number of used locations in |lig_buff|: should be |2*buf_num|}
  delta_h, delta_v: i_word;
  i : integer;
  row_cod: code_object ;

@ @<Const...@>=
  max_buf = 201;

@ @<Do suitable action...@>=
  if row_cod.breadth = hoister then delta_v := delta_v - char_ht
  else if row_cod.breadth = ender then
  else if buf_len + 3 > max_buf then
    abort('overflowed lig_buff array')
    @.Fatal: overflowed |lig_buff|@>

@ @<Esc...@>=
  else if (c = "U" ) and (context = v_bra) then value.breadth := hoister
  else if (c = "U" ) then
  row_warn('U escape out of context')
    @.error: Row: U escape...@>
  else if ((c = o_ket) or (c = h_ket) or (c = v_ket)) and (context = c-2)
  then value.breadth := ender
  else if (c = o_ket) or (c = h_ket) or (c = v_ket)
  then row_warn('mismatching brackets ')
    @.error: Row: mismatching brackets@>

@ Yet another escape is the |kern| escape. If the printer has reasonable
positioning resolution, we may want to move the parts of a multi-character
about to make them fit together better. So a |kern| takes an integer parameter
and moves the next component of the current list by that many |steps| against
the current direction. The reason for going back is that one can easily move
forwards by setting a blank space.

@d h_kern = -32762
@d v_kern = -32761

@<Esc...@>=
  else if (c = "K") and (context = h_bra) then value.breadth := h_kern
  else if (c = "K") and (context = v_bra) then value.breadth := v_kern
  else if (c = "K")
  then row_warn('Kern escape out of context' )
    @.error: Row: Kern escape @>

@ @<Do suitable action...@>=
  else if  row_cod.breadth = h_kern then delta_h := delta_h - row_integer
  else if  row_cod.breadth = v_kern then delta_v := delta_v - row_integer

@ @<Copy...@>=
  buf_num := 0 ;
  if buf_len = 0 then value.breadth :=  bad_char
  else if top_of_ligs + buf_len + 1 >= max_ligs then
  abort ('ligature array overflowed, must recompile with larger array')
    @.Fatal: ligature overflowed@>
  else begin
    for i := 1 to buf_len do
    ligatures[ top_of_ligs + i ] := lig_buff[i] ;
    top_of_ligs := top_of_ligs + buf_len + 1 ;
    buf_num := buf_len div 2 ;
    ligatures[top_of_ligs].num:= buf_num ;
    ligatures[top_of_ligs].guard := sentry ;
    value.multi := top_of_ligs ;
    value.breadth := -20000 ;
    {Provisional: a nonsense value to make sure the correct value does get
      inserted later}
  end;

@ Finally, here are the two functions that actually read the |row_spec|. The
first is |row_integer|. This reads an integer parameter for the |font| and
|kern| escapes. The parameter may have a minus sign and is terminated by the
next non-digit. (If there needs to be another digit immediately after the
parameter, then prefix it with an `L').

@<Lowest...@>=
function row_integer: integer;
  label exit ;
  var neg: boolean ; n: integer ; b: byte ;
  begin
    n := 0 ; neg := false ; b := row_get ;
    if (b = "-") then begin
      b := row_get ; neg := true;
    end
    else if ( b="+") then b := row_get ;
    if (b < "0" ) or (b > "9" ) then
    row_warn( 'no digits found by row_integer')
      @.error: Row: no digits @>
    else
    repeat
      n := n*10 + b - "0" ;
      b := row_get ;
    until (b < "0" ) or (b > "9" ) ;
    if neg then n := -n ;
    exit: row_integer := n ; decr( row_pt) ;
  end;

@ And this function gets the next character from the |row_spec|. I always have
great difficulty with this sort of program, so will go carefully. Recall:
|row_pt| points to the next character we are going to read. |b| is that
character, translated into a byte by |zord|.

@<Forw...@>=
function row_get : byte; forward;

@ @<Lowest...@>=
  function row_get;
  label exit ;
  var b: byte ;
  begin
    b:= 127 ; {any bad character}
    if row_pt > row_length then
    row_warn('fallen off end of row_string')
      @.error: Row: fallen off end@>
    else begin
      b:= zord(row_string[row_pt]) ;
      incr(row_pt);
    end;
    exit: row_get := b ;
  end;

@* Character strings.

In this section I have tried to provide some tolerable string-handling
facilities in despite of the restrictions of \PASCAL. This does not seem to
belong in any particular place in the program, but in view of the horrible
gruesome things that will happen in the next section, it seemed a good idea to
give some light relief. That is why this section is inserted here.

The |var_string| type is principally used for file names and to send command
sequences to the printer. Logically, these procedures should all be functions
and return the results, but stupid \PASCAL\ does not allow this. It would of
course be much cleaner to use the VMS |varying| type, but that would make the
program non-portable.

@<Const...@>= string_length = 100 ; {a guess, of course}

@ @<Types...@>=
  s_ptr = 1..string_length ;
  s_dat = packed array[ s_ptr] of char ;
  var_string= packed record
  len: byte;
  data: s_dat ;
  end ;

@ |@!blank| is used for initialising strings. It should not be altered
anywhere but here.

@<Set init...@>=
  blank.len := 0 ;
  for in_i := 1 to string_length do
  blank.data[in_i] := ' ' ;

@ @<Glob...@>=
blank: var_string ;

@ Here are two small procedures for printing strings.

@<Lowest...@>=
  procedure string_show(ss: var_string);
  var s_n: byte ;
  begin
    for s_n := 1 to ss.len do display(ss.data[s_n]) ;
  end;

@#
  procedure string_print(ss: var_string);
  var s_n: byte ;
  begin
    for s_n := 1 to ss.len do print(ss.data[s_n]) ;
  end;

@ @<Forw...@>=
  procedure upcase(var s: var_string) ; forward;
  {convert to upper case}

@ @<Lowest...@>=
  procedure upcase ;
  var i: s_ptr; k: byte ;
  begin
    for i := 1 to s.len do begin
      k:= zord(s.data[i]) ;
      if (k >= "a" ) and (k <= "z" ) then
      s.data[i] := zchr(k + "A" - "a" ) ;
    end;
  end;

@ We also use |var_strings| for command strings to be sent to the printer.
These nearly always use unprintable ASCII characters, typically ESCAPE. So we
need a special procedure to initialise them. It copies the |in_data| into the
|result|, but changes each |flag| into  the character |zchr(escape)|. We
determine the length by assuming that the |in_data| string is padded with some
character; then we run backwards along it until we hit the latest non-pad
character. Since \.{TANGLE} imposes a limit of 69 on the length of quoted
strings, we use the |row_str| type defined elsewhere.
  @.ASCII@>

@<Lowest...@>=
  procedure set_string
  (var result: var_string; in_data: row_str;
  flag: char; escape: byte );
  var i: byte ; last:char ;
  begin
    result := blank ;
    i := row_length ;
    last := in_data[i] ;
    while (in_data[i] = last) and (i > 1 ) do decr(i) ;
    if (i=1) and (in_data[1] = last ) then i := 0 ;
    {YEUCH! but if I write this in a natural way, it crashes when |i=0|}
    result.len := i ;
    for i := 1 to result.len do
    if in_data[i] = flag then result.data[i] := zchr(escape)
    else result.data[i] := in_data[i] ;
  end;

@#
  procedure add_char(var s: var_string; c: char) ;
  begin
    if s.len >= string_length then warn('string too long')
    else begin
      incr(s.len) ;
      s.data[s.len] := c ;
    end;
  end;
    @.error: string too long@>

@#
  procedure append(var head: var_string; tail: var_string) ;
  var k: integer;
  begin
    if head.len + tail.len > string_length
    then warn('string too long')
    else begin
      for k := 1 to tail.len do
      head.data[ k + head.len] := tail.data[ k] ;
      head.len := head.len + tail.len ;
    end;
  end;
    @.error: string too long@>

@ The next procedures generate substrings. If the character |c| is present in
|s|, then |chop_top| deletes the first |c| from |s|, and all successive
characters. |chop_tail| deletes the last |c| and all previous characters.
|chop_length| chops the string to the stated length.

@<Lowest...@>=
  procedure chop_top(var s: var_string; c:char );
  var t: var_string; i,n: byte ;
  begin
    n := 1 ;
    while (n <= s.len ) and (s.data[n] <> c) do incr(n);
    if n <= s.len then begin
      t := blank ;
      for i := 1 to n-1 do t.data[i] := s.data[i ];
      t.len := n-1 ;
      s := t ;
    end;
  end;

@#
  procedure chop_tail(var s: var_string; c:char );
  var t: var_string; i,n: byte ;
  begin
    n := s.len ;
    while (n >= 1 ) and (s.data[n] <> c) do decr(n);
    if n >= 1 then begin
      t := blank ;
      for i := 1 to s.len - n do t.data[i] := s.data[n+i];
      t.len := s.len - n;
      s := t ;
    end;
  end;

@#
  procedure chop_length(var s: var_string; k: integer);
  var n: integer;
  begin
    if (k < 0 ) or (k > s.len) then
      warn ('Impossible length supplied to chop_length' )
    else begin
      for n:= k+1 to s.len do s.data[n] := ' ' ;
      s.len := k ;
    end;
  end;
    @.error: impossible length@>

@ Printer commands usually have the format (prefix)(parameter)(suffix). These
procedures print the parameter. |s| is one character, and may have the values
`B'(yte), `D'(ecimal), `H'(exadecimal), `O'(ctal), or `W' (a 16-bit signed
word, in twos-complement notation).

@<Lowest...@>=
  procedure string_base(var result:var_string; n:integer; s:byte) ;
  { |n| to base |s| . Note that the integer is appended to |result|}
  var nh : integer ;
  begin
    nh := n ;
    if nh < 0 then begin add_char(result, '-'); nh := - nh ; end ;
    if nh >= s then begin
      string_base(result, nh div s, s) ;
      nh := nh mod s ;
    end ;
    if nh >= 10 then add_char(result, zchr(nh - 10 + "A" ))
    else add_char( result, zchr(nh + "0"  )) ;
  end;

@#
  procedure string_integer (var ss: var_string; n:integer; c:char);
  var nn: integer ;
  begin if c = 'O' then string_base(ss, n, 8)
    else if c = 'H' then string_base(ss, n, 16)
    else if c = 'D' then string_base(ss, n, 10 )
    else if (c = 'B') and (n >= 0) and (n <= 255)
    then add_char(ss, zchr(n))
    else if c = 'B' then warn('out-of-range byte')
    else if c='W' then begin
      if (n>= 0) and (n <= 32767) then begin
        add_char(ss, zchr(n div 256));
        add_char(ss, zchr(n mod 256));
      end
      else if (n<0 ) and (n> -32768) then begin
        nn := n + 65536 ;
        add_char(ss, zchr(nn div 256));
        add_char(ss, zchr(nn mod 256));
      end
      else warn('out-of-range word') ;
    end
    @<Hook for weird parameter types@>
    else warn('string_integer called with illegal type') ;
  end;
    @.error: out-of-range...@> @.error: string_integer called...@>

@#
  procedure print_integer (n:integer; c:char);
  var ss: var_string;
  begin ss := blank ;
    string_integer(ss, n, c) ;
    string_print(ss) ;
  end;

@ @<Hook for weird ...@>=

@* Translating the device-independent file, 5: Movements.

This section considers the problem of deciding where each character has to be
printed on the printer's page. This is by far and away the most difficult (and
unsatisfactory) part of \.{Crudetype}. The current version is not a properly
designed algorithm; it is merely a bodge, obtained by a lot of trial and
error. It does seem to give tolerable results on \.{WEB} files, lineprinter,
and VMS. The main variables are: |@!D_h| is `\TeX's cursor'. It gives the
`exact' horizontal position (in \.{DVI} units) generated by \.{DVI} commands.
This is always updated exactly as in \.{DVItype}. |@!IM_h| is the `page
image's cursor'. It marks the position (in |h_steps|) where the next character
will be set.

The procedure |round_IM_h| is called immediately before we set a character or
a rule. We have to take account of all the movements that occurred since the
last previous character was set.

@<Forw...@>= procedure round_IM_h( code: byte); forward ;

@ @<Lowest...@>=
  procedure round_IM_h ;
  var
  s_top, diff, n, m,
  delta, new_IM_h, rounded_h : integer ;
  begin
    @<Find the new position |new_IM_h|@>
    IM_h := new_IM_h ;
  end;

@ The obvious method is to multiply |D_h| by a factor |h_conv| and round to
nearest integer. This gives extremely bad results, because the characters in
\TeX\ fonts vary enormously in width, while many crude printers have
|fixed_width| characters. If |h_conv| is too large, then you get spaces in the
middle of words. If |h_conv| is too small, then successive characters in a
word get printed on top of each other. With an intermediate value of |h_conv|,
you get both effects at once; in other words, the characters in \TeX\ fonts
vary so much in width that the `too large' and `too small' values of |h_conv|
overlap. In this situation, a great deal of jiggery-pokery is needed to get a
tolerable result (sometimes! I have not been able to make this code work in
general.)
  For a start, here is the algorithm used in \.{DVItype}. |D_h_right| and
|IM_h_right| give the latest value of |D_h| and |IM_h| after the latest
previous character or rule was set. If the horizontal motion is small, like a
kern, |IM_h| changes by rounding the kern; but when the motion is large,
|IM_h| changes by rounding the true position |D_h| so that accumulated
rounding errors disappear. Also, we insist that the total amount of drift is
bounded.

@d h_step_round(#) == round(h_conv * # )
@d max_drift == 2

@<Find the new position |new_IM_h|@>=
  rounded_h := h_step_round(D_h) + l_margin ;
  delta := D_h - D_h_right ;
  if (delta > thin_space) or (delta <= -4*thin_space) then
  new_IM_h := rounded_h
  else new_IM_h := IM_h_right + h_step_round(delta);
  if not fixed_width then begin
    if new_IM_h > rounded_h + max_drift
      then new_IM_h := rounded_h + max_drift
    else if new_IM_h < rounded_h - max_drift
      then new_IM_h := rounded_h - max_drift ;
  end else

@ Calculating |IM_h|  on a |fixed_width| printer is very hairy. If we are not
careful, then the spaces between words will sometimes get rounded to 0. Since
we round `large' movements by rounding |D_h|, the space may even get rounded to
a negative value, if there was previously a lot of drift. So we must re-round
|new_IM_h|. The next idea is that whenever \TeX\ moves right by an amount that
seems large enough to be a space between words, we force |IM_h| to increase.

@<Find the new position |new_IM_h|@>=
  if (delta > thin_space) and (new_IM_h < IM_h_right + gap_width)
    @<Except in some special cases@>
  then new_IM_h := IM_h_right + gap_width
  else if (delta > thin_space) then do_nothing
  else if (delta > - 2*thin_space)
  then new_IM_h := IM_h_right
  else

@ Here are two little fudges which improve the result. First, when \TeX\ puts
out a thin space, it sometimes is a bit too small to be recognised as such. So
we reduce the |font_space| when a font is defined.

@<Read the font parameters...@>=
  font_space[nf] := round(font_space[nf] * 0.99 ) ;

@ The next fudge is needed to handle tables of contents. \TeX\ prints these by
putting out long streams of dots with small spaces in between. If these spaces
all get expanded to a whole character width, the right hand columns get thrown
right off the paper. So dont expand if the next character is a stop or comma.

@<Except in some special cases@>=
  and not ( ( ( code = ".") or ( code = ",") ) and
    ( ( cur_scheme > 0)  and ( cur_scheme <= max_plain )))

@ When these alternatives fail, we have lost contact between |D_h| and
|D_h_right|. This happens when \TeX\ makes a large backspace; in fact \TeX\
seems nearly always to do large backspaces by |pop| rather than an explicit
move left. \TeX\ often expresses boxes by a sequence like this:

\centerline{\tt{
PUS\markarrow{H}  Move right ------------>
\markarrow{[}set characters] \markarrow{P}OP   }}

followed by zero or more |push|es, then by a move either to one of the
positions marked by the arrows, or close by. I try to deal with this by
dropping markers at each of the arrowed positions. The markers are labelled
|D_h_right|, etc, and each marker has a corresponding value of |IM_h|
attached.

@<Glob...@>=
  D_h_left, IM_h_left, D_h_mid, IM_h_mid, D_h_right, IM_h_right,  {the markers}
  IM_h, IM_v, D_dis, IM_dis: integer;
  IM_h_stack, IM_v_stack:
    array [0..max_stack+2] of integer; {pushed down values }

@ Suppose that we are about to set a character, and |D_h-D_h_right| is large
and negative. Then we compare the current value of |D_h| with all the markers.
Let |m| be the closest of these, and |mm| the corresponding rounded value.
Then we re-round |new_IM_h| to force it to lie on the `correct' side of |mm|.
This seems to work fairly often, but it does sometimes slip. First put the
markers on top of the stack...

@<Find the new position |new_IM_h|@>=
  begin s_top := stack_ht ;
    D_h_stack[s_top] := D_h_left;
    IM_h_stack[s_top] := IM_h_left;
    incr(s_top) ;
    D_h_stack[s_top] := D_h_mid;
    IM_h_stack[s_top] := IM_h_mid;
    incr(s_top) ;
    D_h_stack[s_top] := D_h_right;
    IM_h_stack[s_top] := IM_h_right;

@ ...then look for the stacked value closest to |D_h|...

@<Find the new position |new_IM_h|@>=
  m := s_top ;
  for n := s_top downto 1 do begin
    diff := D_h - D_h_stack[n] ;
    if abs(diff) <= abs(delta) then
    begin m := n ; delta := diff; end ;
  end;

@ ...then adjust |new_IM_h| by reference to this point on the stack.

@<Find the new position |new_IM_h|@>=
  if (delta > thin_space ) and ( new_IM_h < IM_h_stack[m] + gap_width) then
  new_IM_h := IM_h_stack[m] + gap_width
  else if (delta < -thin_space )
  and ( new_IM_h > IM_h_stack[m] - gap_width)
  then new_IM_h := IM_h_stack[m] - gap_width
  else if abs(delta) <= thin_space then new_IM_h := IM_h_stack[m];
end;

@ We must assign values to these markers. When we start a page, all the
markers that were left over from the previous page are irrelevant. So we reset
them. This is a good place to consider margins. The standard arrangement given
in the \TeX book (Chapter 23) is that \.{DVI} point $(0,0)$ is about an inch
in from the top and left edges of the paper. But a negative {\tt \BS hoffset}
allows \.{DVI} to address points with negative coordinates, which should still
be on the paper. It seems that the least messy way to implement this is by
adding |l_margin| to |IM_h|, whenever this is set to an absolute value.
@.TeXbook@>@.Margins@>

@<Set up an empty page image@>=
  IM_h := @!l_margin ;
  IM_v := @!top_margin  ;
  D_h_left := 0 ; IM_h_left := l_margin ;
  D_h_mid := 0 ; IM_h_mid := l_margin ;
  D_h_right := 0 ; IM_h_right := l_margin ;

@ So now we consider the three arrows in turn. The left hand arrow will be
marked by |@!D_h_left|. It records the latest horizontal position to be
|push|ed. There might have been a |pop| since then, so it is not necessarily
the value at the top of the stack. If we just record |IM_h| whenever we
|push|, that would give a wrong value whenever there was a sequence
|push..move_right..push|. So we must rectify the pushed value of |IM_h|.

@ @<Some adjustments...@>=
  IM_h_stack[stack_ht]:=IM_h;
  IM_v_stack[stack_ht]:=IM_v;
  if just_pushed and (stack_ht > 0) then begin
    x := h_conv*(D_h_stack[stack_ht] - D_h_stack[stack_ht - 1] );
    if abs(x) > 1.5 {a guess!} then
    IM_h_stack[stack_ht] := IM_h_stack[stack_ht] + round(x) ;
  end;
  D_h_left := D_h ;
  IM_h_left := IM_h_stack[stack_ht] ;

@ The centre arrow will be marked by |@!D_h_mid|. This is defined as the value
of |D_h| just before setting the first character after the latest |push|.

@<Find the new position |new_IM_h|@>=
  if just_pushed then begin
    D_h_mid := D_h ;
    IM_h_mid := new_IM_h;
    just_pushed := false;
  end;

@ The right hand arrow is marked by |@!D_h_right|. At any time, this is
defined as the right hand edge of the latest previous character (or rule) that
has just been set. This equals |D_h + D_dis|, where |D_dis| is the \TeX\ width
of the character. Usually there will follow a |move_right| that updates |D_h|,
but |D_h_right| must be updated even if there is no |move_right|. Now
|@!IM_h_right| must be aligned with the right hand edge of the printed
representation of the character. The idea is that this will usually be the
exact place where the next character has to be set. We hope that all the
characters in each word will be correctly placed next to one another and the
accumulated drift will appear in spaces between the words. So whenever a
character is set, we must assign values to |D_dis| and |IM_dis|. The character
is described by |cod|, and its printed width is written into its |breadth|
field; but if it is a multiple character, then the |breadth| is the negative
of the width.

@<Do messy things...@>=
  D_dis := D_width[D_font, c_num] ;
  if cod.breadth = bad_char then IM_dis := 0
  else IM_dis := abs(cod.breadth) ;
  @<Set |rail_base|@>
  D_h_right := D_h + D_dis ;
  IM_h_right := IM_h + IM_dis ;

@ So the procedure |row| must give the |breadth| field the right value when
assembling a |multi| character. Recall that that character can be either an
|o_list| or an |h_list| or a |v_list|, and |c| tells us which it is. An
|o_list| is assumed to have a width of one |char_width| and the width of a
|v_list| is the width of its widest component. The width of a |h_list| gets
accumulated in |delta_h| as the list is assembled.

@<Copy |lig_buff|...@>=
  if c = o_bra then print_width := char_width
  else if c = h_bra then print_width := delta_h
  else begin
    print_width := char_width ;
    for i := 1 to buf_num do
    with lig_buff[2*i].code do
    if (print_width < -breadth ) and (breadth > -30000 )
    then print_width := -breadth  ;
  end;
  value.breadth := - print_width ;

@ @<Row locals...@>=
  print_width: integer ;

@ We must do the same thing when setting a rule.

@<|Post| set...@>=
  D_dis := D_rul_width ;
  IM_dis := hn * post_width ;
  D_h_right := D_h + D_dis ;
  IM_h_right := IM_h + IM_dis ;

@ @<|Rail| set...@>=
  D_dis := D_rul_width ;
  IM_dis := hn * rail_width ;
  D_h_right := D_h + D_dis ;
  IM_h_right := IM_h + IM_dis ;

@ \.{DVItype} handles vertical motion in the same sort of way as horizontal.

@d v_step_round(#) == round(v_conv * # )

@<Medium...@>=
  procedure move_down(ddd: integer);
  var new_IM_v , delta : integer;
  begin
    D_v:=D_v+ddd;
    delta := v_step_round(ddd) ;
    @<Find a vertical position |new_IM_v|@>
  end;

@ @<Find a vert...@>=
  if delta >= big_drop then begin
    new_IM_v := v_step_round(D_v) + top_margin ;
    if new_IM_v < IM_v + big_drop then
    IM_v := IM_v + big_drop
    else IM_v := new_IM_v ;
    rail_base := IM_v * rail_types ;
  end
  else if delta <= -big_drop then begin
    new_IM_v := v_step_round(D_v) + top_margin ;
    if new_IM_v > IM_v - big_drop then
    IM_v := IM_v - big_drop
    else IM_v := new_IM_v ;
    rail_base := IM_v * rail_types ;
  end else

@ The above calculation fails for small motions. Because \TeX\ expects
subscripts to be about half the size of the main line, it drops them by only a
small amount; with a crude printer, this small amount gets rounded to zero. If
the move is smaller than |@!tiny_drop| \.{DVI} units, we ignore it. If not,
then we force the new value of |IM_v| to be different from the old.

@<Find a vert...@>=
  begin
    IM_v := IM_v + delta ;
    rail_base := rail_base +  v_step_round(ddd * rail_types) ;
    if (ddd >  tiny_drop) and ( delta = 0) then IM_v := IM_v + 1
    else if (ddd < -tiny_drop) and ( delta = 0) then IM_v := IM_v - 1
    else rail_base := IM_v * rail_types ;
  end;

@ The next bit is put in to help catch bugs. Sometimes the \.{DVI} file tries
to address an absurd position; for example, I contrived to make \TeX\ generate
a {\tt \BS hbox} that was 9000 points wide. If we do nothing about this,
\.{Crudetype} will probably crash with an arithmetic error, which is
unacceptable. So any character falling outside the limits |h_min..h_max| and
|v_min..v_max| will generate an error report.

@<Check the position@>=
  if (Set_h < h_min) or ( Set_h > h_max )
  then begin
    warn('out of bounds position') ;
    Set_h := h_min ;
    {Chuck the character somewhere, hopefully out of the way}
  end;
  if (Set_v < v_min) or ( Set_v > v_max )
  then begin
    warn('out of bounds position') ;
    Set_v := v_min ;
  end;
    @.error: out of bounds@>

@ @<Set init...@>=
  h_max := h_resolution * 100 ;
  v_max := v_resolution * 100 ;
  h_min := -10 * h_resolution ;
  v_min := -10 * v_resolution ;

@ Note that since the position fields of a |page_record| are subranges,
|h_max| etc. must be of the same type.

@<Glob...@>= h_max, v_max , h_min, v_min : i_word ;

@* Sorting the page.

Once we have assembled the complete page image, we must sort it. The method
used here is a merge sort based on the country dance called Grand March.

@<Sort the page@>=
  @<The dancers form a long line up the middle of the hall and march
    up towards the Presence@>
  repeat
    @<At the top they split, and alternate groups go to the left and right and
      march down the sides@>
    @<At the bottom of the hall, each group coming from the right hand side
      merges with a group from the left side, and they go up again @>
  until sorted;

@ Since the data being sorted is of unpredictable size and sequentially
processed, it logically ought to be a |file|. But this turned out to make the
program spectacularly slow. So I use linked lists instead--- a sacrifice of
logic to economy. But I continue to use file-like language.

@d send_one_set_to( #)==
  copy_from( mid ) ( # )

@<At the top...@>=
  L_reset( mid) ;
  L_rewrite( left) ;
  L_rewrite( right) ;
  repeat
    send_one_set_to( left) ;
    if not L_eof( mid) then
    send_one_set_to( right) ;
  until L_eof( mid);

@ Eventually everybody comes together in one enormous set and the dance is
finished. The easiest way to detect this is to let it go round one more time.
Then the left side of the hall will be full and the right hand side empty.

 @<At the bottom...@>=
  L_rewrite( mid) ;
  L_reset( left) ;
  L_reset( right) ;
  sorted := L_eof( right) ;
  if sorted then
    page_ptr := son( next( left))
  else repeat
    if L_eof( right) then copy_from( left)  ( mid)
    else if L_eof( left) then  copy_from( right)  ( mid)
    else @<Merge one group from each side@>
  until L_eof( left) and L_eof( right) ;

@ The natural way to assemble the page image is to throw everything into one
huge list, then start sorting. But the code for merging two simple lists was
horribly complicated. (The code given here merely merges two runs.) So the
page image is a list of lists (another sacrifice of logic to economy). Each
top-level entry has a |son|, which points to a sub-list. This is a sorted
subset (a ``run'') of the data. One advantage of the list-of-lists structure
is that we can take advantage of the fact that \TeX\ output is very ``runny''.
I found that this made \.{Crudetype} run at least 3 times faster than before.

@d Add_run == new_tail( mid_ptr) ; son( mid_ptr) := run_ptr ;

@<Merge one group...@>=
  begin
    L_rewrite( run) ;
    L_run_ptr := son( left_ptr) ;
    R_run_ptr := son( right_ptr) ;
    repeat
      if @<The person on the left is more eligible@>
      then copy_from( L_run) ( run)
      else copy_from( R_run) ( run) ;
    until L_eof( R_run) and  L_eof( L_run) ;
    step_wipe( left_ptr) ;
    step_wipe( right_ptr) ;
    L_reset( run) ;
    Add_run ;
  end;

@ So while the page image is being assembled, it must be divided into runs.

@<Add the record...@>=
  begin
    if out_of_sequence then begin {create a new run}
      L_reset( run) ;
      Add_run ;
      L_rewrite( run) ;
    end;
    new_tail( run_ptr ) ;
    with image( run_ptr) do begin {write the data into it}
      hpos := Set_h ; Old_h := Set_h ;
      vpos := Set_v ; Old_v := Set_v ;
      data := cod ;
    end;
    incr(page_size) ;
    if page_size >= page_max then abort(
      'overflowed page: either a bug, or recompile with larger page_max' ) ;
  end
    @.Fatal: overflowed page@>

@ Once the lists are all assembled, we must |reset| them before sorting.

@<The dancers...@>=
  sorted := false;
  L_reset( run) ;
  Add_run ;

@ Now we must specify the desired order!! That is: increasing |vpos| and
|hpos|, |vpos| is more significant.

@d out_of_sequence ==
  ( ( Old_v > Set_v) or ( ( Old_v = Set_v) and ( Old_h > Set_h)))

@<The person on the left is more eligible@>=
  ( ( image( L_run_ptr).vpos < image( R_run_ptr).vpos) or
    ( ( image( L_run_ptr).vpos = image( R_run_ptr).vpos)
      and ( image( L_run_ptr).hpos <= image( R_run_ptr).hpos)))

@ And here we get it all started. Since |garbage| wipes out everything in the
|pool| array above |zzz|, the following code effectively makes |mid..run|
permanent.

@<Set init...@>=
  first_cell ;
  make_new( mid );
  make_new( left );
  make_new( right );
  make_new( run );
  make_new( zzz );
  image(zzz).vpos := max_half;
  next(zzz) := zzz ;
  mid_ptr := zzz ;
  run_ptr := zzz ;

@ @<Set up an empty page image@>=
  garbage ;
  L_rewrite( mid) ;
  L_rewrite( run) ;
  page_size := 0 ;
  Old_v := -max_half ;

@ @<Glob...@>=
  zzz, cell, tempp, page_ptr,
  mid, mid_ptr, run, run_ptr,
  left, left_ptr, L_run_ptr ,
  right, right_ptr, R_run_ptr : link;
  page_size: page_i ;
  Old_v, Old_h : i_word ;
  sorted: boolean ;
  declare_pool

@ Now we must define lots of machinery for handling lists. We could represent
a list by either a big array or dynamic storage. Neither is ideal, because an
array is bound to be either too big or too small; and some \.{PASCAL}s
apparently do not implement pointers. So I have expressed everything in terms
of certain macros, defined in the system dependent part of the program. In
theory, you can switch \.{Crudetype} from array to heap merely by redefining
these as follows:
\begintt
    define image(#) == #^
    define create == new(cell)
    define first_cell ==
    define link_type == ^page_record
    define wipe_out(#) == dispose(#) ; { release data piecemeal}
    define garbage ==
    define declare_pool ==
\endtt

Both array and heap seem to work in VMS. I prefer to use an array because in
VMS, there seems to be no shortage of store, and an array is easier to debug.
Assuming these lowest-level macros, here is some machinery for handling lists.
We must deallocate cells after use. When using arrays, the |garbage| command
does it all in one go. Pointers must be |dispose|d one at a time, and the
obvious time is just after the data was used.

@d next(#) == image(#).prox
@d advance(#) == # := next(#)
@d make_new( #) == create; # := cell ;
@d new_tail( #) ==
  create; next( #) := cell; # := cell ;
@d step_wipe( #) ==
   tempp := # ; advance( #) ; wipe_out( tempp)

@ Suppose |L| is a list; then the actual variable |L| points to a permanently-
allocated cell which in turn points to the head of the list. |L_ptr| points to
the active end. After the list has been assembled, we first mark the tail, by
attaching a special element called |zzz|. Then we move the |L_ptr| round to
the head. |copy_from| must be used in the combination
{\tt copy\_from(A) ( B)}. It copies one element from the head of |A| to the
tail of |B|.

@d L_rewrite( #) ==
  #@&ptr := # ; next( #) := zzz
@d L_reset( #) ==
  next ( # @& ptr) := zzz ; #@&ptr := next( #)
@d L_eof( #) ==
  ( # @& ptr = zzz)
@d copy_end( #) ==
  next( #@&ptr) := tempp ; advance( #@&ptr) ; end
@d copy_from( #)==
begin
  tempp := #@&ptr ;
  advance( #@&ptr ) ;
  copy_end

@ Each top-level entry has the |false| type below; the |prox| field points to
the next top-level entry and the |down| field to a sub-list.

@d son(#) == image(#).down

@<Types...@>=
  page_i = 0..page_max ;
  link = link_type ;
  page_record = packed record
    prox: link ;
    case boolean of
      true: ( hpos : i_word;
        vpos: i_word;
        data: code_object ) ;
      false: ( down : link) ;
    end;

@* Processing a page of output.

The output of \.{Crudetype} is done by the procedure |Send_page|, which takes
the page and translates it for the printer. We shall process it a `line' at a
time, meaning all |page_records| with the same |vpos|. Initially |PR_font|
gets an impossible value so as to force an explicit |set_PR_font|.

@<Top level...@>=
  procedure Send_page;
  var line: link ;
  begin
    @<Pause reset@>;
    PR_font := sentry;
    PR_h := 0;
    PR_v := 0;
    repeat
      line := read_line ;
      do_line(line);
    until L_eof( page) ;
  end;

@ The function |read_line| runs along the page image until the vertical
position changes. It returns a pointer to a sublist which is the next line on
the page. As side effects, it moves the printer into position for this line,
advances |page_ptr| to the first record of the next line, and updates
|PR_v| and |PR_v_next|.

@<Medium...@>=
  function read_line : link ;
  var head, tail: link ; size: integer;
  begin
    head := page_ptr ;
    size := 0 ;
    PR_v_next := image(page_ptr).vpos ;
    @<Move printer vertically to |PR_v_next|, update |PR_v| @>;
    repeat
      tail := page_ptr ;
      advance(page_ptr) ;
      PR_v_next := image(page_ptr).vpos ;
      incr(size) ;
      if size = max_line_size then
      warn('excessively long line, probably this is a bug') ;
        @.error: excessively long line@>
    until ( ( L_eof( page) )
      or (PR_v_next <> PR_v ) ) ;
    next(tail) := zzz ;
    read_line := head ;
  end;

@ These variables all denote the printer fonts, etc.

@<Glob...@>=
  PR_v, PR_v_next,
  PR_h, PR_h_next,
  PR_font : i_word ;

@ This procedure tries to print a line. The main difficulties are: we dont
want to |Backfeed| unless absolutely necessary; and we may have to deal with
overstruck characters. One possible way is to shunt them aside somewhere, then
print the |overflow| after the main line has been printed.

@<Medium...@>=
  procedure do_line (line_ptr: link);
  var overflow : link;
  begin
    overflow := zzz ;
    while line_ptr<>zzz do
    @<Process the character that |line_ptr| points to, and |advance| to
      the next@>;
    @<End the line, trying very hard not to over-feed the paper and print
      the |overflow|@>;
    @<Check pause@>;
  end;

@ We are actually getting almost in sight of the printer!!! Before we can
actually print a character, we must first check if it has to go to the
|overflow|...

@<Process the char...@>=
  with image(line_ptr) do begin
    PR_h_next := hpos ;
    if not b_space_absolute and not b_space_by_string and
    (PR_h_next < PR_h) then
    begin
      {AKT: ignore overflow stuff
      next(overflow) := line_ptr ;
      advance(overflow) ;
      AKT}
      advance(line_ptr) ;
    end
    else begin
      @<Set horizontal position for the next character@>;
      if  data.IM_font <> PR_font then set_PR_font(data.IM_font);
      print(zchr(data.IM_char )) ;
      PR_h := PR_h + data.breadth ;
      step_wipe(line_ptr ) ;
    end ;
  end;

@* Downloading. Not started yet.

@<Download a whole font@>= do_nothing

@ @<Enter a download...@>=

@* Carriage control.

Once the superior software has decided where the printer has to move to next,
this section has the job of translating the desired position into elementary
printer commands. Clearly this mapping depends very much on the range of
functions that the printer can perform. So this section is controlled by
several boolean constants; each asserts that the printer can do the
corresponding action. Here is a list of the most important ones:\item
|@!c_r_feed_dist| is the distance in |v_steps| by which a carriage-return
feeds the paper.\item
|@!w_l_feed_dist| ditto, |write_ln|. Similarly for the other |dist|
values.\item
|@!feed_absolute| says the printer has an absolute position command that takes
a parameter |IM_y|, say, and moves to position |IM_y v_steps| down the
page.\item
|@!b_feed_absolute| ditto, backfeeding.\item
|@!b_feed_by_string| says the printer has a |Backfeed| character that moves it
back by a fixed number |b_feed_dist| of |v_steps|. These booleans should not
be set true unless the printer can backfeed reliably.\item
|@!space_absolute| etc., Ditto, horizontal moves.\item
|@!abs_is_incr| says that in the absolute position commands, the parameter is
actually an incremental move.\item
|@!w_l_does_c_r| says that |write_ln| forces a carriage return.

As mentioned above, it is essential to avoid premature line feeds as much as
possible. Also, many operating systems will choke if the output record gets
too long, so we must do a |print_ln| at intervals. This program tries to
accommodate various types of carriage control, some of which are not in use at
the author's site. This means that several pieces of code have not been
tested. Installers may find that the procedures defined here will need to be
carefully studied in conjunction with the I/O section of their \PASCAL\
manual.

@<Const...@>= @<Carriage control constants@>

@ Now consider what happens at the end of each line. We will want to do a
subset of the following things: carriage-return, print the |overflow|, line
feed, split output records. We must keep a clear separation between these
tasks, and we want to do them in the stated order (but we cannot if
|fortran|). This order puts most of the carriage controls to the ends of the
output records, and (on our machine) makes it easier to examine the output
file with an editor. So first: do we want to do  carriage-return? If so, then
the natural way is to print a carriage-return, but not if it will over-feed
the paper.

@<End the line...@>=
  if  not w_l_does_c_r      {Return is compulsory}
  or (c_r_feed_dist = 0)   {Return is harmless  }
  or b_space_absolute
  or ((not want_split or (overflow <> zzz ) ){We can choose C-R or W-L}
    and (c_r_feed_dist < w_l_feed_dist))
  then begin
    if b_space_absolute and ((c_r_feed_dist > 0) or (l_margin > 0)) then
    set_h_abs(0)
    else begin
      if fortran then print_ln ;
      if not list then             {AKT: only send CR if not list}
      print(c_r_char);
      PR_h := 0;
      PR_v := PR_v + c_r_feed_dist ;
    end;

@ Now for the |overflow|. We will split records if that is harmless.

@<End the line...@>=
  if overflow <> zzz then begin
    if not fortran and (w_l_feed_dist = 0) then print_ln ;
                                           {AKT: must do this if list???}
    next(overflow) := zzz ;
    overflow := next(zzz) ;
    do_line(overflow);
  end;
  @<Reset printer at end of line, if necessary @>
end;

@ @<Reset printer...@>= {hook}

@ Now we decide whether to do any |line_feed|s. But first, we may have to
attempt to |Backfeed|. Sometimes the program will fail; it should not do so
unless the \.{DVI} file calls for overstruck characters and the printer
genuinely cannot do them. If |b_feed_scream|, then print an error message.

@<Move printer...@>=
  if want_split then PR_v_next :=  PR_v_next - w_l_feed_dist ;
  if (PR_v_next < PR_v) then begin
    if b_feed_absolute then set_v_abs(PR_v_next)
    else if b_feed_by_string then
    while PR_v_next < PR_v do @<Backfeed@>
    else if b_feed_scream then begin
      warn('this printer cant feed backwards');
        @.error: printer cant...@>
      display_ln('approximate vertical position is: ', PR_v_next);
      display_ln(' printing over-fed line on line below');
      display_ln(' ');
      PR_v := PR_v_next;
    end;
  end;

@ If we avoided over-feeding, we may want to feed forwards.

@<Move printer...@>=
  if PR_v_next > PR_v then begin
    if feed_absolute then set_v_abs(PR_v_next)
    else begin
      while PR_v_next >= PR_v + feed_dist do @<Line feed@>;
      while PR_v_next > PR_v do @<Tiny feed@> ;
    end;
  end;
  if want_split then begin
    PR_v := PR_v + w_l_feed_dist ;
    PR_v_next := PR_v_next + w_l_feed_dist ;
    if not list then print_ln;                {AKT: was just print_ln;}
    if w_l_does_c_r then PR_h := 0 ;
  end;

@ We set the horizontal position in a similar way, but we do not need to be so
paranoid about backspacing as about back-feeding.

@<Set horiz...@>=
  if PR_h_next = PR_h then
  else begin
    if  (PR_h_next < PR_h) then begin
      if b_space_absolute then set_h_abs(PR_h_next)
      else if b_space_by_string then
      while PR_h_next < PR_h do @<Backspace@>;
    end;
    if space_absolute and (PR_h_next > PR_h )
      then set_h_abs(PR_h_next)
    else begin
      while PR_h_next >= PR_h + space_dist do @<Space@>;
      while PR_h_next > PR_h do @<Tiny space@> ;
    end;
  end;

@* Low level modules for printer control.

Now we have to translate these elementary printer commands into actual strings
of characters to be put into |printfile|. Here is the command for setting a
new printer's font.

@<Lowest...@>=
  procedure set_PR_font(new:integer) ;
  begin
    if (new = PR_font) or only_one_font then
    else
    begin
      string_print(font_prefix) ;
      print_integer(new, param_type);
      string_print(font_suffix) ;
      PR_font := new ;
    end;
  end;

@ Now for |absolute| movements, if the printer can do them. The procedure
|set_v_abs| moves the printer to position |mm h_steps| below the top of the
paper. If |abs_is_incr| then the printers `absolute' command is actually an
incremental command. So the parameter sent to the printer must be decreased by
|PR_v|.

@<Forward...@>=
  procedure set_v_abs(mm: integer) ; forward ;
  procedure set_h_abs(mm: integer) ; forward ;

@ @<Lowest...@>=
  procedure set_v_abs;
  var new_pos :integer ;
  begin
    if abs_is_incr then
    new_pos := mm - PR_v
    else new_pos := mm ;
    string_print (v_abs_prefix) ;
    print_integer(new_pos, param_type) ;
    string_print (v_abs_suffix) ;
    PR_v := mm ;
  end;

@#
  procedure set_h_abs;
  var new_pos :integer ;
  begin
    if abs_is_incr then
    new_pos := mm - PR_h
    else new_pos := mm ;
    string_print (h_abs_prefix) ;
    print_integer(new_pos, param_type) ;
    string_print (h_abs_suffix) ;
    PR_h := mm ;
  end;

@ Now consider commands for printers that can only do simple movements. A
|tiny| movement is usually a movement of one |h_step| or |v_step|. All these
modules should be protected, so they cannot be called unless the printer can
actually do the stated movement. Normally, the command strings for these are
only simple characters, so we can just |print| them.

@<Formfeed@>=
  {AKT: was page(printfile);}
     print_ln;
     print_ln;
     print_ln;
     print('-------------------- [new page] --------------------');
     print_ln;
     print_ln;
  if is_header then
  string_print (page_top);

@ @<Backfeed@>=
  begin string_print(b_feed_string);
    PR_v:=PR_v - b_feed_dist;
  end

@ @<Line feed@>=
  begin
    if fortran then begin
      print_ln ;
      PR_v := PR_v + w_l_feed_dist;
    end;
    if list then           {AKT: was just print(feed_char);}
       print_ln
    else
       print(feed_char);
    PR_v:=PR_v+feed_dist;
  end;

@ @<Tiny feed@>=
  begin print(t_feed_char); PR_v:=PR_v+t_feed_dist; end;

@ @<Backspace@>=
  begin print (b_space_char); PR_h:=PR_h-b_space_dist; end;

@ @<Space@>=
  begin print (space_char); PR_h:=PR_h+space_dist; end;

@ @<Tiny space@>=
  begin print (t_space_char); PR_h:=PR_h+t_space_dist; end;

@* Default declarations for printer.

Here we define a lot of printer-dependent material that is expected to be the
same for most printers. Of course, these will have to be changed if |fortran|,
or on a system that does not use ASCII codes. First, some command characters
for simple movements. |feed| means a vertical movement and |space| horizontal.
Each |thing_char| is the character needed to make the printer do the named
action. Owing to the rules of \.{TANGLE}, the words |back| and |tiny| have to
be abbreviated (to avoid identifier clashes). |c_r_char| etc. must be
consistent with the value of |fortran|.
  @.ASCII@>

@<Set init...@>=
  space_char := chr(32) ;
  t_space_char := chr(32) ;
  feed_char := chr(10) ;
  t_feed_char := chr(10);
  c_r_char := chr(13);
  b_space_char := chr(8);

@ @<Glob...@>=
  space_char,
  t_space_char,
  feed_char ,
  t_feed_char ,
  c_r_char ,
  b_space_char : char ;

@ Next the distances that they normally move, always in |steps|.

@<Carriage control const...@>=
  space_dist = 1;
  b_space_dist = 1;
  t_space_dist = 1;
  feed_dist = 1 ;
  w_l_feed_dist = 0 ;
  b_feed_dist = 0;
  t_feed_dist = 1;
  c_r_feed_dist = 0 ;
  tiny_drop = 500000 ;  {AKT: avoid E in TEX dropping to next line}
  {tiny_drop = 50000 ;   slightly less than a point}
  big_drop = 4 ;

@ |start_stuff|  and |stop_stuff| get written into the start and end of
|printfile|. They are intended to: set printer into correct state for \TeX\
output, and reset printer to standard state afterwards. If the printer needs
to be re-initialised in any way at the top of each page, then set |@!page_top|
to the necessary data and set |is_header| to |true|.

@<Open |printfile|@>=
  string_print(start_stuff) ;
  print_ln ;

@ @<Clean up afterwards@>=
  string_print(stop_stuff);

@ @<Glob...@>=
  start_stuff,
  stop_stuff,
  page_top,
  b_feed_string ,
  font_prefix,
  font_suffix,
  v_abs_prefix,
  v_abs_suffix,
  h_abs_prefix,
  h_abs_suffix : var_string ;
  print_end : var_string ;

@* Printer dependent data.

This section should define masses of data to describe how the printer behaves.
In order to keep the size of each printer's change file within reasonable
bounds, I have replaced this section by a blank.  The missing data is given in
the line printer change file. To set up for another printer, that file will
have to be extensively edited.

*** Attach printer change file here ***

@* Index.

Pointers to error messages appear here together with the section numbers
where each identifier is used.