[exim.git] / doc / doc-scripts / g2t

#! /usr/bin/perl -w

# A Perl script to turn the SGCAL source of the Exim documentation into
# Texinfo input, more or less...

# Supply the source file names as arguments.
# The output goes to the standard output.


##################################################
#         Ensure unique node name                #
##################################################

# Node names must be unique. Occasionally in the Exim spec there are duplicate
# section names, and it's become too much of a hassle to keep them distinct
# manually. So it is now automated.

########### Never really got this working. Abandoned ###############

sub unique {
my($node) = $_[0];
if (defined $node_names{$node})
  {
  $node_names{$node} += 1; 
  $node = "$node ($node_names{$node})"; 
  
print STDERR "+++ $node\n";
 
  }
else
  {
  $node_names{$node} = 0;
  }
$node;
}  


##################################################
#         De-comma a node name                   #
##################################################

# Commas, colons, and apostrophes are not permitted in Texinfo
# node names. I find this incredible, but it is clearly documented.
# The Exim manual has been re-organized not to have colons or
# apostrophes in any chapter or section titles, but I can't manage
# without commas. This function turns "," into " and", which is
# the best that can be done; we can use some clever Perlery to
# just take out commas before "and".

# Sigh. The Sendmail option -p<rval>:<sval> now means that there's a colon
# in the node name for that option. Turn the colon into <colon>. This is also
# done for menus.

# Another thing that causes problems in node names in some versions of
# Texinfo is the use of @sc{xxx} for small caps. Just turn these all into
# real caps. This is also done for menus.

sub decomma {
$_[0] =~ s/,(?!\sand)/ and/g;
$_[0] =~ s/,//g;
$_[0] =~ s/\@sc\{([^}]*)\}/\U$1/g;
$_[0] =~ s/:/<colon>/g;
$_[0];
}


##################################################
#           De-quote a string                    #
##################################################

# @x is turned into x, except when x=@, or when asis is set,
# in which case single @ must turn into @@. A single substitute
# doesn't work in the non-asis case, because of the problems of
# handling things like @@@$, so we do it the hard way.

sub dequote {
if ($asis) { $_[0] =~ s/@/@@/g; } else
  {
  $_[0] =~ s/@@/&at&/g;
  $_[0] =~ s/@([^@])/$1/g;
  $_[0] =~ s/&at&/@@/g;
  }
$_[0];
}


##################################################
#           Get next line                        #
##################################################

# Called from handle_directive, to get the next source line
# into $_.

sub get_next_line {
if ($processing_subsection)
  { return $_ = shift @SUBBUFFER; }
else
  { return $_ = <>; }
}


##################################################
#           Handle text lines                    #
##################################################

# This function is handed whole paragraphs, and we assume that
# SGCAL font changing markup is always complete within a paragraph.
# We have to replace escaped versions of significant characters with
# some magic before performing general transformations, and then
# put them back afterwards. The character & is not common in the text,
# and && is unknown, so we use that.

sub handle_text {
$_ = $_[0];

if ($asis)
  {
  $_ = dequote($_);
  s/(\{|\})/\@$1/g;
  return $_;
  }

while (/~~/)
  {
  $left = $`;
  ($name) = $' =~ /^(\w+)/;
  $right = $';

  $value = $references{$name};
  $value = "" if !defined($value);

  if ($value =~ /\*\*\*\*/)
    {
    $value = ($` eq $current_chapter)? "\"$'\"" :
      "\"$'\" in chapter \"$`\"";
    $value = "" if $value eq "\"\"";
    }
  elsif ($value !~ /^[0-9]+\.[0-9]+$/)   # quote unless version number
    {                                                
    $value = "\"$value\"";                          
    }

  $_ = "${left}${value}${right}";
  }

s/\@\@/&&a/g;         # @@
s/\@\\/&&b/g;         # @\
s/\@</&&l/g;          # @<
s/\@>/&&g/g;          # @>
s/\@\{/&&c/g;         # @{
s/\@\}/&&d/g;         # @}
s/\@#/&&s/g;          # @#

# Now remove all other @'s

$_ = dequote($_);

# Convert SGCAL markup

s/#/ /g;                            # turn #   into a space
s/\$~//g;                           # turn $~  into nothing
s/__/_/g;                           # turn __  into _
s/\$sm\{//g;                        # turn $sm{     into nothing
s/\$sc\{([^\}]*?)\}/$1/g;           # turn $sc{xxx} into xxx
s/\$st\{([^\}]*?)\}/$1/g;           # turn $st{xxx} into xxx
s/\$si\{([^\}]*?)\}/$1/g;           # turn $si{xxx} into xxx
s/\$tt\{([^\}]*?)\}/$1/g;           # turn $tt{xxx} into xxx

s/\$it\{([^\}]*?)\}/$1/g;           # turn $it{xxx} into xxx

s/\$bf\{([^\}]*?)\}/$1/g;           # turn $bf{xxx} into xxx
s/\$rm\{([^\}]*?)\}/$1/g;           # turn $rm{xxx} into xxx
s/\$cb\{([^\}]*?)\}/$1/g;           # turn $cb{xxx} into xxx

# This is a fudge for some specific usages of $<; can't just do a global
# is it occurs in things like $<variable name> as well.

s/\[\$<\]/[]/g;                     # turn [$<]     into []
s/&&b\$<\./&&b./g;                  # turn \$<.     into \.  (\ == &&b by now)
s/(\d)\$<-/$1-/g;                   # turn 0$<-     into 0-

# There is one case where the terminating } of an escape sequence is
# in another paragraph - this follows $sm{ - it can be fixed by
# removing any stray } in a paragraph that contains no { chars.

s/\}//g if !/\{/;

# Any remaining {} must be escaped to prevent Texinfo from complaining

s/(\{|\})/\@$1/g;

# Now to conversions that put {} into the file.
# Change <<..>> from @var to just <...> as the caps that Texinfo
# uses look far too shouty.

s/\\\\([^\\]*?)\\\\/\@sc\{\L$1\}/g; # turn \\xxx\\  into @sc{xxx}
s/\\\(([^)]*?)\)\\/\@file\{$1\}/g;  # turn \(xxx)\  into @file{xxx}
s/\\\"([^\"]*?)\"\\/\@file\{$1\}/g; # turn \"xxx"\  into @file{xxx}

s/\\\?([^?]*?)\?\\/$1/g;            # turn \?URL?\    into URL
s/<<([^>]*?)>>/<$1>/g;              # turn <<xxx>>    into <xxx>
s/\\\$([^\$]*?)\$\\/\$$1/g;         # turn \$xxx$\    into $xxx
s/\\\-([^-]*?)\-\\/\-$1/g;          # turn \-xxx-\    into -xxx
s/\\\*\*([^*]*?)\*\*\\/$1/g;        # turn \**xxx**\  into xxx
s/\[\(([\w\/]*)\)\]//g;             # remove inline HTML

s/\\\*([^*]*?)\*\\/\@dfn\{$1\}/g;     # turn \*xxx*\    into @dfn{xxx}
s/\\%([^*]*?)%\\/\@dfn\{$1\}/g;       # turn \%xxx%\    into @dfn{xxx}
s/:::([^:]*?)::/\@dfn\{:$1:\}/g;      # turn :::xxx::   into @dfn{:xxx:}
s/::([^:]*?)::/\@dfn\{$1:\}/g;        # turn ::xxx::    into @dfn{xxx:}
s/\\([^\\]*?)\\/\@dfn\{$1\}/g;        # turn \xxx\      into @dfn{xxx}
s/\$\*\$/\*/g;                        # turn $*$        into *

# Put back escaped SGCAL specials

s/&&a/\@\@/g;
s/&&b/\\/g;
s/&&l/</g;
s/&&g/>/g;
s/&&c/\@{/g;
s/&&rc/{/g;
s/&&rd/}/g;
s/&&d/\@}/g;
s/&&s/#/g;

# Remove any null flags ($$)

s/\$\$//g;

# If the paragraph starts with $c\b, change this into @center. Assume
# we don't ever get two of these in a row.

s/^\$c\b/\@center /;

# If the paragraph starts with $e\b, stuff some tabs in there, as
# Texinfo can't do this on its own (as far as I can see). They must
# tabs; Texinfo treats them as different to spaces. Sigh.

s/^\$e\b/\t\t\t\t\t\t\t/;

# Handle $t. The Exim spec only ever has one tab per line. Er, not
# quite true, but a good enough assumption. $t is always followed
# by a non-word character.

# The .tabs directive has stashed the value in the $tab variable.
# Don't count Texinfo font chars.

while (/(^|.+?\n)(.+?)\$t(\W.*\n)/)
  {
  $before = $` . $1;
  $after = $';
  $left = $2;
  $right = $3;

  $left =~ s/\s$//;
  $right =~ s/^\s+//;

  $plainleft = $left;
  $plainleft =~ s/\@[a-z]+\{([^}]+?)\}/$1/g;
  $plainleft =~ s/\@//g;

  $_ = $before . $left . (" " x ($tab - length($plainleft))) . $right . $after;

  # Fudge for the one case where there are two tabs

  if ($tab2 != 0)
    {
    $temp = $tab;
    $tab = $tab2;
    $tab2 = $temp;
    }
  }

# Return the new line (paragraph)

$_;
}


##################################################
#           Handle directive lines               #
##################################################

# Use get_next_line() instead of <> because this is called to process
# stacked up subsection lines

sub handle_directive {

my($new_lastwasitem) = 0;

# Chapter directives just require . => @; however, dequoting the
# line thereafter will remove the first @, so just force it back
# afterwards. If the chapter is is one describing a driver, set
# the driver name.

if (/\.chapter/)
  {
  tr/./@/;
  push(@ONESECTION, "@" . &dequote("$_\n"));
  $driver_name = (/The\s+(\S+)\s+(director|router|transport|authenticator)/)? $1 :
    (/Generic options common to both directors and routers/)?
      "director or router" :
    (/[Gg]eneric\s+options for (\S+)s/)? $1 : "";
  $driver_name = &dequote($driver_name);
  }

# Section directives just require . => @; however, dequoting the
# line thereafter will remove the first @, so just force it back
# afterwards. Remove any colons in section titles as they cause
# Texinfo trouble. Also remove any \\ (small caps) markup, which
# appears in a couple of cases.

elsif (/\.section/)
  {
  tr/./@/;
  s/://;
  s"\\\\""g;
  push(@ONESECTION, "@" . &dequote("$_\n"));

  # Horrible magic fudge to cope with the fact that exim_lock has
  # -v and -q options, just like the main program.

  $driver_name = "exim_lock" if /Mailbox maintenance/;
  
  # Similar magic for exiqgrep, which also duplicates options
  
  $driver_name = "exiqgrep" if /Selective queue listing/;  
  }

# .newline must put @* on the end of the previous line, if any, except
# inside a display, where it causes trouble.

elsif (/\.newline/)
  {
  if (@ONESECTION > 0 && ! $indisplay)
    {
    $_ = pop(@ONESECTION);
    s/(\n*)$/\@*$1/;
    push(@ONESECTION, $_);
    }
  }

# .blank turns into @sp, adding 1 if no argument

elsif (/\.blank/)
  {
  s/\.blank\s+(\d+)/\@sp $1/;
  s/\.blank/\@sp 1/;
  push(@ONESECTION, $_);
  }

# .rule turns into a line of hyphens

elsif (/\.rule/)
  {
  push(@ONESECTION, ("-" x ($in_itemize? 68 : 73)) . "\@*\n");
  }

# There's one explicit .tabset setting for two tab stops

elsif (/\.tabset\s*/)
  {
  $rest = $';
  ($first,$second) = $rest =~ /(\d+)em\s+(\d+)em/;
  $tab = ($first * 7)/6;
  $tab2 = $tab + ($second * 7)/6;
  }

# .tabs remembers the first (and only) tab setting

elsif (/\.tabs\s*/)
  {
  $tab = ($' * 7)/6;
  $tab2 = 0;
  }

# .tempindent is used only to align some of the expansion stuff nicely;
# just ignore it. It is used in conjunction with .push/.pop.

elsif (/\.(tempindent|push|pop)\s*/)
  {
  }

# There are some instances of .if ~~sys.fancy in the source. Some of these
# are two-part things, in which case we just keep the non-fancy. For diagrams,
# however, they are in three parts:
#
# .if ~~sys.fancy
# <aspic drawing stuff>
# .elif ~~nothtml
# <ascii art for txt and Texinfo>
# .else
# <HTML instructions for including a gif>
# .fi

elsif (/\.if \~\~sys\.fancy/)
  {
  while (&get_next_line())
    { last if /\.else\b/ || /\.elif\s+\~\~nothtml/ || /\.fi\b/; }

  if (/\.elif/)
    {
    $skip_else = 1;
    }
  }

# There are occasional requirements to do things differently for
# Texinfo/HTML and the PS/txt versions, and there are also some
# HTML-specific things.

elsif (/\.if\s+~~sgcal/ || /\.if\s+~~html/)
  {
  while (&get_next_line()) { last if /\.else\b/ || /\.fi\b/; }
  }

# We may also have Texinfo-specific bits

elsif (/^\.if\s+~~texinfo/)
  {
  $skip_else = 1;
  }

# Ignore any other .if directives

elsif (/\.if/) {}

# Skip else part if flag set

elsif (/\.else/ && $skip_else)
  {
  while (&get_next_line()) { last if /\.fi\b/; }
  $skip_else = 0;
  }

# Ignore other .fi and .else as any .if directives are handled specially

elsif (/\.fi/ || /\.else/) {}

# Ignore .indent

elsif (/\.indent/) {}

# Plain .index goes to @cindex - the "concept" index. Also, there
# are some calls to vindex and findex in the SGCAL source - treated
# as synonymous with .index - which are split into the equivalent
# indexes here.

elsif (/\.(.?)index/)
  {
  $rest = $';
  $letter = ($1 eq "")? "c" : $1;
  tr/./@/;                           # .index -> @index
  
  $rest =~ s/\\\(//g;                # Remove markup
  $rest =~ s/\)\\//g; 
  $rest =~ s/\\%//g;
  $rest =~ s/%\\//g;
  $rest =~ s/\\\*//g;
  $rest =~ s/\*\\//g;    
  $rest =~ s/\\"//g;
  $rest =~ s/"\\//g;
  $rest =~ s/:://g;
  $rest =~ s/\\-/-/g;
  $rest =~ s/-\\//g;
  $rest =~ s/~~//g;     
 
  $rest =~ tr/\\//d;                 # Remove \
   
  $rest =~ s/\@\$/\$/g;              # @$  -> $
  $rest =~ s/\@_/_/g;                # @_  -> _
  $rest =~ s/\@\+/+/g;               # @+  -> +
  $rest =~ s/\$\*\$/\*/g;            # $*$ -> *
  $rest =~ s/\$([^\$]+)\$/\$$1/g;    # $x$ -> $x
   
  $rest =~ s/^\s+//;                 # Remove leading spaces
  $rest =~ s/\s+$//;                 # Remove trailing spaces
  $rest =~ s/\|\|/:/;                # || -> : 
  push(@ONESECTION, "\@${letter}index $rest\n");

  # Duplicate entries for things that were listed as "x see y"

  if (defined $indirections{$rest})
    {
    push(@ONESECTION, "\@${letter}index $indirections{$rest}\n");
    }
  }

# Various flavours of numberpars map to itemize and enumerate.
# Haven't found a way of having a blank space 'bullet' yet, so
# currently using minus.

elsif (/\.numberpars/)
  {
  $rest = $';
  $type = "enumerate";
  $flag = "";

  if    ($rest =~ /\$\./)  { $flag = " \@bullet"; $type = "itemize" }
  elsif ($rest =~ /\" \"/) { $flag = " \@minus";  $type = "itemize"; }
  elsif ($rest =~ /roman/) { $flag = " a"; $type = "enumerate"; }

  push(@ONESECTION, "\n\@$type$flag\n\n\@item\n");
  push(@ENDLIST, $type);
  $in_itemize++;
  }

elsif (/\.nextp/)
  {
  push(@ONESECTION, "\n\@item\n");
  }

elsif (/\.endp/)
  {
  $endname = pop(@ENDLIST);
  push(@ONESECTION, "\@end $endname\n\n");
  $in_itemize--;
  }

# The normal .display (typewriter font) => @example, while the rm
# form goes to @display (no change of font). For Texinfo we need a
# blank line after @display.

elsif (/\.display/)
  {
  $type = /rm/? "display" : "example";
  $asis = 1 if /asis/;
  $indisplay = 1;
  push(@ONESECTION, "\@$type\n\n");
  push(@ENDLIST, $type);
  }

elsif (/\.endd/)
  {
  $asis = 0;
  $indisplay = 0;
  $endname = pop(@ENDLIST);
  push(@ONESECTION, "\@end $endname\n\n");
  }

elsif (/\.conf/)
  {
  ($option, $type, $default) =
    /\.conf\s+(\S+)\s+("(?:[^"]|"")+"|\S+)\s+("(?:[^"]|"")+"|.*)/;

  $option = &dequote($option);

  # If $type ends with $**$ (turned into a dagger for PS version),
  # replace with ", expanded". Remove any surrounding quotes.

  $type =~ s/^"([^"]+)"/$1/;
  $type =~ s/\$\*\*\$/, expanded/;

  # Default may be quoted, and it may also have quotes that are required,
  # if it is a string.

  $default =~ s/^"(.*)"$/$1/;
  $default =~ s/""/"/g;
  $default = &handle_text($default);

  push(@ONESECTION, "\nType: $type\@*\nDefault: $default\n\n");
  }

# Handle .startitems, .enditems, and .item

elsif (/\.startitem/ || /\.enditem/) {}

elsif (/\.item/)
  {
  $arg = $';
  $arg =~ s/^\s*"//;
  $arg =~ s/"\s*$//;
  $arg = &dequote($arg);
  $arg = &handle_text("\\**$arg**\\");

  # If there are two .items in a row, we don't want to put in the
  # separator line.

#  push(@ONESECTION, "\n\@example\n");
  push(@ONESECTION, "\n");
  if (! $lastwasitem)
    {
    push(@ONESECTION, "_" x 75, "\n\n");
    }
#  push(@ONESECTION, "$arg\n\@end example\n\n");
  push(@ONESECTION, "$arg\n\n");
  $new_lastwasitem = 1;
  }

elsif (/\.option/)
  {
  chomp($arg = $');
  $arg =~ s/^\s*//;
  $arg = &dequote("-$arg");
  $arg = &handle_text($arg);
  }

# Texinfo has no facility for emphasis bars.

elsif (/\.em/) {}
elsif (/\.nem/) {}

# Just ignore any .(r)set directives pro tem (or maybe always!)

elsif (/\.r?set/) {}

# Ignore .space, .linelength, and .justify

elsif (/\.space/ || /\.justify/ || /\.linelength/) {}

# Found an SGCAL directive that isn't dealt with. Oh dear.
# Turn the embarrassing characters into question marks and
# output it in an emphasized way.

else
  {
  tr/@{}/???/;
  push(@ONESECTION, "\n\>>>>>>> $_\n") if ! /^\.( |$)/;
  }

$lastwasitem = $new_lastwasitem;
}


##################################################
#             Flush a section                    #
##################################################

# $section_name is the name of the next section.
# $current_section is the name of the one we have buffered up.
# If it is unset, we are at the first section of a chapter.
# $previous_node is the section we last flushed if it was a node.

sub flush_section {

# If there is no text in the section, omit it entirely. However, it
# will have had a pointer set up at the start of the previous section.
# Remember what to replace this with when the chapter gets flushed.

my($skip) = 1;
foreach $s (@ONESECTION)
  {
  if ($s !~ /^(\@cindex|\@section|\s*$)/) { $skip = 0; last }
  }

if ($skip)
  {
  pop @section_list;
  $rewrite{$current_section} = $section_name;
  @ONESECTION = ();
  return;
  }

# There is data in the section: write it out to the chapter file

if ($current_section)
  {
  printf ONECHAPTER ("\@node %s, %s, %s, %s\n",
    &decomma($current_section), &decomma($section_name),
    &decomma($previous_node), &decomma($current_up));
  $previous_node = $current_section;
  while(scalar(@ONESECTION))
    { print ONECHAPTER shift(@ONESECTION); }
  }
else
  {
  while(scalar(@ONESECTION))
    { push(@TOPSECTION, shift(@ONESECTION)); }
  }
}


##################################################
#          Handle a "subsection"                 #
##################################################

# A "subsection" is a set of options that must have their own
# local menu. Do two passes; the first just collects the names
# for the menu. This is called for .conf and .option items.

sub handle_subsection{
my($type) = $_[0];
my($save_up) = $current_up;

$current_up = $current_section? $current_section : $current_chapter;

@sublist = ();
@SUBBUFFER = ();

while (<>)
  {
  last if /^\.end$type/;
  push(@SUBBUFFER, $_);

  # .conf takes the first non-space string as the name, but as there are
  # duplicate confs in various parts of the spec, use the driver name to
  # de-duplicate; .option takes the entire rest of arg as the name, but
  # removes any sequence of ... because this disturbs TexInfo. Also, it
  # turns @- into -.

  if (/^\.$type\s+(\S+)(.*)/)
    {
    if ($type eq "conf")
      {
      $name = &handle_text($1);
      $name .= " ($driver_name)" if ($driver_name ne "");
      }
    else
      {
      chomp($name = &handle_text("-$1$2"));
      $name =~ s/\s*\.\.\.//g;

      $name .= " ($driver_name)" if ($driver_name ne "");

      # There seems to be a major problem in texinfo with the string "--".
      # In the text it gets turned into a single hyphen. This happens if it
      # is used as a menu item, but *not* as a node name. Exim has a command
      # line option "--". With no special action, this appears in the menu
      # as "-", but then the info software complains there is no node called
      # "-". If we triple it in the menu it gets displayed OK, but building
      # software complains about non-existent cross references etc.

      # I have gone for the horrid kludge of turning it into "-<hyphen>"
      # in the menus and nodes.

      # Exim 4 has added --help, which has the same problem.

      $name = "-<hyphen>" if ($name eq "--");
      $name = "-<hyphen>help" if ($name eq "--help");
      }
    push(@sublist, $name);
    }
  }

push (@ONESECTION, "\n\@sp 2\n\@menu\n");
for ($i = 0; $i < scalar(@sublist); $i++)
  {
  $mitem = $sublist[$i];
  $mitem =~ s/\@sc\{([^}]*)\}/\U$1/g;       # Get rid of small caps
  $mitem =~ s/:/<colon>/g;                  # Get rid of colons
  push (@ONESECTION, "* ${mitem}::\n");
  }
push (@ONESECTION, "\@end menu\n\n");

$prevsub = $current_up;
$processing_subsection = 1;
while ($_ = shift(@SUBBUFFER))
  {
  if (/^\.$type\s+(\S+)/)
    {
    $name = shift @sublist;
    $next = (scalar(@sublist))? $sublist[0] : "";
    push @ONESECTION, sprintf("\@node %s, %s, %s, %s\n",
      &decomma($name), &decomma($next), &decomma($prevsub),
      &decomma($current_up));

    if ($name eq "-<hyphen>")    # Fudge for Texinfo
      {
      push(@ONESECTION,
	   "\@findex $name\n",
	   "\@unnumberedsubsec --- option\n");
      push(@ONESECTION,
           "This option consists of two consecutive hyphens. It appears in\n",
           "the menu as \"-<hyphen>\" because otherwise Texinfo gets\n",
           "confused with its cross-referencing.\n");
      }
    elsif ($name eq "-<hyphen>help")    # Fudge for Texinfo
      {
      push(@ONESECTION,
	   "\@findex $name\n",
	   "\@unnumberedsubsec ---help option\n");
      push(@ONESECTION,
           "This option consists of \"help\" preceded by two consecutive\n" .
           "hyphens. It appears in the menu as \"-<hyphen>help\" because\n" .
           "otherwise Texinfo gets confused with its cross-referencing.\n");
      }
    else
      {
      push(@ONESECTION,
	   "\@findex $name\n",
	   "\@unnumberedsubsec $name option\n");
      }

    $prevsub = $name;
    }

  # Then handle as text or directive

  if (substr($_, 0, 1) eq ".")
    { handle_directive(); }
  else
    {
    while($nextline = shift(@SUBBUFFER))
      {
      last if $nextline =~ /^(\.|\s*$)/;
      $_ .= $nextline;
      }
    push(@ONESECTION, handle_text($_));
    $_ = $nextline;
    last if !defined($_);
    redo;
    }
  }

$processing_subsection = 0;
$section_pending = 1;
$current_up = $save_up;
}


##################################################
#            Handle a single chapter             #
##################################################

sub handle_chapter{
chop;
($current_chapter) = /^\.chapter (.*)/;
$current_chapter = &dequote($current_chapter);

$current_chapter = $current_chapter;

my($tmp) = $current_chapter;
$tmp =~ s/\[\[\[\]\]\]/./;
print STDERR "processing chapter: $tmp\n";

# Remember the chapter name for the top-level menu

push(@chapter_list, $current_chapter);

# Open a temporary file to hold the chapter's text while collecting
# all its sections for a chapter-level menu.

$ONECHAPTER = "/tmp/ONECHAPTER.$$";
open(ONECHAPTER, ">$ONECHAPTER") || die "Can't open $ONECHAPTER for writing";

# Initialize for handling sections

@section_list = ();
%rewrite = ();
@ONESECTION = ();
@TOPSECTION = ();
undef $current_section;
undef $next_node;

$processing_subsection = 0;

$previous_node = $current_up = $current_chapter;
$section_pending = 0;

# Handle the .chapter directive as the first text of a section without
# a section title.

handle_directive();

# Loop, handling each section. Assume they are sufficiently short that
# we can buffer the text in store, in an array called ONESECTION, instead
# of thrashing yet another file.

while (<>)
  {
  last if /^\.chapter /;

  # Handle a new section, preserving $_ (handle_text flattens it).
  # It seems we cannot get a fullstop into a Texinfo node name; use a magic
  # character string that gets turned back into a dot by the post-processing.

  if (/^\.section\s+/)
    {
    $save = $_;
    $section_name = $';
    $section_name =~ s/(\s|\n)+$//;
    $section_name =~ s/://;
    $section_name = &handle_text($section_name);
    flush_section();
    push(@section_list, $section_name);
    $current_section = $section_name;
    $next_node = $section_name if !$next_node;
    $section_pending = 0;
    $_ = $save;
    }

  # The .startconf macro introduces a set of .conf's which must have
  # their own local set of menus. Suspend processing the section while
  # we sort out the menu and copy their data. This is all done in a
  # subroutine that is shared with options.

  elsif (/^\.startconf\s+(.*)/)
    {
    $confuse = $1; 
    $confuse = &dequote($confuse); 
    handle_subsection("conf");
    next;
    }

  elsif (/^\.startoption/)
    {
    handle_subsection("option");
    next;
    }

  # Deal with the actual data lines; if there's a section pending
  # start a new section on hitting some text. We hope this happens
  # only once per chapter...

  if (substr($_, 0, 1) eq ".")
    {
    handle_directive();
    }
  else
    {
    while($nextline = <>)
      {
      last if $nextline =~ /^(\.|\s*$)/;
      $_ .= $nextline;
      }
    if ($section_pending && !/^\s*$/)
      {
      $section_name = (defined $current_section)?
        "$current_section (continued)" :
        "$current_chapter (continued)" ;
      flush_section();
      push(@section_list, $section_name);
      $current_section = $section_name;
      $next_node = $section_name if !$next_node;
      $section_pending = 0;
      }

    push(@ONESECTION, handle_text($_));
    $_ = $nextline;
    last if !defined($_);
    redo;
    }
  }

# Flush any pending text, making its next field null.
# and fudging section_name for the final section of the previous.

$section_name = "";
flush_section();

# Set up section name as the start of the next chapter

$section_name = "Concept Index" if (!$doing_filter);

if (defined $_ && /^\.chapter (.*)/)
  {
  $section_name = $1;
  $section_name = &dequote($section_name);
  }
$next_node = $section_name;

# Write out the chapter to the CHAPTERS file, sticking the chapter
# menu after the text that came before the first section heading. This
# will always at least contain the chapter title.

printf CHAPTERS ("\@node %s, %s, %s, Top\n",
  &decomma($current_chapter), &decomma($next_node),
  &decomma($previous_chapter));

# The pre-section stuff; if we hit an @end menu line, it is the menu of
# a "subsection" before the first section. In that case, we need to put
# the chapter's menu one the end of it, and then resume with the rest of
# the TOPSECTION data afterwards. We also need to thread together this
# "subsection"s nodes because they are all at the same level under the
# chapter.

$in_menu = 0;
while(scalar(@TOPSECTION))
  {
  $s = shift(@TOPSECTION);
  if ($s =~ /^\@end menu/)
    {
    $in_menu = 1;
    last;
    }
  print CHAPTERS $s;
  }

# Menu for sections

undef $next_actual_section;
undef $point_back;

if (scalar(@section_list))
  {
  print CHAPTERS "\n\@sp 2\n\@menu\n" if ! $in_menu;
  $next_actual_section = $section_list[0];
  for ($i = 0; $i < scalar(@section_list); $i++)
    {
    $section_name = $section_list[$i];
    $section_name =~ s/\@sc\{([^}]*)\}/\U$1/g;
    print CHAPTERS "* ${section_name}::\n";
    }
  $in_menu = 1;
  }
print CHAPTERS "\@end menu\n\n" if $in_menu;

# Remainder of topsection; we must arrange that the final @node in
# it (which will have a blank "next" field) actually points on to
# the next section, if any. If this happens, then the next section
# must point back to the final @node.

while(scalar(@TOPSECTION))
  {
  $s = shift(@TOPSECTION);
  if ($next_actual_section && $s =~
         /^\@node\s+([^,]+),\s*,\s*([^,]*),\s*(.*)/)
    {
    my($t1, $t2, $t3) = ($1, $2, $3);    # So can be decomma'd
    printf CHAPTERS ("\@node %s, %s, %s, %s\n", &decomma($t1),
      &decomma($next_actual_section), &decomma($t2), &decomma($t3));
    $point_back = $1;
    }
  else { print CHAPTERS $s; }
  }

close(ONECHAPTER);
open(ONECHAPTER, "$ONECHAPTER") || die "Can't open $ONECHAPTER for reading";

# While copying the chapter data, check for node references to empty
# sections that got omitted and correct them, and correct the prev pointer
# in the first node if necessary.

while ($buff = <ONECHAPTER>)
  {
  foreach $key (keys %rewrite)
    {
    $buff =~ s/$key/$rewrite{$key}/;
    }
  if ($point_back && $buff =~ /^\@node\s+([^,]+),\s*([^,]*),\s*([^,]*),\s*(.*)/)
    {
    my($t1, $t2, $t4) = ($1, $2, $4);   # so can be decomma'd
    printf CHAPTERS ("\@node %s, %s, %s, %s\n", &decomma($t1),
      &decomma($t2), &decomma($point_back), &decomma($t4));
    undef $point_back;
    }
  else { print CHAPTERS $buff; }
  }

$previous_chapter = $current_chapter;

close(ONECHAPTER);
unlink($ONECHAPTER);
}


##################################################
#                Main Program                    #
##################################################

# This is a two-pass algorithm. The first pass goes through and gets the
# variable names for cross references. The second pass does the real work,
# but we can't just read through doing the translation in one pass. We need
# to know the list of chapters in order to build a top-level menu, and for
# each chapter we need to know the sections in order to build a section
# menu. Consequently, make use of temporary files to buffer things.

# This script is used for the filter document and the overview as well;
# flags tell it if it is doing one of them.

$doing_filter = 0;
$skip_else = 0;
$in_itemize = 0;
$lastwasitem = 0;
$confuse = "";

$chapter_number = 0;
$section_number = 0;

if ($#ARGV >= 0 && $ARGV[0] eq "-filter")
  {
  $doing_filter = 1;
  shift @ARGV;
  }

# First pass: Just fish out variable settings. Save the arguments so that
# they can be reinstated for a second pass.

print STDERR "Scanning for references\n";
@save_argv = @ARGV;

# Pick up any .set directives right at the very start

while (<>)
  {
  last if ! /^\.set\s+(\S+)\s+(.+)$/;
  $name = $1;
  $value = $2;
  $value =~ s/^\"?(.*?)\"?\s*$/$1/;
  $references{$name} = $value;
  }

# Now skip everything before the first .chapter except for
# .index lines that set up indirections. Save these so that
# the relevant index entries can be duplicated.

while (<>)
  {
  if (/^\.chapter\s+(.+)$/)
    {
    $chapter_number++;
    $section_number = 0;
    $current_chapter = $1;
    $current_chapter = $current_chapter;
    $current_section = "";
    last;
    }

  if (/^\.index\s+([^\$]+)\s+\$it\{see\s+([^}]+)\}\s*$/)
    {
    $indirections{"$2"} = $1;
    }
  }

# Do the business

while (<>)
  {
  if (/^\.chapter\s+(.+)$/)
    {
    $current_chapter = $1;
    $current_chapter = &dequote($current_chapter);
    $current_section = "";
    }
  elsif (/^\.section\s+(.+)$/)
    {
    $current_section = $1;
    $current_section = &dequote($current_section);
    $current_section =~ s/://;
    }
  elsif (/^\.r?set\s+(\S+)\s+(.+)$/ && $1 ne "runningfoot")
    {
    $name = $1;
    $value = $2;

    # Only set the first time. This handles a few special cases in part2
    # which is included in the filter text as well.

    if (!exists($references{$name}))
      {
      $value =~ s/^\"?(.*?)\"?\s*$/$1/;
      $value =~ s/~~chapter\./~~chapter****/;
      $value =~ s/~~chapter/$current_chapter/;
      $value =~ s/~~section/$current_section/;
      $references{$name} = $value;
      }
    }
  }

$final_chapter = defined($current_chapter)? $current_chapter : "";

# Reinstate ARGV with the list of files and proceed to the main pass

@ARGV = @save_argv;

# $asis is set true when processing .display asis blocks, to stop
# characters getting interpreted.

$asis = 0;

# $indisplay is set true while processing .display blocks, to stop
# .newlines being handled therein (adding @* wrecks alignment)

$indisplay = 0;

# $tab is set to the value of the tab stop - only one stop is ever used
# in the Exim source.

$tab = 0;

# Current driver name, for disambiguating nodes

$driver_name = "";

# $section_pending is set if a new section is to be started on hitting
# any data lines.

$section_pending = 0;

# Open a file to buffer up the entire set of chapters

$CHAPTERS = "/tmp/CHAPTERS.$$";
open(CHAPTERS, ">$CHAPTERS") || die "Can't open $CHAPTERS for writing";

# Skip everything before the first .chapter

while (<>) { last if /^\.chapter /; }

# Loop, handling each chapter

$current_up = "";
$previous_chapter = "Top";
$previous_node = "Top";

$chapter_number = 0;
$section_number = 0;

while (defined ($_) && /^\.chapter /)
  {
  handle_chapter();
  }

# Output the stuff at the start of the file

print "\\input texinfo\n";

print "\@set{wmYear} 2003\n";
print "\@set{wmAuthor} Philip Hazel\n";
print "\@set{wmAuthor_email} <ph10\@\@cus.cam.ac.uk>\n";
print "\@set{COPYRIGHT1} Copyright \@copyright{} \@value{wmYear} University of Cambridge\n";

print "\@c %**start of header\n";

if (!$doing_filter)
  {
  print "\@setfilename spec.info\n";
  print "\@settitle Exim Specification\n";
  }
else
  {
  print "\@setfilename filter.info\n";
  print "\@settitle Exim Filter Specification\n";
  }

print "\@paragraphindent 0\n";
print "\@c %**end of header\n\n";


print "\@titlepage\n";
print "\@title The Exim Mail Transfer Agent\n";
print "\@author \@value{wmAuthor}\n";

print "\@page\n";
print "\@vskip 0pt plus 1filll\n";

print "Permission is granted to make and distribute verbatim copies of this manual provided the\n";
print "copyright notice and this permission notice are preserved on all copies.\n";

print "\@sp2\n";
print "\@value{COPYRIGHT1}\@*\n";

print "\@end titlepage\n\n";

# Output the top-level node and its introductory blurb

print "\@node       Top,       $chapter_list[0], (dir), (dir)\n";
print "\@top\n";

if (!$doing_filter)
{
print <<End;
The Exim Mail Transfer Agent\@*
****************************

The specification of the Exim Mail Transfer Agent is converted mechanically
into Texinfo format from its original marked-up source. Some typographic
representations are changed, chapters and sections cannot be numbered, and
Texinfo lacks the ability to mark updated parts of the specification with
change bars.

Because the chapters and sections are unnumbered, cross references are set to
their names. This makes the English a bit odd, with phrases like \`see chapter
\"Retry configuration\"\' but it seemed very cumbersome to change this to \`see
the chapter entitled \"Retry configuration\"\' each time.

Each chapter, section, and configuration option has been placed in a separate
Texinfo node. Texinfo doesn\'t allow commas, colons, or apostrophes in node
names, which is a rather nasty restriction. I have arranged not to use colons
or apostrophes in section titles, but cannot bring myself to omit them from
titles such as \"The foo, bar and baz commands\". For the corresponding node
names I have just used multiple occurrences of \"and\", though it looks very
ugly.

If a chapter or section continues after a list of configuration options that is
not in a new section, a new node is started, using the chapter\'s or section\'s
name plus \`(continued)\'. The \`Up\' operation from a section or configuration
option returns to the start of the current chapter; the \`Up\' operation at a
chapter start returns to the top of the document; the \`Up\' in a list of
configuration options within a section returns to the top of that section.

A number of drivers have options with the same name, so they have been
disambiguated by adding the name of the driver to its option names in order to
create node names. Thus, for example, the specification of the \`command\'
options of the \`lmtp\' and \`pipe\' transports are in nodes called \`command
(lmtp)\' and \`command (pipe)\', respectively.

End
}

else
{
print <<End;
Filtering with the Exim Mail Transfer Agent\@*
*******************************************

The specifications of the Exim Mail Transfer Agent\'s filtering facility is
converted mechanically into Texinfo format from its original marked-up source.
Some typographic representations are changed, chapters and sections cannot be
numbered, and Texinfo lacks the ability to mark updated parts of the
specification with change bars.

Because the chapters and sections are unnumbered, cross references are set to
their names. This makes the English a bit odd, with phrases like \`see section
\"Multiple personal mailboxes\"\' but it seemed very cumbersome to change this to
\`see the section entitled \"Multiple personal mailboxes\"\' each time.

End
}

# Output the top-level menu

print "\@menu\n";

while (scalar(@chapter_list))
  {
  $name = &decomma(shift(@chapter_list));
  print "* ${name}::\n";
  }
print "* Concept Index::\n" if (!$doing_filter);
print "\@end menu\n\n";

# Copy the chapters, then delete the temporary file

close(CHAPTERS);
open(CHAPTERS, "$CHAPTERS") || die "Can't open $CHAPTERS for reading";
print $buff while($buff = <CHAPTERS>);
close(CHAPTERS);
unlink($CHAPTERS);

# Output the finishing off stuff

if (!$doing_filter)
  {
  print "\@node Concept Index, , $final_chapter, Top\n";
  print "\@chapter Concept Index\n\@printindex cp\n";
  print "\@chapter Function Index\n\@printindex fn\n";
  }
print "\@contents\n";
print "\@bye\n";

# End