% $StyleId: cwebbase.doc,v 1.4 1995/11/07 17:55:19 schrod Exp $
%----------------------------------------------------------------------
% Written by Joachim Schrod <schrod@iti.informatik.th-darmstadt.de>.
% Copyright conditions see below (GPL).

%
% LaTeX macros cwebbase
% support for incorporation of CWEAVE markup in LaTeX documents
%
% [LaTeX in MAKEPROG]
% (history at end)




\documentclass{progltx}

\usepackage{cweb-doc}		% document-specific markup
\usepackage{fullpage}


\RCS $StyleRevision: 1.4 $
\RCS $StyleDate: 1995/11/07 17:55:19 $


\begin{document}


\title{Typesetting \cweave{} Output}
\author{%			% LaTeX does not discard unnecessary glue...
    Joachim Schrod%
    \thanks{%
	\protect\raggedright
	TU~Darmstadt, Computer Science Department, WG Systems Programming,
	Alexanderstr.~10, D-64283~Darmstadt, Germany.
	Email: \texttt{schrod@iti.informatik.th-darmstadt.de}.%
	}%
    }
\date{%
    Revision \RCSStyleRevision\\
    (as of \RCSStyleDate)%
    }

\maketitle


% doesn't work with progltx yet
%\tableofcontents




\chap Introduction.

This code shall interpret and render the tag set output by \cweave{}.
That's realized in this module as it's needed in multiple places --
when we want to use \LaTeX{} markup in \cweb{}, when we want to
incorporate \cweb{} pieces in \LaTeX{} documents (as examples, or in
textbooks), and when we want to incorporate whole \cweb{} documents in
a \LaTeX{} document. (That's by far the most difficult task.)

As \cweb{} operates in modes, for obvious reasons our implementation
is based on a finite state automaton (FSA). The whole tag set is
separated into subsets that are valid in certain states.

The complete set of tags does not obey any name conventions, some tags
are even in conflict with standard \LaTeX{} tags (e.g.\ |\\|). Most
tags are used to render code pieces. With the introduction of states
we can restrict the visibilities of these tags to those states where
we typeset program parts. The needed functionality will be available
in cseqs of our namespace, we'll map that to \cweave{}'s names in a
group.

Then we have to implement the rendering of chunks and sections, that's
(more or less) a straight-forward task. Actually, we only set up the
basic code pieces, the integration of section divisions in the actual
\LaTeX{} document is a task of higher-level code; that integration
depends heavily on the type of document. That also implies that we
won't implement front or back matter material (document markup, table
of contents, index, list of refinements, etc.)\ in this module.


\sect Let's identify this module.

\noindent The code below is explained in the implementation
documentation of the \pkg{rcs} package.

\beginprog
\begingroup
    \def\RCSFile#1#2 $#3: #4 #5\endRCS $#6: #7 #8\endRCS{%
	\def\date{#4}\def\id{v#7}%
	\ProvidesFile{#1}[\date\space\id\space #2]%
	}

  \RCSFile{cwebbase.tex}{CWEAVE tags for LaTeX markup}
  $StyleDate: 1995/11/07 17:55:19 $: 9999/00/00 \endRCS
  $StyleRevision: 1.4 $: 0.0 \endRCS
\endgroup
\endprog


\sect This module is part of the supported \textsl{cweb-sty} bundle.
Send bug reports, comments and repairs.

It assumes that the canonical setup of my modules has happened and can
be used herein. (In particular, the underscore is a letter.)


\sect This module reserves the namespace |cwbb|. It assumes that
|\cweb_warning| is bound either to |\PackageWarningNoLine| or
|\ClassWarningNoLine|, depending on the environment.

\beginprog
\ifx \cwbb@loaded\undefined
    \def\cwbb@loaded{$StyleRevision: 1.4 $}
\else
    \cweb_warning{cwebbase}%
        {Some other package already uses namespace `cwbb'}
\fi
\endprog


\sect This is freely distributable software; you can redistribute it
and/or modify it under the terms of the GNU General Public License as
published by the Free Software Foundation; either version~2 of the
License, or (at your option) any later version.

This software is distributed in the hope that it will be useful, but
\textbf{without any warranty}; without even the implied warranty of
\textbf{merchantability} or \textbf{fitness for a particular purpose}.
See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License in
the file |License| along with this package; if not, write to the Free
Software Foundation, Inc., 675~Mass Ave, Cambridge, MA~02139,~USA.




% ------------------------------------------------------------
%
% subdocument: The interface between CWEAVE and TeX
%

\input{cweave-spec}

%
% ------------------------------------------------------------




\chap Interface of \pkg{cwebbase} module.

\pkg{cwebbase} implements most of \cweave{} tags. It sets up tons of
redefinable cseqs, all of them start with |Cweb|.
%
\begin{fixme}
  How do I document them? We need a configuration document anyhow. Do
  I present here just a list of those cseqs realized in this module
  and refer to the config docs? Need to wait until I have it, then
  we'll see how the stuff here will be added.
\end{fixme}


\sect This module supports both hierarchic and flat section
structuring. You have to bind |\cweb@structure| to 0 (hierarchic) or 1
(flat) before importing it, to show which structure is wanted.


\sect This module supports suppression of different document parts.
You have to define flags |@cweb@suppress@|\<foo>|@| with
$\mbox{\<foo>} \in \{ |changehints|, |unchanged|, |format| \}$ to
denote if suppression shall happen.


\sect The counter |secnumdepth| must be~11 if chunks shall be
numbered. That counter doesn't need to be set when this module is
imported, it can be set locally during typesetting \cweb{} code.

The definitions for flat structure assume that |secnumdepth| is~11
during typesetting.


\sect |\part|, |\chapter| (if available), and |\section| are used for
\cweb{} section divisions (those with the lowest ranks). These
commands will change the respective counters globally and will
add toc lines and marks. You might want to prevent that by redefining
|\addcontentsline| and |\mark|.

Table of contents macros are not defined for the section divisions
introduced by this module. You must either prevent that they are
written to the |AUX| file at all; or you must define them.
%
\begin{fixme}
  Add explanation of introduced section divisions.
\end{fixme}

For flat structure, all section divisions are of type `|chunk|', like
chunks. The section names start with `|\cwbbRank{|\<rank>|}|'. By
default, |\cwbbRank| takes one argument and expands to the empty list
-- this can be redefined for the table of contents.


\sect The document end, as explained in the \cweave{} interface specs
(chunk~\ref{spec:sec:docend}), is not implemented. That must be done
by the user of this module -- if the back matter is to be processed at
all.




\chap Processing states.

We have to typeset five different categories of material:
Documentation, program pieces---embedded within the documentation and
as large chunks, \TeX{} material within program pieces (i.e., comments
and refinement names), and cross reference information. Since we need
a complete other environment for the program pieces than for the rest
we design ``states'' where we switch to appropriately.
%
\begin{enumerate}

\item A chunk starts in the \textsl{documentation state}.

\item |\B| switches to \textsl{program state}. This can happen in
documentation and program state.

\item While we process the argument of |\PB| we're in \textsl{restricted
program state}; |\PB| may appear in documentation and in \TeX{}
state. Since \TeX{} state can be switched on within restricted
program state, |\PB| can appear within the argument of itself.

\item In the arguments of |\C|, |\SHC|, and |\X| we switch to
\textsl{\TeX{} state}. All these cseqs appear only in (restricted)
program state, their official names are actually different. I.e., only
in (restricted) program state these cseq are bound to the meaning
described here.

\item Cross reference information are attached to most chunks with
refinements. This information is processed in \textsl{CR state}.
After CR state material
comes always the next chunk or the document end, i.e., material in
documentation state.

\end{enumerate}
%
This FSA is illustrated by the diagram in figure~\ref{fig:fsa}.

\begin{figure}

\begin{center}
\chardef\\=`\\
\input{cweb-fsa.latex}
\end{center}

\caption{The processing state's FSA. The automaton starts and ends in
the ``documentation'' state.}
\label{fig:fsa}

\end{figure}

|\cwbb@documentation| will switch to documentation state,
|\cwbb@program| to program state, |\cwbb@Rprogram| to restricted
program state, |\cwbb@tex| to \TeX{} state, and |\cwbb@CR| to CR state.
If we're already in a state, the switch to this state shall be a
permissible null operation.


\sect The basic difference between these states can be named with
two parameters: (1)~The cseq bindings in effect and (2)~the layout
parameters used for paragraph makeup.

In (restricted) program state and in CR state the text is output
under the control of \cweave{}, and tagged by \cweave{}. The used
tags are from a global namespace and should only be in effect during
these states. We call this tag set the \textsl{\cweave{} bindings}. In the
other two states the tags are largely defined by the user, the tag
set is called the \textsl{user bindings}. The switch to another binding
is always done locally, i.e., if we switch from documentation to
restricted program state within a group we don't have to bother about
the restauration of the user binding; it will be done automatically
by \TeX{} at the end of the group. Nevertheless we must be able to
switch from the \cweave{} bindings back to the user bindings which
were in effect when we activated the \cweave{} bindings. This is
needed for the \TeX{} state which is always activated within
(restricted) program state.

The parameters for \textsl{program layout} are really special ones
since they need to support the indentation which shows the program
structure. These parameters are used in program and in \TeX{} state.
The \textsl{document layout} parameters established by the user are
used in the other three states.

The following table shall summarize this. $C$~denotes \cweave{}
bindings, $U$~user bindings, $P$~program layout, and $D$~document
layout. If an entry is empty, its value is not changed on entry in
this state.
%
\begin{center}
  \DeleteShortVerb\|
\begin{tabular}{r|c|c}
\multicolumn1{c|}{\textsc{State}} & \textsc{Binding} & \textsc{Layout} \\[1pt]
\hline
documentation & $U$ & $D$ \\
program & $C$ & $P$ \\
restricted program & $C$ & \\
\TeX{} & $U$ & \\
CR & $C$ & $D$ \\
\end{tabular}
  \MakeShortVerb\|
\end{center}


\sect Since the user bindings and the document layout is defined
initially, we don't have to do anything if it's requested. Only if we
change it, i.e., within |\cwbb@cweave_bindings| and
|\cwbb@program_layout|, we redefine |\cwbb@user_bindings| and
|\cwbb@doc_layout|. If they are eventually executed, they shall
rebind themselves back to |\relax|. This way we can switch to
documentation state as often as we want.

\beginprog
\let\cwbb@user_bindings=\relax
\let\cwbb@doc_layout=\relax

\def\cwbb@documentation{%
    \cwbb@user_bindings
    \cwbb@doc_layout
    }
\def\cwbb@Rprogram{\cwbb@cweave_bindings}

\def\cwbb@program{%
    \cwbb@cweave_bindings
    \cwbb@program_layout
    }
\def\cwbb@tex{\cwbb@user_bindings}

\def\cwbb@CR{%
    \cwbb@cweave_bindings
    \cwbb@doc_layout
    }
\endprog




\chap Saving and restoring control sequences.

We have a lot of cseqs which are defined within the namespace of this
module and which will be used with other names. This usage is in a
controlled environment, namely neither in documentation nor in \TeX{}
state. (I.e., the text processed is tagged by \cweave{}, not by humans;
therefore we have a precise specification of the cseqs we have to
accept.) We cannot work with groups where a cseq is just redefined
and \TeX{} takes care for establishing the old binding again; when we
switch from program state to \TeX{} state all the bindings which were
in effect before the program state get active, i.e., in the
documentation state, must be in effect again. We cannot simply
consider the \TeX{} state as something parallel to program state, it
must be a hierarchical relationship: In the program state values are
set up which must be available after switching back from \TeX{} to
program state.

We save the current binding of a cseq in another cseq, but only if
there exists a binding currently. This is done to save valuable
\TeX{} main memory. (Why don't all those people who talk about
extending \TeX{} not just implement dynamic memory management for the
existing \TeX{}? It would still be \TeX{} and would \emph{really} be
a worthy activity. But not enough fun, I assume\,\dots) Actually, one
can assume that nearly no cseq bindings must be saved at all---the
used names are strange enough. The bindings of |\foo|, i.e., of the
cseq with the name \fbox{\ttfamily foo}, is saved as the cseq with the
name \fbox{\ttfamily cweb\char `\\ foo}, i.e., as
|\csname cweb\string\foo \endcsname|.


\sect The save process is not done statically, but by the macro
|\cwbb@rebind| which interprets a list of tuples $({\it old\_name},
{\it new\_name})$, terminated by the tuple $(|\stop|, |\stop|)$.
Eventually it constructs two new lists, |\cwbb@to_restore| with the
cseqs which had a binding, and |\cwbb@undefined| with the names which
didn't have one.

The saving is actually done by |\cwbb@save_binding|, |\cwbb@rebind|
is responsible for the initialization, |\cwbb@do_rebind| for the
effective rebinding and the tail recursion on the list.

\begin{fixme}
  We could pull the |\next| assignment in the |\else| branch out of
  the loop to get a better performance. Should measure if this is of
  interest.
\end{fixme}

\beginprog
\newtoks\cwbb@undefined
\newtoks\cwbb@to_restore

\def\cwbb@rebind{%
    \cwbb@undefined{}%
    \cwbb@to_restore{}%
    \cwbb@do_rebind
    }
\def\cwbb@do_rebind#1#2{%
    \ifx #1\stop
	\let\next\relax
    \else
	\cwbb@save_binding #2%
	\let #2=#1%
	\let\next\cwbb@do_rebind
    \fi
    \next
    }
\endprog


\sect If the cseq to be saved is undefined, it may just be added to
the ``undefined list.'' Otherwise its binding is saved and it's added
to the ``to-be-restored list.''

\TeX{}nical note: The cseq-name for the saved binding must be created
before the |\let| is executed.

\beginprog
\def\cwbb@save_binding#1{%
    \ifx #1\undefined
	\cwbb@undefined \expandafter{\the\cwbb@undefined #1}%
    \else
	\expandafter\let \csname cweb\string#1\endcsname =#1%
	\cwbb@to_restore \expandafter{\the\cwbb@to_restore #1}%
    \fi
    }
\endprog


\sect The restoration of rebound cseqs is a two-tied activity: All
previously undefined cseqs must be made undefined again, and all saved
cseqs must be restored. Actually, we don't need to reset the two token
lists, but we do it to save space.

Both |\cwbb@undefine| and |\cwbb@restore_binding| iterate over a list
of cseqs terminated by |\stop|.

\begin{fixme}
  And here the |\next| assignment could be prepended to the loop as
  well.
\end{fixme}

\beginprog
\def\cwbb@restore_bindings{%
    \expandafter\cwbb@undefine \the\cwbb@undefined \stop
    \cwbb@undefined{}%
    \expandafter\cwbb@restore_binding \the\cwbb@to_restore \stop
    \cwbb@to_restore{}%
    }

\def\cwbb@undefine#1{%
    \ifx #1\stop
	\let\next\relax
    \else
	\let#1\undefined
	\let\next\cwbb@undefine
    \fi
    \next
    }
\endprog


\sect \TeX{}nical note: As in |\cwbb@save_binding|, the cseq-name for
the saved binding must be created before the |\let|s are executed.

\begin{fixme}
  Another |\next| assignment.
\end{fixme}

\beginprog
\def\cwbb@restore_binding#1{%
    \ifx #1\stop
	\let\next\relax
    \else
	\expandafter\let \expandafter#1\csname cweb\string#1\endcsname
	\expandafter\let \csname cweb\string#1\endcsname \undefined
	\let\next\cwbb@restore_binding
    \fi
    \next
    }
\endprog




\chap Chunks.

All chunks are numbered subsequently. The chunk number is supplied as
part of the markup. In addition, the `changed' tag |\*| may be
appended to the chunk number.

By default, chunks have approximately two picas vertical space in
front.

If we use an hierarchic structure, we output the chunk number at the
refinement. (It's only needed to identify the refinement for cross
references.) But chunks are important structural units and thus should
be marked by typographic means that's more visible than the space
mentioned above, we add a `chunk start marker'.

For a flat structure, we output the chunk number as the chunk start
marker, that is the traditional \WEB{} layout.

In any case, the documentation part shall be run in after the chunk
start marker.

\beginprog
\newskip\CwebChunkPreSkip
	\CwebChunkPreSkip=2pc plus 1pc minus 6pt
\endprog


\sect In \LaTeX{}, structural divisions are best realized by
|\@startsection|, we'll use it also for chunk starts. The chunk start
marker is the section heading. For hierarchic structure, we use a
traditional paragraph sign, used since medieval times to tag new text
units. Our first test showed that the paragraph sign has typically an
underlength, that doesn't look good at the start of a chunk. We raise
the sign by its depth.

For flat structure, we don't have text as the section heading,
but we have to cope for a deficiency of \LaTeX{}: It always leaves a
quad after the section number (by the way, this is not mentioned in
the documentation---you have to look in the macros to figure this
out). So we use a backskip of $\rm -1\,em$ as the chunk start marker
then.

Changed chunks are marked by a changeflag, by default a star lapping
to the right. We need the information if that chunk is changed to
decide if we shall output the changeflag. We demand from the chunk
macros that they record that info in the flag |@cwbb_changed_chunk@|.

Each structural division must be categorized, it belongs to a
\textsl{type}. This type determines also the name of the counter, the
mark handling for this category, and how this section tag is
represented in the table of contents. The type of chunks is `|chunk|'
(trivial, isn't it?).

Furthermore, a structural division in \LaTeX{} has an associated
\textsl{level}. That level is used to determine if that division get
numbered in text or table of contents. For sections, we'll compute the
level from the rank, it will not be larger than the rank.%
    \footnote{Check the next section for a more precise definition of
	`rank' and `level'.}
Since the highest section rank is~10, we may use 11 as chunk level.

For flat structure, we output the number for all section headings,
including chunks. For hierarchic structure we use the default maximum
level as supplied by our document class (e.g., 3~for \cls{article},
and 2~for \cls{report} or \cls{book}).

As a last point, we have to care already for sections in flat
structure. They will use the |chunk| category as well, we have to
distinguish them in the table of contents. For that, we'll record the
level in the text, it will get written to the table of contents.

\beginprog
\newif\if@cwbb_changed_chunk@
\let\CwebChangeFlag=*
\if@cweb@suppress@changehints@
    \let\cwbb@change_flag=\relax
\else
    \let\cwbb@change_flag=\CwebChangeFlag
\fi
\def\cwbb@lap_change_flag{\rlap{\cwbb@change_flag}}

\ifcase \cweb@structure
    %% hierarchic
    \let\CwebChunkHeadingStyle\relax
    \newbox\cweb@chunk_marker
    \setbox\cweb@chunk_marker=\hbox{\P}	% tradition from medivial times
    \setbox\cweb@chunk_marker=\hbox{%
	\raise\dp\cweb@chunk_marker \box\cweb@chunk_marker
	}
    \def\CwebChunkHeadingMarker{\copy\cweb@chunk_marker}
    \def\CwebChunkHeading#1{%
	\CwebChunkHeadingMarker
	\if@cwbb_changed_chunk@ \cwbb@lap_change_flag \fi
	}
  \or
    %% flat
    \def\cwbbRank#1{}
    \def\CwebChunkHeadingStyle{\bfseries}
    \def\CwebChunkHeading#1{%
	\protect\cwbbRank{#1}%
	\kern -1em
	\if@cwbb_changed_chunk@ \cwbb@lap_change_flag \fi
	}
\fi
\endprog


\sect Chunks are tagged with |\M| and get passed the chunk number as
argument. Before we evaluate |\@startsection|, we'll start the chunk.
That's common code to be used by sections as well, as they also
implicitely start a chunk.

\beginprog
\def\M#1{%
    \cwbb@start_chunk{#1}%
    \@startsection
	{chunk}%		% category, == counter etc.
	{11}%			% level
	{\z@}%			% no indentation for heading
	{\CwebChunkPreSkip}%	% skip above heading
	{\m@ne em}%		% run-in heading, 1em distance to text
	{\normalfont\CwebChunkHeadingStyle}%	% layout
	{\CwebChunkHeading{11}}% % text, level might go to toc file
    }
\endprog


\sect But let's first declare how a chunk number is represented. The
principle rendering is controlled by |\thechunk|, namely, we get
arabic numbers.  This rendering is used both for creation of the
section header (if in flat mode) and for cross references. I.e.,
|\ref| expands into it.

But for flat structure we need different expansions: In the section
header a period shall be appended, a cross reference shall appear
without such a period. We use two different macros: |\thechunk|
creates the number with the period, |\cwbb@refchunk| only the number.
The latter is used for a redefinition of |\@currentlabel|. Of course,
for hierarchic structure, both are the same.

\beginprog
\newcounter{chunk}

\def\cwbb@refchunk{\arabic{chunk}}

\ifcase \cweb@structure
    %% hierarchic
    \let\thechunk=\cwbb@refchunk
  \or
    %% flat
    \def\thechunk{\arabic{chunk}.}
\fi
\endprog


\sect Each chunk is started by |\cwbb@start_chunk|, which does the
common actions. First of all, we switch to documentation state.

Then we store the chunk number in the |chunk| counter. That implies
that the chunk start must not step this counter! We'll take care for
that later by redefining |\refstepcounter|. The actual number is
passed as the argument, including an optional |\*| used to denote
changed chunks. We'll redefine that cseq to update the flag
|@cwbb_changed_chunk@|.

Each chunk must be typeset with an open |\if| since it is finished by
|\fi|. If we shall suppress the output of unchanged chunks, we have
to use |\iffalse| if not |@cwbb_changed_chunk@|, otherwise |\iftrue|.
The flag |@cwbb_print_chunk@| is set up to provide an appropriate
test.

When we typeset a chunk, we evaluate |\cwbb@init_print_chunk|. For
hierarchic structure, that cseq will be used to initialize the
typesetting of the chunk number at program parts.

\beginprog
\if@cweb@suppress@unchanged@
    \def\if@cwbb_print_chunk@{\if@cwbb_changed_chunk@}
    \@gobble\fi				% close \if in case it's skipped
\else
    \let\if@cwbb_print_chunk@=\iftrue
    \@gobble\fi				% same here
\fi

\def\cwbb@start_chunk#1{%
    \cwbb@documentation
    \begingroup
        \def\*{\global\@cwbb_changed_chunk@true}%
        \global\@cwbb_changed_chunk@false
        \global \c@chunk #1\relax	% might expand \*
    \endgroup
    \if@cwbb_print_chunk@
        \cwbb@init_print_chunk
    }
\endprog


\sect |\@startsection| will step the section counter (i.e., |chunk|).
Since we just assigned it in |\cwbb@start_chunk|, that would leave us
with a counter that's one too high. And our output in cross references
may be wrong, as a period is appended in flat structure.

That damage all comes from one macro, |\refstepcounter|; let's
advice it to do the Right Thing for |chunk| counters.

\beginprog
\let\cwbb@refstepcounter=\refstepcounter

\def\cwbb@string@chunk{chunk}
\def\refstepcounter#1{%
    \def\@tempa{#1}%
    \ifx \@tempa\cwbb@string@chunk
        \protected@edef\@currentlabel{\cwbb@refchunk}%
    \else
    	\cwbb@refstepcounter{#1}%
    \fi
    }
\endprog




\chap Sections.

Important sections are started on a new page, normal sections have the
same space as chunks in front. The very first section does not
automatically start on a new page since a title may be in front of it.

Each section division has a \textsl{rank}, that denotes the section
depth as seen by \cweave{}.

For this depth we compute a \textsl{level}, used to denote the section
depth for \LaTeX{}. This section depth is used to decide where
numbering shall stop, or if that entry shall be presented in the table
of contents, etc.

These two numbers are not the same, due to the support of document
classes with chapters.

We consider a section as important if it's rank is smaller than~3,
i.e., then we'll start a new page. This limit is controlled by the
number |\CwebRankNoEject|. The rank is used as the section level
and is therefore available to configure depth of typeset numbering and
appearance in table of contents.

\beginprog
\def\CwebRankNoEject{3}
\endprog


\sect Sections are started with the macro |\N|. It's first parameter
is the section's rank. The second parameter is the chunk number, the
third the title of the section. This last parameter must be terminated
by a dot.

Each section implicitly starts a chunk. As explained above, major
sections may eject the page, and we add our actual section heading
with |\CwebSection|.

We also tell the user that we have reached the next section by telling
him about the current chunk number. The message is output after the
section header is set, it shall be on the correct page. Since the
token list to be output is expanded at shipout time, we must take care
for the immediate expansion ourselves.

\beginprog
\def\N#1#2#3.{%
    \cwbb@start_chunk{#2}%
    \CwebEjectSection{#1}%
    \CwebSection{#1}{#3}%
    \begingroup
        \def\*{}%
        \edef\tmp{*#2}%
        \message \expandafter{\tmp}%
    \endgroup
    }
\endprog


\sect At the start of a section we usually eject the page if our rank
is not too low. But if the user had a title above the first section,
he won't want this page break. So we try to guess it. If we guess
wrong, the user may either add |\newpage| himself or he may redefine
|\CwebEjectSection| in a package.

\beginprog
\def\CwebEjectSection#1{%		% #1 == rank
    \gdef\CwebEjectSection##1{%
	\relax
        \ifnum ##1<\CwebRankNoEject\relax
            \newpage
        \fi
	}%
    }

\if@titlepage
    \CwebEjectSection{}
\fi
\endprog


\sect Now comes a heary piece. We have 11~section ranks, and need 11
different section categories, section counters, etc. For flat
structure, they shall all do the same; for hierarchic structure, they
shall be really different.

\noindent Let's have a look at the hierarchic structure first, it's
more complex.
%
\begin{itemize}

\item We associate section rank~0 (that's `|@**|') with document
  parts, i.e., map it to |\part|. Many smaller programs won't use it.

\item If we are used in a class that has a bound cseq |\chapter|, we
  associate rank~1 (`|@*0|') with it and rank~2 (`|@*1|') with
  |\section|. Otherwise we associate rank~1 with |\section|. These
  tags (|\chapter| and/or |\section|) must be defined because many
  macros that want to create section divisions will use them.

\item All other section ranks are placed below |\section|.
  |\subsection| and other \LaTeX{} style markup is not supported.

  For hierarchic structure, we identify these sections with category
  names `|section|$l$', where $l$ is the respective level, computed
  from the rank. Those section names do not exist yet, we have to
  create them.

\end{itemize}

Now we come to the point why we can't use the rank as the section
level passed to \LaTeX{}: When we assume the level usage of \LaTeX{}'s
standard classes, |\section| is \emph{always} on level~1, independent
of the existence of |\chapter|. The latter is placed on level~0, its
existence only influences the section level of |\part|, which is
either $-1$ or~0.

This implies that |\subsection| in a standard class is always on
level~2, but our equivalent section division is either `|@*0|'
(rank~1) or `|@*1|' (rank~2). I.e., when we have chapters, we use a
level that's one lower than the rank, otherwise we use the rank.

For flat structure, we use type |chunk|. This way we'll use the chunk
numbers, they are already set up.

For flat structure, we will map all ranks to the same section markup.
This markup will also be accessible by |\section| or |\chapter|. They
will be declared to be on a level that is the same as the rank.


\sect The printed representation of a section divison number is simply
the presentation from the next upper level, a full stop, and an arabic
section number. When we compute the name of the previous level we must
pay attention to our first section division created this way. This
division must be placed below |section|, without any following number.

Since registers are a valuable resource in \TeX{}, we have to take
care not to allocate a superfluous one.

\beginprog
\ifcase \cweb@structure
    %% hierarchic
    \newcounter{section2}[section]
    \@namedef{thesection2}{\@nameuse{thesection}.\arabic{section2}}
    \newcounter{section3}[section2]
    \@namedef{thesection3}{\@nameuse{thesection2}.\arabic{section3}}
    \newcounter{section4}[section3]
    \@namedef{thesection4}{\@nameuse{thesection3}.\arabic{section4}}
    \newcounter{section5}[section4]
    \@namedef{thesection5}{\@nameuse{thesection4}.\arabic{section5}}
    \newcounter{section6}[section5]
    \@namedef{thesection6}{\@nameuse{thesection5}.\arabic{section6}}
    \newcounter{section7}[section6]
    \@namedef{thesection7}{\@nameuse{thesection6}.\arabic{section7}}
    \newcounter{section8}[section7]
    \@namedef{thesection8}{\@nameuse{thesection7}.\arabic{section8}}
    \newcounter{section9}[section8]
    \@namedef{thesection9}{\@nameuse{thesection8}.\arabic{section9}}
    \ifx \chapter\undefined
        \newcounter{section10}[section9]
	\@namedef{thesection10}{\@nameuse{thesection9}.\arabic{section10}}
    \fi
\fi
\endprog


\sect Sections are typeset by |\CwebSection|; rank and title are the
arguments.

For hierarchic structure, we select the proper cseq according to rank:
|\part|, |\chapter| if this is a structure with chapters, |\section|,
and |\cwbb@section{|\<rank>|}| for other ranks. For the last cseq we
also have to supply the section category, it's `|section|$l$' where
$l$ is the rank~$r$ if we don't have chapters, $r-1$ otherwise.

That category cannot be computed by |\cwbb@section| itself since we
want to use that macro also for flat structure. But there all sections
are basically chunks, we have to use the |chunk| category there. Since
we want to differentiate them in the table of contents nevertheless,
we record the current rank in the title.

\beginprog
\ifcase \cweb@structure
    %% hierarchic
    \ifx \chapter\undefined
        \def\CwebSection#1#2{%
	    \ifcase #1
	        %% 0
	        \let\next\part
	      \or  %% 1
		\let\next\section
	    \else  %% level = rank
		\def\cwbb@sect_name{section#1}%
		\def\next{\cwbb@section{#1}}%
	    \fi
	    \next{#2}%
	    }
    \else
        \def\CwebSection#1#2{%
	    \ifcase #1
	        %% 0
	        \let\next\part
	      \or  %% 1
		\let\next\chapter
	      \or  %% 2
		\let\next\section
	    \else  %% level = rank - 1
		\count@=#1  \advance \count@\m@ne
		\edef\cwbb@sect_name{section\number\count@}%
		\edef\next{\noexpand\cwbb@section{\number\count@}}%
	    \fi
	    \next{#2}%
	    }
    \fi
  \or
    %% flat
    \def\CwebSection#1#2{%
	\def\cwbb@sect_name{chunk}%
	\cwbb@section{#1}{\protect\cwbbRank{#1}#2}%
	}
    \def\section{%
	\def\cwbb@sect_name{chunk}%
	\cwbb@section{1}%
	}
\fi
\endprog


\sect We use the same layout for all minor sections. That layout can
be changed by ammended by defining |\CwebSect|$r$|Hook| to redefine
|\CwebSectPreSkip| and/or |\CwebSectLayout| locally. (Of couse, it may
be redefined once globally, too.) These are the values to be used for
arguments 4 and~5 of |\@startsection|, pay attention to the skip sign,
telling about indentation of the first paragraph. And don't forget
that the skip is only used for sections on ranks below
|\CwebRankNoEject|, the others start on a new page anyhow.

The default definition sets the section title in bold and leaves the
same space as before chunks. The first paragraph of the following
documentation has no indentation.

Other layout changes (indentation for heading, etc.)\ must be made by
redefining |\cwbb@section|. Or---email me, if you want to do it. Then
I might be enclined to introduce further possibilities.

\beginprog
\newskip\CwebSectPreSkip
	\CwebSectPreSkip = -\CwebChunkPreSkip
\def\CwebSectLayout{\normalfont\bfseries}

\def\cwbb@section#1{%			% #2 (next token) is title
    \csname CwebSect#1Hook\endcsname	% \relax if undefined
    \expandafter\@startsection
	\expandafter{\cwbb@sect_name}%	% category; for counter, etc.
	{#1}%				% level is rank
	\z@				% no indentation for heading
	\CwebSectPreSkip		% skip above, maybe no next par indent
	\medskipamount			% skip below heading
        \CwebSectLayout
    }
\endprog




\chap Marks.

\begin{fixme}
  Since I don't know yet how to handle marks I ignore most of them for
  the time being.

  Have to change this later.
\end{fixme}

\beginprog
\let\chunkmark\@gobble

\count@=2
\@whilenum \count@<11 \do {
    \expandafter\let \csname section\number\count@ mark\endcsname \@gobble
    \advance \count@ by 1
    }
\endprog




\chap Typesetting programs.

Program pieces come in two flavours: as argument of |\PB|, or as
material after |\B|. In the former case we can use a group for the
switch to the restricted program state, the group end will restore the
previous state again. In the latter case we use |\B| for the switch to
program state, the cross reference list or the next chunk will go to
another state. Since we have to do more for the material after |\B|,
we define this cseq later.

\beginprog
\def\PB#1{%
    \begingroup
	\cwbb@Rprogram
	\leavevmode
	#1%
    \endgroup
    }
\endprog


\sect Note, that |\Y| cannot just be |\smallskip|, as in Plain
\cweb{}. We must assert that the current paragraph is ended before the
vertical glue is inserted, and that's not done by the \LaTeX{}
definition of |\smallskip|. In addition we add some negative penalty,
here is a good place for a page break. As the penalty value we use
half the chunk break penalty---of course a chunk start is an even
better place for a page break\,\dots

It might be that |\Y| is the very first token in a chunk. Then the
chunk start marker isn't set already since it is regarded as a run-in
section heading. In that case we don't set the vertical skip but
simply start the chunk with the program part.

\TeX{}nical note: The flag |@noskipsec| may be used to test if we're
immediately after a run-in section heading.

\beginprog
\newcount\CwebProgPenalty
	\CwebProgPenalty=\@secpenalty
	\divide\CwebProgPenalty by 2
\def\Y{%
    \if@noskipsec
    \else
	\par
	\penalty\CwebProgPenalty
	\smallskip
    \fi
    }
\endprog


\sect Yep, let's unfold the ``official'' names of the cseqs used in
program state.

If we turn on the \cweave{} bindings, they might be in effect already,
we don't need to establish them again. We can test
|\cwbb@user_bindings| for this case, it will be redefined then.

\beginprog
\def\cwbb@cweave_bindings{%
    \ifx \cwbb@user_bindings\relax
	\cwbb@rebind
	    % indentation and paragraph layout
	    \cwbb@incr_indent	\1%
	    \cwbb@decr_indent	\2%
	    \cwbb@expr_break	\3%
	    \cwbb@backup	\4%
	    \cwbb@opt_break	\5%
	    \cwbb@break		\6%
	    \cwbb@big_break	\7%
	    \cwbb@noindent	\8%
	    % C/C++ tokens
	    \CwebRel		\?%
	    \CwebAddress	\AND
	    \CwebComplement	\CM
	    \CwebScope		\DC
	    \CwebEquiv		\E
	    \CwebGe		\G
	    \CwebRightShift	\GG
	    \CwebNe		\I
	    \CwebAssign		\K
	    \CwebLeftShift	\LL
	    \CwebMod		\MOD
	    \CwebNull		\NULL
	    \CwebNot		\R
	    \CwebBinOr		\OR
	    \CwebMemberRef	\PA
	    \CwebThis		\this
	    \CwebOr		\V
	    \CwebAnd		\W
	    \CwebXor		\XOR
	    \CwebLe		\Z
	    \CwebPointer	\MG
	    \CwebPointerMemberRef \MGA
	    \CwebDecr		\MM
	    \CwebIncr		\PP
	    % more tokens
	    \CwebId		\\%
	    \CwebIdLetter	\|%
	    \CwebRes		\&%
	    \CwebString		\.%		%% ( ...Emacs...
	    \CwebStringBreak	\)%
	    \CwebNumber		\T
	    \CwebCombinedOp	\MRL
	    % goes to TeX state
	    \CwebComment	\C
	    \CwebCxxComment	\SHC
	    \CwebRefName	\X
	    % CWEB tokens
	    \CwebMacrosHere	\ATH
	    \CwebDefine		\D
	    \cwbb@format	\F
	    \CwebIdCat		\J
	    \CwebVerbString	\vb
	    % cross reference tags
	    \cwbb@change_flag	\*%
	    \CwebCRAlso		\A
	    \CwebCRsAlso	\As
	    \CwebCRCite		\Q
	    \CwebCRsCite	\Qs
	    \CwebCRUse		\U
	    \CwebCRsUse		\Us
	    \CwebCREt		\ET
	    \CwebCRsEt		\ETs
	    % finish the list
	    \stop\stop
	\def\cwbb@user_bindings{%
	    \cwbb@restore_bindings
	    \let\cwbb@user_bindings\relax
	    }%
    \fi
    }
\endprog


\sect Since most of the \cweave{} bindings are simple and tedious
coding, we'll have a look at the program layout next. Between two
paragraphs there must not be any skip, the skip used in document
layout is saved in |\cwbb@save@parskip|. A few other layout
parameters from the document layout must be saved as well.

\beginprog
\newskip\cwbb@save@parskip
\newskip\cwbb@save@rightskip
\newcount\cwbb@save@sem_sfcode
\newcount\cwbb@save@pretolerance
\newcount\cwbb@save@hyphenpenalty
\newcount\cwbb@save@exhyphenpenalty

\def\cwbb@save_doc_layout{%
    \cwbb@save@parskip\parskip
    \cwbb@save@rightskip\rightskip
    \cwbb@save@sem_sfcode\sfcode`\;
    \cwbb@save@pretolerance\pretolerance
    \cwbb@save@hyphenpenalty\hyphenpenalty
    \cwbb@save@exhyphenpenalty\exhyphenpenalty
    }
\def\cwbb@restore_doc_layout{%
    \parskip\cwbb@save@parskip
    \rightskip\cwbb@save@rightskip
    \sfcode`\;=\cwbb@save@sem_sfcode
    \pretolerance\cwbb@save@pretolerance
    \hyphenpenalty\cwbb@save@hyphenpenalty
    \exhyphenpenalty\cwbb@save@exhyphenpenalty
    }
\endprog


\sect When |\cwbb@program_layout| is called for the very first time
we have to set up some variables. This cannot be done in advance
since they depend on values which might be changed by the user in the
preamble. This setup is done in |\cwbb@layout_init|.

If the program layout is already active, we must not switch it on
another time. This can be tested by the current binding of
|\cwbb@doc_layout|.

\beginprog
\def\cwbb@program_layout{%
    \ifx \cwbb@doc_layout\relax
	\cwbb@layout_init
	\cwbb@save_doc_layout
	\def\cwbb@doc_layout{%
	    \cwbb@restore_doc_layout
	    \let\cwbb@doc_layout\relax
	    }%
	% set new values
	\parskip\z@skip
	\rightskip\z@ plus 100\p@ minus 10\p@
	\sfcode`\;=3000
	\pretolerance\@M
	\hyphenpenalty 1000	% strings can be broken this way
	\exhyphenpenalty\@M
    \fi
    }
\endprog


\sect The unit of the basic indentation is stored in
|\CwebIndentUnit|. Continuation lines are indented two units further,
i.e., the initial hanging indentation is 3~units. The current hanging
indentation is kept in |\cwbb@indent|.

We need undiscardable items which can be used as backspaces, of one
and two units, respectively. This is done best by boxes. The
initialization of the boxes must be postponed until the user had the
chance to change |\CwebIndentUnit|, it's done in the program state
initialization.

\beginprog
\newdimen\CwebIndentUnit
	\CwebIndentUnit=1em
\newdimen\cwbb@indent

\newbox\cwbb@bak	% backspace one unit
\newbox\cwbb@bakk	% backspace two units

\def\cwbb@layout_init{%
    \global\setbox\cwbb@bak \hbox to -1\CwebIndentUnit{}%
    \global\setbox\cwbb@bakk \hbox to -2\CwebIndentUnit{}%
    \global\let\cwbb@layout_init\relax
    }
\endprog


\sect Now we can formulate how to start typesetting program pieces: We
switch to program state, set the initial indentation, and add the
basic indentation of the first code line. Then we might add a marker
to this program part; for hierarchic structure we want to typeset the
chunk number in the margin of the chunk's program part. That's not
really possible, 'though: definitions and format requests are also
introduced with |\B|, so the chunk number will happen to be at the
start of the definition part if one exists.

If the code line is the very first text to be typeset in this chunk,
we don't add the basic indentation---we're already indented from the
run-in chunk start marker. But we must not start our chunk with
|\noindent|, the marker would be typeset in the left margin then. (I
consider this as a \LaTeX{} error.) So we just emulate an empty
documentation part, the marker is now set correctly. |\B| can test for
this case: |@noskipsec| is still true then.

\beginprog
\def\B{%
    \cwbb@program
    \cwbb@indent 3\CwebIndentUnit  \hangindent\cwbb@indent
    \ifvmode
	\if@noskipsec
	    \indent		% add an empty documentation part
	\else
	    \noindent \kern\CwebIndentUnit
	\fi
    \fi
    \CwebNumberProgramPart
    }
\endprog


\sect The chunk number shall be added at the first program part. We
have a hook in the chunk start where we can set up
|\CwebNumberProgramPart|. It will redefine itself at first call, to
disable further output on more |\B| occurences in a chunk.

The chunk number should always be in the left margin. If it's in the
right margin, there would be too much horizontal space between the
code start and the actual number. (The first code line is typically
not very long.) \LaTeX{} does not allow to select the margin for
marginpars, therefore we have to typeset the marginal note `by hand'.
As we know that we're in horizontal mode when we're called, |\vadjust|
is used to create the note. The standard \LaTeX{} dimen register for
the distance between marginal notes and text matter is used, of
course.

Note that marginal notes may overwrite the chunk number. As it's very
unlikely that somebody issues marginal notes in program mode (that's
quite difficult and involves ingenious usage of the implementation of
this macros), I won't care for that case.

\begin{fixme}
  The baseline of the marginal note shall be aligned to the baseline
  of the preceding line. That does not happen by default. Currently, I
  assume that the preceding line has a depth of appropriately the
  depth of a~`g'. Experiments showed that this gives good results for
  most \cweb{} documents. But nevertheless one should determine the
  exact depth of the preceding line, to know how much |box0| should be
  raised. If anybody knows how this can be done, please tell me. (I
  know about the route over the output routine, implementing an own
  marginal note category, but this seems too much effort.)
\end{fixme}

For flat structure, we don't add any program part marker and we
don't need to initialize it.

\beginprog
\ifcase \cweb@structure
    %% hierarchic
    \def\cwbb@marginal_chunkno{%
	\vadjust{%
	    \setbox\z@ \hbox to \z@{%
		\hss \vphantom{g}%
		\small \thechunk
		\hskip \marginparsep
		}%
	    \vbox to \z@{%
		\hsize\z@
		\vss
		\noindent \raise \dp\z@ \box\z@
		}%
	    }%
	}
    \def\cwbb@init_print_chunk{%
       \def\CwebNumberProgramPart{%
	   \cwbb@marginal_chunkno
	   \let\CwebNumberProgramPart\relax
	   }%
       }
  \or
    %% flat
    \let\cwbb@init_print_chunk\relax
    \let\CwebNumberProgramPart\relax
\fi
\endprog


\sect If a statement is finished, a new paragraph with the basic
indentation has to be started.

An optional statement break is implemented by a low penalty which
will be selected if the line has to be broken. We assume that the
hanging indentation of the new line is already set correctly, and
have to backup two units to get the basic indentation. It might be
that the line break is not chosen by \TeX{}, we compensate the
backspace for this case. This compensation is discarded at the start
of a new line.

\beginprog
\def\cwbb@break{%	% forced break, between statements
    \ifmmode\else
	\endgraf	% in LaTeX it isn't sure what \par is really...
	\noindent
	\hangindent\cwbb@indent  \kern\cwbb@indent
	\copy\cwbb@bakk	% go back to basic indentation
	\ignorespaces
    \fi
    }
\def\cwbb@opt_break{%	% optional break between statements
    \hfil \penalty\m@ne
    \hfilneg  \kern .5em  \kern 2\CwebIndentUnit   % discarded on line break
    \copy\cwbb@bakk
    \ignorespaces
    }
\def\cwbb@big_break{%	% forced break and a little extra space
    \Y
    \cwbb@break
    }
\def\cwbb@expr_break#1{%	% break with penalty #1 * 10
    \hfil \penalty#10
    \hfilneg		% discarded on line break
    }
\endprog


\sect When we increment the indentation, we must not forget to set
the hanging indentation immediately since |\cwbb@opt_break| relies on
the new value. If a continuation line is needed it will also be
indented one unit more, which is ok since it should be distinguishable
from the next line.
%
\begin{fixme}
  Shouldn't the hanging indentation be set anew at the end of
  |\cwbb@decr_indent| as well?
\end{fixme}

\beginprog
\def\cwbb@incr_indent{%
    \global\advance\cwbb@indent\CwebIndentUnit
    \hangindent\cwbb@indent
    }
\def\cwbb@decr_indent{\global\advance\cwbb@indent -\CwebIndentUnit}
\def\cwbb@backup{\copy\cwbb@bak}
\def\cwbb@noindent{%				% no indentation
    \hskip -\cwbb@indent \hskip 2\CwebIndentUnit
    }
\endprog




\chap Program (C or \C++) tokens.

Since the user might want to change the (somewhat unusual) way some
of the operators are typeset, we supply names for them.

\beginprog
\let\CwebAnd=\land		% logical and, &&
\let\CwebEquiv=\equiv		% equiv sign, for ==
\let\CwebGe=\ge			% greater or equal
\let\CwebLe=\le			% less or equal
\let\CwebNe=\ne			% unequal, !=
\let\CwebNull=\Lambda		% NULL pointer
\let\CwebNot=\lnot		% logical not, !
\let\CwebOr=\lor		% logical or, ||
\let\CwebXor=\oplus		% bitwise exclusive or, ^
\endprog


\sect Some symbols have to be shifted around, to save computation time
we put them in boxes.

\begin{fixme}
  It's not sure if the following symbols still look good in other font
  families than Computer Modern. The movements might need to get
  adapted when more experience is available. They were chosen from
  Plain \cweb{}, where these were explicit movements measured in
  points. I changed the dimension to quads, to achieve at least the
  support for different sizes in the Computer Modern family.
\end{fixme}

\TeX{}nical note: The pointer symbol must not be declared as a math
symbol (e.g., with |\DeclareMathSymbol|). Then \TeX{} would insert an
italic correction behind it, according to rule~17 from appendix~G.
(Please send email if you know of an easy way to prevent that
addition.) So I use the symbol as an accent, but then I have to supply
a nucleus for this math list. That nucleus also determines the total
width of this construct. As the vector accent has a width of
approximately 5\,pt in |cmmi10|, I'm going to use 0.5\,em. (That
heuristic is as good as any other I can think of.)

\beginprog
%% pointer to struct component (`->'), use symbol of \vec accent
\newbox\cwbb@pointer
\setbox\cwbb@pointer=\hbox{%
	    \kern -.2em
	    \lower .3em \hbox{$\vec{\kern .5em}$}%
	    \kern .1em
	    }

\newbox\cwbb@decr		% decrement, --
\setbox\cwbb@decr=\hbox{%
	    \kern .05em
	    \raise .1em \hbox{$\scriptstyle {-}\kern -.1em{-}$}%
	    \kern .05em
	    }
\newbox\cwbb@incr		% increment, ++
\setbox\cwbb@incr=\hbox{%
	    \kern .05em
	    \raise .1em \hbox{$\scriptstyle {+}\kern -.1em{+}$}%
	    \kern .05em
	    }
\endprog


\sect Although the following symbols are typeset like an ``ordinary
\C++ programmer'' would expect them, we provide own module names
nevertheless. They can now be changed as well, i.e., orthogonality is
enhanced. (And we can use |\cwbb@rebind| for assigning them to their
names while we switch to program state\,\dots).

Note: The comment at |\CwebMod| says ``modulo/remainder'', as this is
not defined by the C standard, i.e., implementation-dependent.

\beginprog
\mathchardef\CwebAddress="2026	% `&', as binary op
\let\CwebAssign==		% assignment
\let\CwebBinOr=\mid		% binary or
\def\CwebComplement{{\sim}}	% `~', as ordinary symbol
\def\CwebDecr{\copy\cwbb@decr}	% decrement
\def\CwebIncr{\copy\cwbb@incr}	% increment
\let\CwebLeftShift=\ll		% left shift, <<
\def\CwebMod{\mathbin{\hbox{\footnotesize\rm\%}}}	% modulo/remainder, %
\def\CwebMemberRef{\mathbin{.*}}	% ptr to member (on object)
\def\CwebPointer{\copy\cwbb@pointer}
\def\CwebPointerMemberRef{\mathbin{\CwebPointer*}} % ptr to member (on ptr)
\def\CwebRel{\mathrel?}		% relation operator
\let\CwebRightShift=\gg		% right shift, >>
\def\CwebScope{\kern.1em{::}\kern.1em}	% scope resolution

\def\CwebThis{\CwebRes{this}}	% reserved identifier `this'
\endprog


\sect The special identifier |\TeX| remains. In math mode it shall be
the identifier, in other modes the usual logo. We use the definition
supplied by DEK (or \textsc{Silvio Levy}?)\ from Plain \cweb{}.

\beginprog
\def\TeX{%
    {%
	\ifmmode\it\fi
	\mbox{T\kern-.1667em\lower.424ex\hbox{E}\hskip-.125emX}%
    }}
\endprog


\sect Some tokens don't have constant names, the name is supplied as
the argument.

Identifiers are typeset in italics, reserved words and type names in
boldface, and strings in typewriter. For an underscore in the bold
words we use a line that's 50\,\% thicker as the usual line thickness.
-- But what is the usual line thickness. As Knuth uses 0.4\,pt for a
10\,pt Computer Modern font, we'll use 0.06\,em for a CM font in an
arbitrary size. If that line thickness is still OK for other fonts is
doubtful, as it should fit to the stem width. (CM is a rather light
font, after all.) But since there is no font dimension that gives a
hint to an appropriate line thickness, it's our best try.

\beginprog
\def\CwebCombinedOp#1{\mathrel{\let\K==#1}}  % e.g., += operator
\def\CwebId#1{\hbox{\it#1\/\kern.05em}} % identifier, more than one char
\def\CwebIdLetter#1{\hbox{$#1$}}	% identifier, one letter
\def\CwebRes#1{%			% reserverd words and type names
    \hbox{\bf
	\def\_{\kern.04em\vbox{\hrule width.3em height.06em}\kern.08em}%
	#1\/\kern .05em
	}%
    }
\endprog


\sect In Plain \cweb{} strings are typeset in the font |cmtex10|,
which is a typewriter font with extended ASCII characters. This font
does not have the usual accents and can therefore not used for
typesetting national characters. (Very often they are input in some
8-bit encoding, the respective character code is made active and is
substituted by a cseq which expands to the correct glyph. Of course,
there are better ways, but that's the reality we have to cope with.)
Instead of this special font we use the standard typewriter font.
This has the further advantage that |cmtex10| may not be introduced
to NFSS\@.

\beginprog
\def\CwebString{%
    \hbox \bgroup
        \tt
	\cwbb@string_setup
	\cwbb@string
    }
\def\cwbb@string#1{%
	#1\kern .05em
	\egroup
    }
\endprog


\sect Within strings certain cseqs have a special meaning; this is
introduced by |\cwbb@string_setup|. The cseqs within strings are
mostly special \TeX{} characters that are to be typeset by cseqs with
the same name.

But a problem remains: active characters. They will still expand, and
will mess up the string visualization. This is a particular problem
with |german.sty| where the double quote is active. (Of course, that's
a problem -- I'm German. :--) When one typesets strings, a double
quote shall be typeset as double quotes, and must not expand to the
special meaning. Remember that double quotes happen to appear quite
often in strings. As the string argument is already tokenized, one has
to redefine the binding of $(\textit{\textup{symbol}}\ \textbf{.}\
|#\"|)$.

\begin{fixme}
  Perhaps one might use a loop over all characters except the braces,
  and set their catcode to other. That would give us enough security.

  For now, we handle only the double quote.
\end{fixme}

Every once in a while we have discretionary breaks in a string,
denoted by |\CwebStringBreak|. This break is shown with the C
convention of escaped newlines.

\beginprog
\def\cwbb@string_setup{%
    \chardef\ =`\  %		% <-- two spaces !
    \chardef\&=`\&
    \chardef\\=`\\
    \chardef\^=`\^
    \chardef\_=`\_
    \chardef\{=`\{
    \chardef\}=`\}
    \chardef\~=`\~
    \catcode`\"=\CatOther
    }

\def\CwebStringBreak{\discretionary{\hbox{\tt\char`\\}}{}{}}
\endprog


\sect Numbers are typeset in different ways. We use the definition of
Plain \cweb{}.
%
\begin{fixme}
  Should add a specification of the possible input and an explanation
  of the macros below. In particular, that the closing brace after
  |\aftergroup| is used much later is probably not grokked by most
  \TeX{} programmers.
\end{fixme}

\beginprog
\def\cwbb@oct{\hbox{$^\circ$\kern-.1em\it\aftergroup\?\aftergroup}}
\def\cwbb@hex{\hbox{$^{\scriptscriptstyle\#}$\tt\aftergroup}}

\def\CwebNumber#1{%		% octal, hex, or decimal constant
    \hbox{%
	$%
	    \def\?{\kern.2em}%
	    \def\$##1{\egroup\sb{\,\rm##1}\bgroup}% suffix to constant
	    \def\_{\cdot 10^{\aftergroup}}% power of ten (via dirty trick)
	    \let\~\cwbb@oct \let\^\cwbb@hex
	    {#1}%
	$}%
    }
\endprog


\sect Comments are typeset in \TeX{} state. We add a hook, the user
shall be able to change the layout (e.g., he might want another font).
%
\begin{fixme}
  Currently \C++ comments are typeset like C comments. This is
  horrible in usual circumstances, i.e., when complete blocks of text
  are prefixed with |//|. We should simply catenate all these text and
  typeset it as one paragraph, each line prefixed by |//|. But then we
  have to implement an |\everyline| first, and since that's not so
  easy we postpone it\,\dots
\end{fixme}
%
In front of a C comment there is an optional stmt break, with 2\,quad
in front if the line is not broken and 1.5\,quad if the line is broken.

\beginprog
\def\CwebComment#1{%
    \5%				% 0.5em will be discarded on line break
    \hskip 1.5em
    $/\ast\,$%
    {\cwbb@tex
	\CwebCommentHook
	#1%
    }%
    $\,\ast/$%
    }
\let\CwebCommentHook\relax
\let\CwebCxxComment\CwebComment
\endprog





\chap \cweb{} tokens.

We distinguish three categories of \cweb{} tokens: (1) Those which
output constant text, (2) those which have attributes to be displayed
in a special way, and (3) those which start a new structure element,
namely |@d| and~|@f|. Let's consider them in this order.


\sect \cweb{} tokens which expand in a constant string are the
identifier catenation operator~(`|@&|') and the macro placement
directive~(`|@h|').

\beginprog
\def\CwebIdCat{\CwebString{@\&}}
\def\CwebMacrosHere{%
    \begingroup
	\def\CwebRefNumber##1{}%
	\CwebRefName :Preprocessor definitions\X
    \endgroup
    }
\endprog


\sect Verbatim program strings, i.e., strings passed verbatim by
\ctangle{}~(`|@=|') are typeset like normal strings, but within a box.
We use 2\,pt as the separating distance, this is set locally.

\beginprog
\def\CwebVerbString#1{{\fboxsep\tw@\p@ \fbox{\CwebString{#1}}}}
\endprog


\sect The refinement names are typeset in angles, this has a long
tradition. We burry the typesetting of the chunk number (which is
the first parameter) in a macro call; the user may change this to
achieve special effects.
%
\begin{fixme}
  One could think about handling special values of the chunk numbers
  differently. E.g., an empty argument is not typeset at all, a~0
  triggers a marginal note about a missing definition, etc.
\end{fixme}

A refinement name may be typeset both in math and horizontal mode.
The name itself is typeset in horizontal mode, of course; for the
angles we need math mode. Therefore we assert at the start of the
macro that we're not in math mode any more. At the end we switch back
to math mode if we've started in it. This conditional switch from and
back to math mode is done by |\cwbb@toggle_text|.

\TeX{}nical note: If we're in math mode |\cwbb@toggle_text| must be
defined globally, as it will turn off math mode and the definition
would be un-made then. The second invocation of |\cwbb@toggle_text|
would be undefined then.

A refinement may also consist of the file name the expansion of this
refinement shall be written to.
%
\label{sec:refname-dot}
%
This file name is tagged with |\.|, it shall be typeset like a string.
But the user shall be able to use the dot accent in the refinement
name as well. We check if the text consists solely of the tag and its
argument; in this case we substitute |\.| with |\CwebString|.
Otherwise we leave it as it is.  Then a refinement name may not
consist of a single dot-accented expression---well, that's highly
unlikely. (Nevertheless I'll document it in the user's manual\,\dots)

\begin{fixme}
  One should be able to use |\EnsureMath| here. But the refinement
  name must not be set in math mode and must not be set in an hbox;
  the latter would prevent line breaks.
\end{fixme}

\beginprog
\def\CwebRefName#1:#2\X{%
    \ifmmode  \gdef\cwbb@toggle_text{\null$\null}%
    \else  \let\cwbb@toggle_text\relax
    \fi
    \cwbb@toggle_text
    $\langle\,${\cwbb@tex \cwbb@check_dot{#2}\CwebRefNumber{#1}}$\,\rangle$%
    \cwbb@toggle_text
    }
\def\CwebRefNumber#1{%
    {\reset@font \footnotesize \kern .5em#1}%
    }
\endprog


\sect The next macro, |\cwbb@check_dot| is a hassle to implement. We
must check if it's argument consists solely of a `|\.|'~tag, together
with the argument to this tag. Then the `|\.|'-argument shall be
typeset like a string, |\.| itself must be ignored.

The check for |\.| must not evaluate the parameter. Especially |\PB|
must not be evaluated within, it would lead to havoc with the
redefinition of all bindings, etc. (At least that's the empirical
result---it happened; although I don't know where the problems are.
Anyhow, it wasn't in the specs of the rebinding process.) All other
cseqs which shouldn't be used in moving arguments may cause problems
here, too. Note that |\protect| is no solution here, the user often is
not aware of the presence of \cweave{}-generated tags. As an example,
consider the refinement \verb"@< local variables of |foo| @>|" which
will lead to a call with the argument `|local variables of \PB{\\{foo}}|'.
|\\| must not be evaluated here.

I use an approach where I hope that it will work---but to be honest,
I'm not sure. I want to define a macro |\next| which shall have an
empty expansion iff |#1| was a `|\.|'~tag with its argument. Empty
refinement names are not allowed so we can assume that |#1| consists
of at least one token. I evaluate the first token of |#1| before the
|\def| is done. With an appropriate binding a |\.| will disappear,
perhaps leaving an empty token list. I also take care that |\PB| is
not evaluated then, by temporary rebinding it to |\relax|. Thereby I
have taken care for all cseqs \cweave{} might have introduced at the
very front of |#1|. Other cseqs are user-tags. If they cause
problems, the user might circumstance them by adding |\protect|.

\beginprog
\def\cwbb@check_dot#1{%
    \begingroup
	\let\.\@gobble
	\let\PB\relax
	\expandafter\def \expandafter\next \expandafter{#1}%
	\ifx \next\empty
	    \gdef\next##1{\CwebString}%
	\else
	    \global\let\next\relax
	\fi
    \endgroup
    \next #1%
    }
\endprog


\sect \cweb{} macro definitions (done by~|@d|) and format directives are
basically program text. When they are started we are already in
program state, i.e., |\B| has appeared in front of it. But since we have
an introductionary identifier (either |#define| or |format|) to set at
the very front we increase the indentation by three units.

The output of format directives may be suppressed.  Then we have to
gather both identifiers that will appear after |\F|.  Both of them
consist of one group-token (a cseq and the identifier name), after
each identifier one other token separates them. I.e., we have to
gobble six group-tokens. As the very last token is |\par| if a format
directive follows directly on another one, the definition must be
long.

\TeX{}nical note: |\cwbb@format| must expand to |\CwebFormat|, not to
its expansion. Otherwise a redefinition of |\CwebFormat| in a package
would not be used.

\beginprog
\def\CwebDefine{\cwbb@macro{\#define}}
\def\CwebFormat{\cwbb@macro{format}}
\if@cweb@suppress@format@
    \long\def\cwbb@format#1#2#3#4#5#6{}
\else
    \def\cwbb@format{\CwebFormat}
\fi

\def\cwbb@macro#1{%
    \global\advance\cwbb@indent \tw@\CwebIndentUnit
    \cwbb@incr_indent
    \CwebRes{#1 }%		% <-- blank!
    }
\endprog




\chap Refinement cross references and changes.

At the end of chunks with refinements, \cweave{} may output cross
reference information: In which chunks additional definitions for
this refinement are found, where this refinement is used, and where it
is cited. This cross reference information is always introduced by a
tag followed by a non-empty list of chunk numbers. Different tags
are used for lists with only one element and for lists with more than
one element, this way the introductionary text may be adapted.

Within a list of $n$~chunk numbers the first $n-1$~numbers are
separated by commas with one following blank. The last two numbers
are separated by |\ET| if~$n=2$, and by |\ETs| if~$n>2$. If a chunk
is changed by the changefile, a changeflag~(`|\*|') is appended to
its number. The list is eventually terminated by a full stop.


\sect We separate the cross reference information by a smallskip from
the refinement or from a previous cross reference information. The
information itself is typeset in a smaller font, as it is auxilliary,
inserted stuff. The number list has a hanging indentation of
|\CwebNumberListHangindent|. But beware: This isn't a dimension
register, it's a macro. This way one can use ems as dimensions.

The actual size change is represented by a macro. We will redefine it
when we need an other size, e.g., in the lists at the end of the
document.

We must assert that we're in CR state while we set a cross reference.

The first parameter of a cross reference information unit is the
introductionary text, the second is the number list. The parameters
must be evaluated in a group---local parameter changes therein must
not influence the environment.

\beginprog
\def\CwebNumberListHangindent{2em}
\let\CwebCRSize\footnotesize
\def\CwebCrossRef#1#2.{%
    \par\smallskip
    \cwbb@CR
    \begingroup
	\reset@font \CwebCRSize
	\noindent \hangindent\CwebNumberListHangindent\relax
	#1~#2.\par
    \endgroup
    }
\endprog


\sect Well, let's define all introducing tags, which in fact start the
cross reference information unit. The number list is gathered by
|\CwebCrossRef|.

\beginprog
\def\CwebCRAlso{\CwebCrossRef{See also chunk}}
\def\CwebCRsAlso{\CwebCrossRef{See also chunks}}

\def\CwebCRCite{\CwebCrossRef{This code is cited in chunk}}
\def\CwebCRsCite{\CwebCrossRef{This code is cited in chunks}}

\def\CwebCRUse{\CwebCrossRef{This code is used in chunk}}
\def\CwebCRsUse{\CwebCrossRef{This code is used in chunks}}

\def\CwebCREt{ and~}
\def\CwebCRsEt{, and~}
\endprog




\chap Bells and whistles.

\begin{fixme}
  Just copied from Plain \cweb{}. It doesn't work
  anyhow---|\char"|\<xy> usually doesn't typeset the correct
  character. That's because this character is most probably an active
  character; at least that's typical for the way \TeX{} systems are
  used in Europe. I have to think about the implementation.
\end{fixme}

\beginprog
\def\ATL{\par\noindent\bgroup\catcode`\_=12 \postATL} % print @l in limbo
\def\postATL#1 #2 {\bf letter \\{\uppercase{\char"#1}}
   tangles as \tentex "#2"\egroup\par}
\def\noATL#1 #2 {}
\def\noatl{\let\ATL=\noATL} % suppress output from @l
\endprog



\chap The end.

Well, after all we're finished with this module

\beginprog
\endinput
\endprog



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%

\vskip \PltxPreSectSkip

\rcsLogIntro{After the actual revision log of this module, we give an
    extract of log messages of the \cls{cweb} class. This module was
    extracted from revision~2.8 of that class, we present to you the
    log entries that concern the code of the new module.}

\begin{rcslog}
$StyleLog: cwebbase.doc,v $
\Revision 1.4 1995/11/07 17:55:19 schrod
    \LaTeX{} \cweb{} should work at least with the last two \LaTeX{}
versions; make it work with the previous-to-last one, version
\mbox{$\langle$1994/12/01$\rangle$}. For that, one has to install the
\mbox{$\langle$1995/06/01$\rangle$} (non-outer) definition of
|\newif|, and |\hb@xt@| must not be used.\\
Problem reported by Laurent Desnogues
\path|<laurent.desnogues@aiguemarine.unice.fr>| and somebody else (XXX
-- add name).

\Revision 1.3 1995/11/07 14:13:12 schrod
Reset section numbers when higher-level section begins.\\
Error reported by Bronne Louis \path|<bronne@montefiore.ulg.ac.be>|.

\Revision 1.2 1995/11/06 10:58:47 schrod
    Support active double quotes in strings. That happens when
|german.sty| is used. Problems with other active characters may still
appear.

\Revision 1.1 1995/09/12 23:02:22 schrod
Moved all code that does the actual typesetting of \cweave{} tags and
is therefore also needed for a |cweb| environment. It's now an own
module named \pkg{cwebbase}.

Chunk markers and chunk marginal numbers don't hang below the
baseline any more.


\item[] \noindent\hrulefill\break
  \textbf{Extract of the revision log for |cweb.doc|:}

  \vskip 2ex

\Revision 2.5 1995/08/29 02:07:26 schrod
Discard dependencies on 10\,pt fonts.

Support suppression of format directives.

\Revision 2.4 1995/08/27 19:31:43 schrod
Make configuration of change flag easier.

Put section title of changed chunks list in marks, and tell the user
that it got typeset.

\Revision 2.3 1995/08/27 17:24:46 schrod
Make usage of baseclass with chapters work.

\Revision 2.2 1995/08/27 13:26:22 schrod
Add possibility to suppress change hints.

\Revision 2.1 1995/08/25 19:11:18 schrod
Hierarchic structures are supported now, in addition to the flat
structure of the beta-test version. One can choose with an option. For
that step, the terminology was cleaned up, too: Chunks are not named
sections any more. (That change involved reimplementation of almost
all the structure and toc stuff.)

The chunk number supplied by \cweave{} is used now, not some computed
number. Change flags are printed, too.

One can suppress output of unchanged sections.

\Revision 1.12  1993/08/10  14:15:43  schrod
New page on main section only if group level $<$ |\cwebSecNoEject|.
Default for the latter is 3.

\Revision 1.11  1993/08/10  11:21:07  schrod
Reference to section number does not render a period after the
number any more.

\Revision 1.10  1993/08/09  20:08:20  schrod
|\cweb@cweave_bindings| is now a no-op if \cweave{} bindings are in
effect already.\\
(Problem reported by Michael M\"uller \path|<mimu@mpi-sb.mpg.de>|.)

\Revision 1.9  1993/08/09  18:05:28  schrod
Left shift operator wasn't defined correctly.\\
(Problem reported by Michael M\"uller \path|<mimu@mpi-sb.mpg.de>|.)

\Revision 1.6  1993/06/15  08:49:23  schrod
|\cweb@check_dot| must not evaluate its argument in an |\edef|, this
causes problems if a |\PB| is within. Now I try hard not to evaluate any
tokens outside of my control.

Can use |\@defpar| for an empty line in |\cweb@has_entries|, don't
need an own macro.

\Revision 1.4  1993/06/14  15:54:18  schrod
Add FSA diagram about processing states. CR state is also switched
to from \TeX{} state (that happens with |\ch| at the document end).

\Revision 1.3  1993/05/13  17:51:21  schrod
Refinements may also be filenames (`|@(|'). Then the complete name
consists of a |\.| macro call, which is handled now.\\
(Problem reported by Michael M\"uller \path|<mimu@mpi-sb.mpg.de>|.)

Made the detection of `|@.|' index entries more robust.

\Revision 1.2  1993/05/12  18:28:59  schrod
Adapted to recent changes of \cweave{} (of April 93):

Main sections have a group level, represented in the table of
contents. This changed the complete implementation of section tags.

New C token cseqs: |\Z| and |\MRL|, implemented as |\CwebLe| and
|\CwebCombinedOp|.

\Revision 1.1  1993/04/09  15:00:37  schrod
Initial revision

\end{rcslog}



\end{document}



%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
Local Variables:
mode: LaTeX
TeX-brace-indent-level: 4
indent-tabs-mode: t
TeX-auto-untabify: nil
TeX-auto-regexp-list: LaTeX-auto-regexp-list
compile-command: "make cwebbase.tex"
End: