\documentstyle{report}

\setlength{\oddsidemargin}{0in}
\setlength{\evensidemargin}{0in}
\setlength{\topmargin}{0in}
\addtolength{\topmargin}{-\headheight}
\addtolength{\topmargin}{-\headsep}
\setlength{\textheight}{8.9in}
\setlength{\textwidth}{6.5in}
\setlength{\marginparwidth}{0.5in}

\title{Nuweb \\ A Simple Literate Programming Tool}
\author{\sl Preston Briggs}
\date{(Version 0.6pc)}

\begin{document}
\maketitle
\tableofcontents

\chapter{Introduction}

In 1984, Knuth introduced the idea of {\em literate programming\/} and
described a pair of tools to support the practise~\cite{knuth:84}.
His approach was to combine Pascal code with \TeX\ documentation to
produce a new language, {\tt WEB}, that offered programmers a superior
approach to programming. He wrote several programs in {\tt WEB},
including {\tt weave} and {\tt tangle}, the programs used to support
literate programming.
The idea was that a programmer wrote one document, the web file, that
combined documentation (written in \TeX~\cite{texbook}) with code
(written in Pascal).

Running {\tt tangle} on the web file would produce a complete
Pascal program, ready for compilation by an ordinary Pascal compiler.
The primary function of {\tt tangle} is to allow the programmer to
present elements of the program in any desired order, regardless of
the restrictions imposed by the programming language. Thus, the
programmer is free to present his program in a top-down fashion,
bottom-up fashion, or whatever seems best in terms of promoting
understanding and maintenance.

Running {\tt weave} on the web file would produce a \TeX\ file, ready
to be processed by \TeX\@@. The resulting document included a variety of
automatically generated indices and cross-references that made it much
easier to navigate the code. Additionally, all of the code sections
were automatically pretty printed, resulting in a quite impressive
document. 

Knuth also wrote the programs for \TeX\ and {\small\sf METAFONT}
entirely in {\tt WEB}, eventually publishing them in book
form~\cite{tex:program,metafont:program}. These are probably the
largest programs ever published in a readable form.

Inspired by Knuth's example, many people have experimented with
{\tt WEB}\@@. Some people have even built web-like tools for their own
favorite combinations of programming language and typesetting
language. For example, {\tt CWEB}, Knuth's current system of choice,
works with a combination of C and \TeX~\cite{levy:90}. Another system,
FunnelWeb, is independent of any programming language and only mildly
dependent on \TeX~\cite{funnelweb}. Inspired by the versatility
of FunnelWeb and by the daunting size of its documentation, I decided
to write my own, very simple, tool for literate programming.%
\footnote{There is another system similar to mine, written by Norman
Ramsey, called {\em noweb}~\cite{noweb}. It perhaps suffers from being
overly Unix-dependent and requiring several programs to use. On the
other hand, its command syntax is very nice. In any
case, nuweb certainly owes its name to his inspiration.}


\section{Nuweb}

Nuweb works with any programming language and \LaTeX~\cite{latex}. I
wanted to use \LaTeX\ because it supports a multi-level sectioning
scheme and has facilities for drawing figures. I wanted to be able to
work with arbitrary programming languages because my friends and I
write programs in many languages (and sometimes combinations of
several languages), {\em e.g.,} C, Fortran, C++, yacc, lex, Scheme,
assembly, Postscript, and so forth. The need to support arbitrary
programming languages has many consequences:
\begin{description}
\item[No pretty printing] Both {\tt WEB} and {\tt CWEB} are able to
  pretty print the code sections of their documents because they
  understand the language well enough to parse it. Since we want to use
  {\em any\/} language, we've got to abandon this feature.
\item[No index of identifiers] Because {\tt WEB} knows about Pascal,
  it is able to construct an index of all the identifiers occurring in
  the code sections (filtering out keywords and the standard type
  identifiers). Unfortunately, this isn't as easy in our case. We don't
  know what an identifiers looks like in each language and we certainly
  don't know all the keywords. (On the other hand, see the end of
  Section~1.3)
\end{description}
Of course, we've got to have some compensation for our losses or the
whole idea would be a waste. Here are the advantages I can see:
\begin{description}
\item[Simplicity] The majority of the commands in {\tt WEB} are
  concerned with control of the automatic pretty printing. Since we
  don't pretty print, many commands are eliminated. A further set of
  commands is subsumed by \LaTeX\  and may also be eliminated. As a
  result, our set of commands is reduced to only four members (explained
  in the next section). This simplicity is also reflected in
  the size of this tool, which is quite a bit smaller than the tools
  used with other approaches.
\item[No pretty printing] Everyone disagrees about how their code
  should look, so automatic formatting annoys many people. One approach
  is to provide ways to control the formatting. Our approach is simpler
  -- we perform no automatic formatting and therefore allow the
  programmer complete control of code layout.
\item[Control] We also offer the programmer complete control of the
  layout of his output files (the files generated during tangling). Of
  course, this is essential for languages that are sensitive to layout;
  but it is also important in many practical situations, {\em e.g.,}
  debugging.
\item[Speed] Since nuweb doesn't do to much, the nuweb tool runs
  quickly. I combine the functions of {\tt tangle} and {\tt weave} into
  a single program that performs both functions at once.
\item[Multiple file output] The programmer may specify more than one
  output file in a single nuweb file. This is required when constructing
  programs in a combination of languages (say, Fortran and C)\@@. It's also
  an advantage when constructing very large programs that would require
  a lot of compile time.
\end{description}
This last point is very important. By allowing the creation of
multiple output files, we avoid the need for monolithic programs.
Thus we support the creation of very large programs by groups of
people. 

A further reduction in compilation time is achieved by first
writing each output file to a temporary location, then comparing the
temporary file with the old version of the file. If there is no
difference, the temporary file can be deleted. If the files differ,
the old version is deleted and the temporary file renamed. This
approach works well in combination with {\tt make} (or similar tools),
since {\tt make} will avoid recompiling untouched output files.


\section{Writing Nuweb}

The bulk of a nuweb file will be ordinary \LaTeX\@@. In fact, any \LaTeX\
file can serve as input to nuweb and will be simply copied through
unchanged to the {\tt .tex} file -- unless a nuweb command is
discovered. All nuweb commands begin with an ``at-sign'' ({\tt @@}).
Therefore, a file without at-signs will be copied unchanged.
Nuweb commands are used to specify {\em output files,} define 
{\em macros,} and delimit {\em scraps}. These are the basic features
of interest to the nuweb tool -- all else is simply text to be copied
to the {\tt .tex} file.

\subsection{The Major Commands}

Files and macros are defined with the following commands:
\begin{description}
\item[\tt @@o {\em file-name flags scrap\/}] Output a file. The file name is
  terminated by whitespace.
\item[\tt @@d {\em macro-name scrap\/}] Define a macro. The macro name
  is terminated by a return or the beginning of a scrap.
\end{description}
A specific file may be specified several times, with each definition
being written out, one after the other, in the order they appear.
The definitions of macros may be similarly divided.

\subsubsection{Scraps}

Scraps have specific begin markers and end markers to allow precise
control over the contents and layout. Note that any amount of
whitespace (including carriage returns) may appear between a name and
the beginning of a scrap.
\begin{description}
\item[\tt @@\{{\em anything\/}@@\}] where the scrap body includes every
  character in {\em anything\/} -- all the blanks, all the tabs, all the
  carriage returns.
\end{description}
Inside a scrap, we may invoke a macro.
\begin{description}
\item[\tt @@<{\em macro-name\/}@@>] Causes the macro {\em macro-name\/}
  to be expanded inline as the code is written out to a file. It is an
  error to specify recursive macro invocations.
\end{description}
Note that macro names may be abbreviated, either during invocation or
definition. For example, it would be very tedious to have to
repeatedly type the macro name
\begin{quote}
{\tt @@d Check for terminating at-sequence and return name if found}
\end{quote}
Therefore, we provide a mechanism (stolen from Knuth) of indicating
abbreviated names.
\begin{quote}
{\tt @@d Check for terminating...}
\end{quote}
Basically, the programmer need only type enough characters to uniquely
identify the macro name, followed by three periods. An abbreviation
may even occur before the full version; nuweb simply preserves the
longest version of a macro name. Note also that blanks and tabs are
insignificant in a macro name; any string of them are replaced by a
single blank.

When scraps are written to an output or {\tt .tex} file, tabs are
expanded into spaces by default. Currently, I assume tab stops are set
every eight characters. Furthermore, when a macro is expanded in a scrap,
the body of the macro is indented to match the indentation of the
macro invocation. Therefore, care must be taken with languages 
({\em e.g.,} Fortran) that are sensitive to indentation.
These default behaviors may be changed for each output file (see
below).

\subsubsection{Flags}

When defining an output file, the programmer has the option of using
flags to control output of a particular file. The flags are intended
to make life a little easier for programmers using certain languages.
They introduce little language dependences; however, they do so only
for a particular file. Thus it is still easy to mix languages within a
single document. There are three ``per-file'' flags:
\begin{description}
\item[\tt -d] Forces the creation of \verb|#line| directives in the
  output file. These are useful with C (and sometimes C++ and Fortran) on
  many Unix systems since they cause the compiler's error messages to
  refer to the web file rather than the output file. Similarly, they
  allow source debugging in terms of the web file.
\item[\tt -i] Suppresses the indentation of macros. That is, when a
  macro is expanded in a scrap, it will {\em not\/} be indented to
  match the indentation of the macro invocation. This flag would seem
  most useful for Fortran programmers.
\item[\tt -t] Suppresses expansion of tabs in the output file. This
  feature seems important when generating {\tt make} files.
\end{description}


\subsection{The Minor Commands}

We have two very low-level utility commands that may appear anywhere
in the web file.
\begin{description}
\item[\tt @@@@] Causes a single ``at sign'' to be copied into the output.
\item[\tt @@i {\em file-name\/}] Includes a file. Includes may be
  nested, though there is currently a limit of 10~levels. The file name
  should be complete (no extension will be appended) and should be
  terminated by a carriage return.
\end{description}
Finally, there are three commands used to create indices to the macro
names, file definitions, and user-specified identifiers.
\begin{description}
\item[\tt @@f] Create an index of file names.
\item[\tt @@m] Create an index of macro name.
\item[\tt @@u] Create an index of user-specified identifiers.
\end{description}
I usually put these in their own section
in the \LaTeX\ document; for example, see Chapter~\ref{indices}.

Identifiers must be explicitely specified for inclusion in the 
{\tt @@u} index. By convention, each identifier is marked at the point
of its definition; all references to each identifier (inside scraps)
will be discovered automatically. To ``mark'' an identifier for
inclusion in the index, we must mention it at the end of a scrap.
For example,
\begin{quote}
\begin{verbatim}
@@d a scrap @@{
Let's pretend we're declaring the variables FOO and BAR
inside this scrap.
@@| FOO BAR @@}
\end{verbatim}
\end{quote}
I've used alphabetic identifiers in this example, but any string of
characters (not including whitespace) will do.  Therefore, it's
possible to add index entries for things like {\tt <<=} if desired.
An identifier may be declared in more than one scrap.

In the generated index, each identifier appears with a list of all the
scraps using and defining it, where the defining scraps are
distinguished by underlining. Note that the identifier doesn't
actually have to appear in the defining scrap; it just has to be in
the list of definitions at the end of a scrap.


\section{Running Nuweb}

Nuweb is invoked using the following command:
\begin{quote}
{\tt nuweb} {\em flags file-name}\ldots
\end{quote}
One or more files may be processed at a time. If a file name has no
extension, {\tt .w} will be appended. While a file name may specify a
file in another directory, the resulting {\tt .tex} file will always
be created in the current directory. For example,
\begin{quote}
{\tt nuweb /foo/bar/quux}
\end{quote}
will take as input the file {\tt /foo/bar/quux.w} and will create the
file {\tt quux.tex} in the current directory.

By default, nuweb performs both tangling and weaving at the same time.
Normally, this is not a bottleneck in the compilation process;
however, it's possible to achieve slightly faster throughput by
avoiding one or another of the default functions using command-line
flags. There are currently three possible flags:
\begin{description}
\item[\tt -t] Suppress generation of the {\tt .tex} file.
\item[\tt -o] Suppress generation of the output files.
\item[\tt -c] Avoid testing output files for change before updating them.
\end{description}
Thus, the command
\begin{quote}
{\tt nuweb -to /foo/bar/quux}
\end{quote}
would simply scan the input and produce no output at all.


\section{Restrictions and Extensions}

Because nuweb is intended to be a simple tool, I've established a few
restrictions. Over time, some of these may be eliminated; others seem
fundamental.
\begin{itemize}
\item The scraps don't print well with older versions of \TeX\ 
  (pre~3.0). I'm not sure why and I'm not sure it's always true
  (I don't have a version to experiment with).
\item The handling of errors is not completely ideal. In some cases, I
  simply warn of a problem and continue; in other cases I halt
  immediately. This behavior should be regularized.
\item I warn about references to macros that haven't been defined, but
  don't halt. This seems most convenient for development, but may change
  in the future.
\item File names and index entries should not contain any {\tt @@}
  signs.
\item Macro names may be (almost) any well-formed \TeX\ string.
  It makes sense to change fonts or use math mode; however, care should
  be taken to ensure matching braces, brackets, and dollar signs.
\item Anything is allowed in the body of a scrap; however, very
  long scraps (horizontally or vertically) may not typeset well.
\end{itemize}
Very long scraps may be allowed to break across a page if declared
with {\tt @@O} or {\tt @@D} (instead of {\tt @@o} or {\tt @@d}).
This doesn't work very well as a default, since far too many short
scraps will be broken across pages; however, as a user-controlled
option, it seems very useful.


\chapter{The Overall Structure}

Processing a web requires three major steps:
\begin{enumerate}
\item Read the source, accumulating file names, macro names, scraps,
and lists of cross-references.
\item Reread the source, copying out to the {\tt .tex} file, with
protection and cross-reference information for all the scraps.
\item Traverse the list of files names. For each file name:
\begin{enumerate}
\item Dump all the defining scraps into a temporary file. 
\item If the file already exists and is unchanged, delete the
temporary file; otherwise, rename the temporary file.
\end{enumerate}
\end{enumerate}


\section{Files}

I have divided the program into several files for quicker
recompilation during development. Rather than use {\tt .h} files
for shared declarations, I'll simply create a shared scrap containing
all the global declarations. This isn't really best from the point of
view of disk space; but it does simplify the makefile (and allow me to
play with my new toy).
@d Shared declarations
@{@<Include files@>
@<Type declarations@>
@<Global variables@>
@<Function prototypes@>
@}

We'll need at least two of the standard system include files.
@d Include files
@{#include <stdio.h>
#include <string.h>
#include <ctype.h>
@| FILE stderr exit fprintf fputs fopen fclose getc putc strlen
toupper isupper islower isgraph isspace @}

\newpage
\noindent
I also like to use {\tt TRUE} and {\tt FALSE} in my code.
I'd use an {\tt enum} here, except that some systems seem to provide
definitions of {\tt TRUE} and {\tt FALSE} be default.  The following
code seems to work on all the local systems.
@d Type dec...
@{#ifndef FALSE
#define FALSE 0
#endif
#ifndef TRUE
#define TRUE (!0)
#endif
@| FALSE TRUE @}


\subsection{The Main Files}

The code is divided into four main files (introduced here) and five
support files (introduced in the next section).
The file {\tt main.c} will contain the driver for the whole program
(see Section~\ref{main-routine}).
@o main.c
@{@<Shared...@>
@}


The first pass over the source file is contained in {\tt pass1.c}.
It handles collection of all the file names, macros names, and scraps
(see Section~\ref{pass-one}).
@o pass1.c
@{@<Shared...@>
@}

The {\tt .tex} file is created during a second pass over the source
file. The file {\tt latex.c} contains the code controlling the
construction of the {\tt .tex} file 
(see Section~\ref{latex-file}).
@o latex.c
@{@<Shared...@>
@}


The code controlling the creation of the output files is in {\tt output.c}
(see Section~\ref{output-files}).
@o output.c
@{@<Shared...@>
@}


\subsection{Support Files}

The support files contain a variety of support routines used to define
and manipulate the major data abstractions.
The file {\tt input.c} holds all the routines used for referring to
source files (see Section~\ref{source-files}).
@o input.c
@{@<Shared...@>
@}

Creation and lookup of scraps is handled by routines in {\tt scraps.c}
(see Section~\ref{scraps}).
@o scraps.c
@{@<Shared...@>
@}


The handling of file names and macro names is detailed in {\tt names.c}
(see Section~\ref{names}).
@o names.c
@{@<Shared...@>
@}

Memory allocation and deallocation is handled by routines in {\tt arena.c}
(see Section~\ref{memory-management}). I don't bother including all
the shared declarations since they aren't required.
@o arena.c
@{@<Include...@>
@}


\section{The Main Routine} \label{main-routine}

The main routine is quite simple in structure.
It wades through the optional command-line arguments,
then handles any files listed on the command line.
@o main.c
@{void main(argc, argv)
     int argc;
     char **argv;
{
  int arg = 1;
  @<Interpret command-line arguments@>
  @<Process the remaining arguments (file names)@>
  exit(0);
}
@| main @}


\subsection{Command-Line Arguments}

There are three possible command-line arguments:
\begin{description}
\item[{\tt -t}] Suppresses generation of the {\tt .tex} file.
\item[{\tt -o}] Suppresses generation of the output files.
\item[{\tt -c}] Forces output files to overwrite old files of the same
name without comparing for equality first.
\item[{\tt -v}] Prints the current version of {\tt nuweb}.
\end{description}
Global flags are declared for each of the arguments.
@d Global...
@{int tex_flag;      /* if FALSE, don't emit the .tex file */
int output_flag;   /* if FALSE, don't emit the output files */
int compare_flag;  /* if FALSE, overwrite without comparison */
@| tex_flag output_flag compare_flag @}

\newpage
\noindent
The flags are all initialized to {\tt TRUE} for correct default behavior.
@d Interpret com...
@{tex_flag = TRUE;
output_flag = TRUE;
compare_flag = TRUE;
@}


We save the invocation name of the command in a global variable
{\tt command\_name} for use in error messages.
@d Global...
@{char *command_name;
@| command_name @}

The invocation name is conventionally passed in {\tt argv[0]}.
@d Interpret com...
@{command_name = argv[0];
@}


We need to examine the remaining entries in {\tt argv}, looking for
command-line arguments.
@d Interpret com...
@{while (arg < argc) {
  char *s = argv[arg];
  if (*s++ == '-') {
    @<Interpret the argument string {\tt s}@>
    arg++;
  }
  else break;
}@}

\newpage
\noindent
Several flags can be stacked behind a single minus sign; therefore,
we've got to loop through the string, handling them all.
@d Interpret the...
@{{
  char c = *s++;
  while (c) {
    switch (c) {
      case 'c': compare_flag = FALSE;
		break;
      case 'o': output_flag = FALSE;
		break;
      case 't': tex_flag = FALSE;
	 	break;
      case 'v': printf("nuweb version 0.6\n");
		break;
      default:  fprintf(stderr, "%s: unexpected argument ignored.  ",
			command_name);
		fprintf(stderr, "Usage is: %s [-cot] file...\n",
			command_name);
		break;
    }
    c = *s++;
  }
}@}


\subsection{File Names}

We expect at least one file name. While a missing file name might be
ignored without causing any problems, we take the opportunity to report
the usage convention.
@d Process the remaining...
@{if (arg < argc)
  do {
    @<Handle the file name in {\tt argv[arg]}@>
    arg++;
  } while (arg < argc);
else {
  fprintf(stderr, "%s: expected a file name.  ", command_name);
  fprintf(stderr, "Usage is: %s [-cot] file-name...\n", command_name);
  exit(-1);
}@}

\newpage
\noindent
The code to handle a particular file name is rather more tedious than
the actual processing of the file. A file name may be an arbitrarily
complicated path name, with an optional extension. If no extension is
present, we add {\tt .w} as a default. The extended path name will be
kept in a local variable {\tt source\_name}. The resulting {\tt .tex}
file will be written in the current directory; its name will be kept
in the variable {\tt tex\_name}.
@d Handle the file...
@{{
  char source_name[100];
  char tex_name[100];
  @<Build {\tt source\_name} and {\tt tex\_name}@>
  @<Process a file@>
}@}


I bump the pointer {\tt p} through all the characters in {\tt argv[arg]},
copying all the characters into {\tt source\_name} (via the pointer
{\tt q}). 

At each slash, I update {\tt trim} to point just past the
slash in {\tt source\_name}. The effect is that {\tt trim} will point
at the file name without any leading directory specifications.

The pointer {\tt dot} is made to point at the file name extension, if
present. If there is no extension, we add {\tt .w} to the source name.
In any case, we create the {\tt tex\_name} from {\tt trim}, taking
care to get the correct extension.
@d Build {\tt sou...
@{{
  char *p = argv[arg];
  char *q = source_name;
  char *trim = q;
  char *dot = NULL;
  char c = *p++;
  while (c) {
    *q++ = c;
    if (c == '/') {
      trim = q;
      dot = NULL;
    }
    else if (c == '.')
      dot = q - 1;
    c = *p++;
  }
  *q = '\0';
  if (dot) {
    *dot = '\0';
    sprintf(tex_name, "%s.tex", trim);
    *dot = '.';
  }
  else {
    sprintf(tex_name, "%s.tex", trim);
    *q++ = '.';
    *q++ = 'w';
    *q = '\0';
  }
}@}

Now that we're finally ready to process a file, it's not really too
complex.  We bundle most of the work into three routines {\tt pass1}
(see Section~\ref{pass-one}), {\tt write\_tex} (see
Section~\ref{latex-file}), and {\tt write\_files} (see
Section~\ref{output-files}). After we're finished with a
particular file, we must remember to release its storage (see
Section~\ref{memory-management}).
@d Process a file
@{{
  pass1(source_name);
  if (tex_flag)
    write_tex(source_name, tex_name);
  if (output_flag)
    write_files(file_names);
  arena_free();
}@}


\section{Pass One} \label{pass-one}

During the first pass, we scan the file, recording the definitions of
each macro and file and accumulating all the scraps. 

@d Function pro...
@{void pass1();
@}


The routine {\tt pass1} takes a single argument, the name of the
source file. It opens the file, then initializes the scrap structures
(see Section~\ref{scraps}) and the roots of the file-name tree, the
macro-name tree, and the tree of user-specified index entries (see 
Section~\ref{names}). After completing all the
necessary preparation, we make a pass over the file, filling in all
our data structures. Next, we seach all the scraps for references to
the user-specified index entries. Finally, we must reverse all the
cross-reference lists accumulated while scanning the scraps.
@o pass1.c
@{void pass1(file_name)
     char *file_name;
{
  source_open(file_name);
  init_scraps();
  macro_names = NULL;
  file_names = NULL;
  user_names = NULL;
  @<Scan the source file, looking for at-sequences@>
  if (tex_flag)
    search();
  @<Reverse cross-reference lists@>
}
@| pass1 @}

\newpage
\noindent
The only thing we look for in the first pass are the command
sequences. All ordinary text is skipped entirely.
@d Scan the source file, look...
@{{
  int c = source_get();
  while (c != EOF) {
    if (c == '@@')
      @<Scan at-sequence@>
    c = source_get();
  }
}@}

Only four of the at-sequences are interesting during the first pass.
We skip past others immediately; warning if unexpected sequences are
discovered.
@d Scan at-sequence
@{{
  c = source_get();
  switch (c) {
    case 'O':
    case 'o': @<Build output file definition@>
	      break;
    case 'D':
    case 'd': @<Build macro definition@>
	      break;
    case '@@':
    case 'u':
    case 'm':
    case 'f': /* ignore during this pass */
	      break;
    default:  fprintf(stderr,
		      "%s: unexpected @@ sequence ignored (%s, line %d)\n",
		      command_name, source_name, source_line);
	      break;
  }
}@}


\subsection{Accumulating Definitions}

There are three steps required to handle a definition:
\begin{enumerate}
\item Build an entry for the name so we can look it up later.
\item Collect the scrap and save it in the table of scraps.
\item Attach the scrap to the name.
\end{enumerate}
We go through the same steps for both file names and macro names.
@d Build output file definition
@{{
  Name *name = collect_file_name(); /* returns a pointer to the name entry */
  int scrap = collect_scrap();	    /* returns an index to the scrap */
  @<Add {\tt scrap} to...@>
}@}


@d Build macro definition
@{{
  Name *name = collect_macro_name();
  int scrap = collect_scrap();
  @<Add {\tt scrap} to...@>
}@}

Since a file or macro may be defined by many scraps, we maintain them
in a simple linked list. The list is actually built in reverse order,
with each new definition being added to the head of the list.
@d Add {\tt scrap} to {\tt name}'s definition list
@{{
  Scrap_Node *def = (Scrap_Node *) arena_getmem(sizeof(Scrap_Node));
  def->scrap = scrap;
  def->next = name->defs;
  name->defs = def;
}@}


\subsection{Fixing the Cross References}

Since the definition and reference lists for each name are accumulated
in reverse order, we take the time at the end of {\tt pass1} to
reverse them all so they'll be simpler to print out prettily.
The code for {\tt reverse\_lists} appears in Section~\ref{names}.
@d Reverse cross-reference lists
@{{
  reverse_lists(file_names);
  reverse_lists(macro_names);
  reverse_lists(user_names);
}@}



\section{Writing the Latex File} \label{latex-file}

The second pass (invoked via a call to {\tt write\_tex}) copies most of
the text from the source file straight into a {\tt .tex} file.
Definitions are formatted slightly and cross-reference information is
printed out.

Note that all the formatting is handled in this section.
If you don't like the format of definitions or indices or whatever,
it'll be in this section somewhere. Similarly, if someone wanted to
modify nuweb to work with a different typesetting system, this would
be the place to look.

@d Function...
@{void write_tex();
@}

\newpage
\noindent
We need a few local function declarations before we get into the body
of {\tt write\_tex}.

@o latex.c
@{static void copy_scrap();		/* formats the body of a scrap */
static void print_scrap_numbers();	/* formats a list of scrap numbers */
static void format_entry();		/* formats an index entry */
static void format_user_entry();
@}

The routine {\tt write\_tex} takes two file names as parameters: the
name of the web source file and the name of the {\tt .tex} output file.
@o latex.c
@{void write_tex(file_name, tex_name)
     char *file_name;
     char *tex_name;
{
  FILE *tex_file = fopen(tex_name, "w");
  if (tex_file) {
    source_open(file_name);
    @<Copy {\tt source\_file} into {\tt tex\_file}@>
    fclose(tex_file);
  }
  else
    fprintf(stderr, "%s: can't open %s\n", command_name, tex_name);
}
@| write_tex @}

We make our second (and final) pass through the source web, this time
copying characters straight into the {\tt .tex} file. However, we keep
an eye peeled for {\tt @@}~characters, which signal a command sequence.

@d Copy {\tt source\_file}...
@{{
  int scraps = 1;
  int c = source_get();
  while (c != EOF) {
    if (c == '@@')
      @<Interpret at-sequence@>
    else {
      putc(c, tex_file);
      c = source_get();
    }
  }
}@}

@d Interpret at-sequence
@{{
  int big_definition = FALSE;
  c = source_get();
  switch (c) {
    case 'O': big_definition = TRUE;
    case 'o': @<Write output file definition@>
	      break;
    case 'D': big_definition = TRUE;
    case 'd': @<Write macro definition@>
	      break;
    case 'f': @<Write index of file names@>
	      break;
    case 'm': @<Write index of macro names@>
	      break;
    case 'u': @<Write index of user-specified names@>
	      break;
    case '@@': putc(c, tex_file);
    default:  c = source_get();
	      break;
  }
}@}


\subsection{Formatting Definitions}

We go through a fair amount of effort to format a file definition.
I've derived most of the \LaTeX\ commands experimentally; it's quite
likely that an expert could do a better job. The \LaTeX\ for
the previous macro definition should look like this (perhaps modulo
the scrap numbers):
\begin{verbatim}
\begin{flushleft}
\begin{minipage}{\linewidth}
$\langle$Interpret at-sequence {\footnotesize 35}$\rangle\equiv$
\vspace{-1.5ex}
\begin{quote}
\verb@@{@@\\
\verb@@  c = source_get();@@\\
\verb@@  switch (c) {@@\\
\verb@@    case 'O': big_definition = TRUE;@@\\
\verb@@    case 'o': @@$\langle$Write output file definition {\footnotesize 36}$\rangle$\verb@@@@\\
\end{verbatim}
\vdots
\begin{verbatim}
\verb@@    case '@@{\tt @@}\verb@@': putc(c, tex_file);@@\\
\verb@@    default:  c = source_get();@@\\
\verb@@              break;@@\\
\verb@@  }@@\\
\verb@@}@@$\Diamond$
\end{quote}
\vspace{-2ex}
{\footnotesize Macro referenced in scrap 34.}
\end{minipage}\\[4ex]
\end{flushleft}
\end{verbatim}
The {\em flushleft\/} environment is used to avoid \LaTeX\ warnings
about underful lines. The {\em minipage\/} environment is used to
avoid page breaks in the middle of scraps. The {\em verb\/} command
allows arbitrary characters to be printed (however, note the special
handling of the {\tt @@} case in the switch statement).

Macro and file definitions are formatted nearly identically.
I've factored the common parts out into separate scraps.

@d Write output file definition
@{{
  Name *name = collect_file_name();
  @<Begin the scrap environment@>
  fprintf(tex_file, "\\verb@@\"%s\"@@", name->spelling);
  fprintf(tex_file, " {\\footnotesize %d }$\\equiv$\n", scraps++);
  @<Fill in the middle of the scrap environment@>
  @<Write file defs@>
  @<Finish the scrap environment@>
}@}

I don't format a macro name at all specially, figuring the programmer
might want to use italics or bold face in the midst of the name.

@d Write macro definition
@{{
  Name *name = collect_macro_name();
  @<Begin the scrap environment@>
  fprintf(tex_file, "$\\langle$%s", name->spelling);
  fprintf(tex_file, " {\\footnotesize %d}$\\rangle\\equiv$\n", scraps++);
  @<Fill in the middle of the scrap environment@>
  @<Write macro defs@>
  @<Write macro refs@>
  @<Finish the scrap environment@>
}@}


@d Begin the scrap environment
@{{
  fputs("\\begin{flushleft}\n", tex_file);
  if (!big_definition)
    fputs("\\begin{minipage}{\\linewidth}\n", tex_file);
}@}

The interesting things here are the $\Diamond$ inserted at the end of
each scrap and the various spacing commands. The diamond helps to
clearly indicate the end of a scrap. The spacing commands were derived
empirically; they may be adjusted to taste.

@d Fill in the middle of the scrap environment
@{{
  fputs("\\vspace{-1.5ex}\n\\begin{quote}\n", tex_file);
  copy_scrap(tex_file);
  fputs("$\\Diamond$\n\\end{quote}\n\\vspace{-2ex}\n", tex_file);
}@}

\newpage
\noindent
We've got one last spacing command, controlling the amount of white
space after a scrap.

Note also the whitespace eater. I use it to remove any blank lines
that appear after a scrap in the source file. This way, text following
a scrap will not be indented. Again, this is a matter of personal taste.

@d Finish the scrap environment
@{{
  if (!big_definition)
    fputs("\\end{minipage}\\\\[4ex]\n", tex_file);
  fputs("\\end{flushleft}\n", tex_file);
  do
    c = source_get();
  while (isspace(c));
}@}


\subsubsection{Formatting Cross References}

@d Write file defs
@{{
  if (name->defs->next) {
    fputs("{\\footnotesize File defined by scraps ", tex_file);
    print_scrap_numbers(tex_file, name->defs);
    fputs("}\n", tex_file);
  }
}@}

@d Write macro defs
@{{
  fputs("{\\footnotesize ", tex_file);
  if (name->defs->next) {
    fputs("Macro defined by scraps ", tex_file);
    print_scrap_numbers(tex_file, name->defs);
    fputs("\\\\[-1ex]\n", tex_file);
  }
}@}

@d Write macro refs
@{{
  if (name->uses) {
    if (name->uses->next) {
      fputs("Macro referenced in scraps ", tex_file);
      print_scrap_numbers(tex_file, name->uses);
      fputs("}\n", tex_file);
    }
    else
      fprintf(tex_file, "Macro referenced in scrap %d.}\n", name->uses->scrap);
  }
  else {
    fputs("Macro never referenced.}\n", tex_file);
    fprintf(stderr, "%s: <%s> never referenced.\n",
	    command_name, name->spelling);
  }
}@}


@o latex.c
@{static void print_scrap_numbers(tex_file, scraps)
     FILE *tex_file;
     Scrap_Node *scraps;
{
  fprintf(tex_file, "%d", scraps->scrap);
  scraps = scraps->next;
  while (scraps->next) {
    fprintf(tex_file, ", %d", scraps->scrap);
    scraps = scraps->next;
  }
  fprintf(tex_file, " and~%d.", scraps->scrap);
}
@| print_scrap_numbers @}


\subsubsection{Formatting a Scrap}

@o latex.c
@{static void copy_scrap(file)
     FILE *file;
{
  int indent = 0;
  int c = source_get();
  fputs("\\verb@@", file);
  while (1) {
    switch (c) {
      case '@@':  @<Check at-sequence for end-of-scrap@>
		 break;
      case '\n': fputs("@@\\\\\n\\verb@@", file);
		 indent = 0;
		 break;
      case '\t': @<Expand tab into spaces@>
		 break;
      default:   putc(c, file);
		 indent++;
		 break;
    }
    c = source_get();
  }
}
@| copy_scrap @}


@d Expand tab into spaces
@{{
  int delta = 8 - (indent % 8);
  indent += delta;
  while (delta > 0) {
    putc(' ', file);
    delta--;
  }
}@}

@d Check at-sequence...
@{{
  c = source_get();
  switch (c) {
    case '@@': fputs("@@{\\tt @@}\\verb@@", file);
	      break;
    case '|': @<Skip over index entries@>
    case '}': putc('@@', file);
	      return;
    case '<': @<Format macro name@>
	      break;
    default:  /* ignore these since pass1 will have warned about them */
	      break;
  }
}@}

There's no need to check for errors here, since we will have already
pointed out any during the first pass.
@d Skip over index entries
@{{
  do {
    do
      c = source_get();
    while (c != '@@');
    c = source_get();
  } while (c != '}');
}@}


@d Format macro name
@{{
  Name *name = collect_scrap_name();
  fprintf(file, "@@$\\langle$%s {\\footnotesize ", name->spelling);
  if (name->defs)
    fprintf(file, "%d", name->defs->scrap);
  else {
    putc('?', file);
    fprintf(stderr, "%s: scrap never defined <%s>\n",
	    command_name, name->spelling);
  }
  fputs("}$\\rangle$\\verb@@", file);
}@}



\subsection{Generating the Indices}

@d Write index of file names
@{{
  if (file_names) {
    fputs("\n\\begin{list}{}{\\setlength{\\itemsep}{-\\parsep}}\n", tex_file);
    format_entry(file_names, tex_file, TRUE);
    fputs("\\end{list}", tex_file);
  }
  c = source_get();
}@}


@d Write index of macro names
@{{
  if (macro_names) {
    fputs("\n\\begin{list}{}{\\setlength{\\itemsep}{-\\parsep}}\n", tex_file);
    format_entry(macro_names, tex_file, FALSE);
    fputs("\\end{list}", tex_file);
  }
  c = source_get();
}@}


@o latex.c
@{static void format_entry(name, tex_file, file_flag)
     Name *name;
     FILE *tex_file;
     int file_flag;
{
  while (name) {
    format_entry(name->llink, tex_file, file_flag);
    @<Format an index entry@>
    name = name->rlink;
  }
}
@| format_entry @}


@d Format an index entry
@{{
  fputs("\\item \\hspace{-\\leftmargin}", tex_file);
  if (file_flag) {
    fprintf(tex_file, "\\verb@@\"%s\"@@ ", name->spelling);
    @<Write file's defining scrap numbers@>
  }
  else {
    fprintf(tex_file, "$\\langle$%s {\\footnotesize ", name->spelling);
    @<Write defining scrap numbers@>
    fputs("}$\\rangle$ ", tex_file);
    @<Write referencing scrap numbers@>
  }
  putc('\n', tex_file);
}@}


@d Write file's defining scrap numbers
@{{
  Scrap_Node *p = name->defs;
  fputs("{\\footnotesize Defined by scrap", tex_file);
  if (p->next) {
    fputs("s ", tex_file);
    print_scrap_numbers(tex_file, p);
  }
  else
    fprintf(tex_file, " %d.", p->scrap);
  putc('}', tex_file);
}@}

@d Write defining scrap numbers
@{{
  Scrap_Node *p = name->defs;
  if (p) {
    while (1) {
      fprintf(tex_file, "%d", p->scrap);
      p = p->next;
      if (!p) break;
      fputs(", ", tex_file);
    }
  }
  else
    putc('?', tex_file);
}@}

@d Write referencing scrap numbers
@{{
  Scrap_Node *p = name->uses;
  fputs("{\\footnotesize ", tex_file);
  if (p) {
    fputs("Referenced in scrap", tex_file);
    if (p->next) {
      fputs("s ", tex_file);
      print_scrap_numbers(tex_file, p);
    }
    else
      fprintf(tex_file, " %d.", p->scrap);
  }
  else
    fputs("Not referenced.", tex_file);
  putc('}', tex_file);
}@}


@d Write index of user-specified names
@{{
  if (user_names) {
    fputs("\n\\begin{list}{}{\\setlength{\\itemsep}{-\\parsep}}\n", tex_file);
    format_user_entry(user_names, tex_file);
    fputs("\\end{list}", tex_file);
  }
  c = source_get();
}@}


@o latex.c
@{static void format_user_entry(name, tex_file)
     Name *name;
     FILE *tex_file;
{
  while (name) {
    format_user_entry(name->llink, tex_file);
    @<Format a user index entry@>
    name = name->rlink;
  }
}
@| format_user_entry @}


@d Format a user index entry
@{{
  Scrap_Node *uses = name->uses;
  Scrap_Node *defs = name->defs;
  fprintf(tex_file, "\\item \\hspace{-\\leftmargin}\\verb@@%s@@: ",
	  name->spelling);
  do {
    if (uses && (!defs || uses->scrap < defs->scrap)) {
      fprintf(tex_file, "%d", uses->scrap);
      uses = uses->next;
    }
    else if (defs && (!uses || defs->scrap <= uses->scrap)) {
      if (uses && defs->scrap == uses->scrap)
	uses = uses->next;
      fprintf(tex_file, "\\underline{%d}", defs->scrap);
      defs = defs->next;
    }
    if (uses || defs) fputs(", ", tex_file);
  } while (uses || defs);
  fputs(".\n", tex_file);
}@}



\section{Writing the Output Files} \label{output-files}

@d Function pro...
@{void write_files();
@}

@o output.c
@{void write_files(files)
     Name *files;
{
  while (files) {
    write_files(files->llink);
    @<Write out {\tt files->spelling}@>
    files = files->rlink;
  }
}
@| write_files @}

Due to the filename restrictions on the PC, just use {\tt nuweb~~~.tmp} as
the temporary filename. Since files are processed serially, this should
cause any harm to unix platforms.

@d Write out {\tt files->spelling}
@{{
  FILE *temp_file;
  char temp_name[12]="nuweb~~~.tmp";
/*  sprintf(temp_name, "%s~", files->spelling);  */
  temp_file = fopen(temp_name, "w");
  if (temp_file) {
    char indent_chars[500];
    write_scraps(temp_file, files->defs, 0, indent_chars,
		 files->debug_flag, files->tab_flag, files->indent_flag);
    fclose(temp_file);
    if (compare_flag)
      @<Compare the temp file and the old file@>
    else
      rename(temp_name, files->spelling);
  }
  else {
    fprintf(stderr, "%s: can't open %s\n", command_name, temp_name);
    exit(-1);
  }
}@}


@d Compare the temp file...
@{{
  FILE *old_file = fopen(files->spelling, "r");
  if (old_file) {
    int x, y;
    temp_file = fopen(temp_name, "r");
    do {
      x = getc(old_file);
      y = getc(temp_file);
    } while (x == y && x != EOF);
    fclose(old_file);
    fclose(temp_file);
    if (x == y)
      unlink(temp_name);
    else
      rename(temp_name, files->spelling);
  }
  else
    rename(temp_name, files->spelling);
}@}



\chapter{The Support Routines}

\section{Source Files} \label{source-files}

\subsection{Global Declarations}

We need two routines to handle reading the source files.
@d Function pro...
@{void source_open(); /* pass in the name of the source file */
int source_get();   /* no args; returns the next char or EOF */
@}


There are also two global variables maintained for use in error
messages and such.
@d Global...
@{char *source_name;  /* name of the current file */
int source_line;    /* current line in the source file */
@| source_name source_line @}

\subsection{Local Declarations}


@o input.c
@{static FILE *source_file;  /* the current input file */
static int source_peek;
static int double_at;
static int include_depth;
@| source_peek source_file double_at include_depth @}


@o input.c
@{struct {
  FILE *file;
  char *name;
  int line;
} stack[10];
@| stack @}


\subsection{Reading a File}

The routine {\tt source\_get} returns the next character from the
current source file. It notices newlines and keeps the line counter 
{\tt source\_line} up to date. It also catches {\tt EOF} and watches
for {\tt @@} characters. All other characters are immediately returned.
@o input.c
@{int source_get()
{
  int c = source_peek;
  switch (c) {
    case EOF:  @<Handle {\tt EOF}@>
	       return c;
    case '@@':  @<Handle an ``at'' character@>
	       return c;
    case '\n': source_line++;
    default:   source_peek = getc(source_file);
	       return c;
  }
}
@| source_get @}


This whole {\tt @@}~character handling mess is pretty annoying.
I want to recognize {\tt @@i} so I can handle include files correctly.
At the same time, it makes sense to recognize illegal {\tt @@}~sequences
and complain; this avoids ever having to check anywhere else.
Unfortunately, I need to avoid tripping over the {\tt @@@@}~sequence;
hence this whole unsatisfactory {\tt double\_at} business.
@d Handle an ``at''...
@{{
  c = getc(source_file);
  if (double_at) {
    source_peek = c;
    double_at = FALSE;
    c = '@@';
  }
  else
    switch (c) {
      case 'i': @<Open an include file@>
		break;
      case 'f': case 'm': case 'u':
      case 'd': case 'o': case 'D': case 'O':
      case '{': case '}': case '<': case '>': case '|':
		source_peek = c;
		c = '@@';
		break;
      case '@@': source_peek = c;
		double_at = TRUE;
		break;
      default:  fprintf(stderr, "%s: bad @@ sequence (%s, line %d)\n",
			command_name, source_name, source_line);
		exit(-1);
    }
}@}

@d Open an include file
@{{
  if (include_depth < 10) {
    char name[100];
    @<Collect include-file name@>
    stack[include_depth].name = source_name;
    stack[include_depth].file = source_file;
    stack[include_depth].line = source_line + 1;
    include_depth++;
    source_line = 1;
    source_name = save_string(name);
    source_file = fopen(source_name, "r");
    if (!source_file) {
      fprintf(stderr, "%s: can't open include file %s\n",
	      command_name, source_name);
      exit(-1);
    }
    source_peek = getc(source_file);
    c = source_get();
  }
  else {
    fprintf(stderr, "%s: include nesting too deep (%s, %d)\n",
	    command_name, source_name, source_line);
    exit(-1);
  }
}@}

@d Collect include-file name
@{{
    char *p = name;
    do 
      c = getc(source_file);
    while (c == ' ' || c == '\t');
    while (isgraph(c)) {
      *p++ = c;
      c = getc(source_file);
    }
    *p = '\0';
    if (c != '\n') {
      fprintf(stderr, "%s: unexpected characters after file name (%s, %d)\n",
	      command_name, source_name, source_line);
      exit(-1);
    }
}@}

If an {\tt EOF} is discovered, the current file must be closed and
input from the next stacked file must be resumed. If no more files are
on the stack, the {\tt EOF} is returned.
@d Handle {\tt EOF}
@{{
  fclose(source_file);
  if (include_depth) {
    include_depth--;
    source_file = stack[include_depth].file;
    source_line = stack[include_depth].line;
    source_name = stack[include_depth].name;
    source_peek = getc(source_file);
    c = source_get();
  }
}@}


\subsection{Opening a File}

The routine {\tt source\_open} takes a file name and tries to open the
file. If unsuccessful, it complains and halts. Otherwise, it sets 
{\tt source\_name}, {\tt source\_line}, and {\tt double\_at}.
@o input.c
@{void source_open(name)
     char *name;
{
  source_file = fopen(name, "r");
  if (!source_file) {
    fprintf(stderr, "%s: couldn't open %s\n", command_name, name);
    exit(-1);
  }
  source_name = name;
  source_line = 1;
  source_peek = getc(source_file);
  double_at = FALSE;
  include_depth = 0;
}
@| source_open @}




\section{Scraps} \label{scraps}


@o scraps.c
@{#define SLAB_SIZE 500

typedef struct slab {
  struct slab *next;
  char chars[SLAB_SIZE];
} Slab;
@| Slab SLAB_SIZE @}


@o scraps.c
@{typedef struct {
  char *file_name;
  int file_line;
  Slab *slab;
} ScrapEntry;
@| ScrapEntry @}

@o scraps.c
@{ScrapEntry *SCRAP[256];

#define scrap_array(i) SCRAP[(i) >> 8][(i) & 255]

int scraps;
@| scraps scrap_array SCRAP @}


@d Function pro...
@{void init_scraps();
int collect_scrap();
int write_scraps();
@}


@o scraps.c
@{void init_scraps()
{
  scraps = 1;
  SCRAP[0] = (ScrapEntry *) arena_getmem(256 * sizeof(ScrapEntry));
}
@| init_scraps @}


@o scraps.c
@{typedef struct {
  Slab *scrap;
  int index;
} Manager;
@| Manager @}



@o scraps.c
@{static void push(c, manager)
     char c;
     Manager *manager;
{
  Slab *scrap = manager->scrap;
  int index = manager->index;
  scrap->chars[index++] = c;
  if (index == SLAB_SIZE) {
    Slab *new = (Slab *) arena_getmem(sizeof(Slab));
    scrap->next = new;
    manager->scrap = new;
    index = 0;
  }
  manager->index = index;
}
@| push @}

@o scraps.c
@{static void pushs(s, manager)
     char *s;
     Manager *manager;
{
  while (*s)
    push(*s++, manager);
}
@| pushs @}


@o scraps.c
@{int collect_scrap()
{
  Manager writer;
  @<Create new scrap...@>
  @<Accumulate scrap and return {\tt scraps++}@>
}
@| collect_scrap @}

@d Create new scrap, managed by {\tt writer}
@{{
  Slab *scrap = (Slab *) arena_getmem(sizeof(Slab));
  if ((scraps & 255) == 0)
    SCRAP[scraps >> 8] = (ScrapEntry *) arena_getmem(256 * sizeof(ScrapEntry));
  scrap_array(scraps).slab = scrap;
  scrap_array(scraps).file_name = save_string(source_name);
  scrap_array(scraps).file_line = source_line;
  writer.scrap = scrap;
  writer.index = 0;
}@}


@d Accumulate scrap...
@{{
  int c = source_get();
  while (1) {
    switch (c) {
      case EOF: fprintf(stderr, "%s: unexpect EOF in scrap (%s, %d)\n",
			command_name, scrap_array(scraps).file_name,
			scrap_array(scraps).file_line);
		exit(-1);
      case '@@': @<Handle at-sign during scrap accumulation@>
		break;
      default:  push(c, &writer);
		c = source_get();
		break;
    }
  }
}@}


@d Handle at-sign during scrap accumulation
@{{
  c = source_get();
  switch (c) {
    case '@@': pushs("@@@@", &writer);
	      c = source_get();
	      break;
    case '|': @<Collect user-specified index entries@>
    case '}': push('\0', &writer);
	      return scraps++;
    case '<': @<Handle macro invocation in scrap@>
	      break;
    default : fprintf(stderr, "%s: unexpected @@%c in scrap (%s, %d)\n",
		      command_name, c, source_name, source_line);
	      exit(-1);
  }
}@}


@d Collect user-specified index entries
@{{
  do {
    char new_name[100];
    char *p = new_name;
    do 
      c = source_get();
    while (isspace(c));
    if (c != '@@') {
      Name *name;
      do {
	*p++ = c;
	c = source_get();
      } while (c != '@@' && !isspace(c));
      *p = '\0';
      name = name_add(&user_names, new_name);
      if (!name->defs || name->defs->scrap != scraps) {
	Scrap_Node *def = (Scrap_Node *) arena_getmem(sizeof(Scrap_Node));
	def->scrap = scraps;
	def->next = name->defs;
	name->defs = def;
      }
    }
  } while (c != '@@');
  c = source_get();
  if (c != '}') {
    fprintf(stderr, "%s: unexpected @@%c in scrap (%s, %d)\n",
	    command_name, c, source_name, source_line);
    exit(-1);
  }
}@}


@d Handle macro invocation in scrap
@{{
  Name *name = collect_scrap_name();
  @<Save macro name@>
  @<Add current scrap to {\tt name}'s uses@>
  c = source_get();
}@}


@d Save macro name
@{{
  char *s = name->spelling;
  int len = strlen(s) - 1;
  pushs("@@<", &writer);
  while (len > 0) {
    push(*s++, &writer);
    len--;
  }
  if (*s == ' ')
    pushs("...", &writer);
  else
    push(*s, &writer);
  pushs("@@>", &writer);
}@}


@d Add current scrap to...
@{{
  if (!name->uses || name->uses->scrap != scraps) {
    Scrap_Node *use = (Scrap_Node *) arena_getmem(sizeof(Scrap_Node));
    use->scrap = scraps;
    use->next = name->uses;
    name->uses = use;
  }
}@}


@o scraps.c
@{static char pop(manager)
     Manager *manager;
{
  Slab *scrap = manager->scrap;
  int index = manager->index;
  char c = scrap->chars[index++];
  if (index == SLAB_SIZE) {
    manager->scrap = scrap->next;
    index = 0;
  }
  manager->index = index;
  return c;
}
@| pop @}



@o scraps.c
@{static Name *pop_scrap_name(manager)
     Manager *manager;
{
  char name[100];
  char *p = name;
  int c = pop(manager);
  while (TRUE) {
    if (c == '@@')
      @<Check for end of scrap name and return@>
    else {
      *p++ = c;
      c = pop(manager);
    }
  }
}
@| pop_scrap_name @}


@d Check for end of scrap name...
@{{
  c = pop(manager);
  if (c == '@@') {
    *p++ = c;
    c = pop(manager);
  }
  else if (c == '>') {
    if (p - name > 3 && p[-1] == '.' && p[-2] == '.' && p[-3] == '.') {
      p[-3] = ' ';
      p -= 2;
    }
    *p = '\0';
    return prefix_add(&macro_names, name);
  }
  else {
    fprintf(stderr, "%s: found an internal problem (1)\n", command_name);
    exit(-1);
  }
}@}


@o scraps.c
@{int write_scraps(file, defs, global_indent, indent_chars,
		   debug_flag, tab_flag, indent_flag)
     FILE *file;
     Scrap_Node *defs;
     int global_indent;
     char *indent_chars;
     char debug_flag;
     char tab_flag;
     char indent_flag;
{
  int indent = 0;
  while (defs) {
    @<Copy {\tt defs->scrap} to {\tt file}@>
    defs = defs->next;
  }
  return indent + global_indent;
}
@| write_scraps @}


@d Copy {\tt defs->scrap}...
@{{
  char c;
  Manager reader;
  int line_number = scrap_array(defs->scrap).file_line;
  @<Insert debugging information if required@>
  reader.scrap = scrap_array(defs->scrap).slab;
  reader.index = 0;
  c = pop(&reader);
  while (c) {
    switch (c) {
      case '@@':  @<Check for macro invocation in scrap@>
		 break;
      case '\n': putc(c, file);
		 line_number++;
		 @<Insert appropriate indentation@>
		 break;
      case '\t': @<Handle tab...@>
		 break;
      default:	 putc(c, file);
		 indent_chars[global_indent + indent] = ' ';
		 indent++;
		 break;
    }
    c = pop(&reader);
  }
}@}


@d Insert debugging information if required
@{if (debug_flag) {
  fprintf(file, "\n#line %d \"%s\"\n",
	  line_number, scrap_array(defs->scrap).file_name);
  @<Insert appropr...@>
}@}


@d Insert approp...
@{{
  if (indent_flag) {
    if (tab_flag)
      for (indent=0; indent<global_indent; indent++)
	putc(' ', file);
    else
      for (indent=0; indent<global_indent; indent++)
	putc(indent_chars[indent], file);
  }
  indent = 0;
}@}


@d Handle tab characters on output
@{{
  if (tab_flag)
    @<Expand tab...@>
  else {
    putc('\t', file);
    indent_chars[global_indent + indent] = '\t';
    indent++;
  }
}@}



@d Check for macro invocation...
@{{
  c = pop(&reader);
  switch (c) {
    case '@@': putc(c, file);
	      indent_chars[global_indent + indent] = ' ';
	      indent++;
	      break;
    case '<': @<Copy macro into {\tt file}@>
	      @<Insert debugging information if required@>
	      break;
    default:  /* ignore, since we should already have a warning */
	      break;
  }
}@}


@d Copy macro into...
@{{
  Name *name = pop_scrap_name(&reader);
  if (name->mark) {
    fprintf(stderr, "%s: recursive macro discovered involving <%s>\n",
	    command_name, name->spelling);
    exit(-1);
  }
  if (name->defs) {
    name->mark = TRUE;
    indent = write_scraps(file, name->defs, global_indent + indent,
			  indent_chars, debug_flag, tab_flag, indent_flag);
    indent -= global_indent;
    name->mark = FALSE;
  }
  else if (!tex_flag)
    fprintf(stderr, "%s: macro never defined <%s>\n",
	    command_name, name->spelling);
}@}


\section{Names} \label{names}

@d Type de...
@{typedef struct scrap_node {
  struct scrap_node *next;
  int scrap;
} Scrap_Node;
@| Scrap_Node @}


@d Type de...
@{typedef struct name {
  char *spelling;
  struct name *llink;
  struct name *rlink;
  Scrap_Node *defs;
  Scrap_Node *uses;
  int mark;
  char tab_flag;
  char indent_flag;
  char debug_flag;
} Name;
@| Name @}

@d Global...
@{Name *file_names;
Name *macro_names;
Name *user_names;
@| file_names macro_names user_names @}

@d Function pro...
@{Name *collect_file_name();
Name *collect_macro_name();
Name *collect_scrap_name();
Name *name_add();
Name *prefix_add();
char *save_string();
void reverse_lists();
@}

@o names.c
@{enum { LESS, GREATER, EQUAL, PREFIX, EXTENSION };

static int compare(x, y)
     char *x;
     char *y;
{
  int len, result;
  int xl = strlen(x);
  int yl = strlen(y);
  int xp = x[xl - 1] == ' ';
  int yp = y[yl - 1] == ' ';
  if (xp) xl--;
  if (yp) yl--;
  len = xl < yl ? xl : yl;
  result = strncmp(x, y, len);
  if (result < 0) return GREATER;
  else if (result > 0) return LESS;
  else if (xl < yl) {
    if (xp) return EXTENSION;
    else return LESS;
  }
  else if (xl > yl) {
    if (yp) return PREFIX;
    else return GREATER;
  }
  else return EQUAL;
}
@| compare LESS GREATER EQUAL PREFIX EXTENSION @}


@o names.c
@{char *save_string(s)
     char *s;
{
  char *new = (char *) arena_getmem((strlen(s) + 1) * sizeof(char));
  strcpy(new, s);
  return new;
}
@| save_string @}

@o names.c
@{Name *prefix_add(root, spelling)
     Name **root;
     char *spelling;
{
  Name *node = *root;
  while (node) {
    switch (compare(node->spelling, spelling)) {
    case GREATER:   root = &node->rlink;
		    break;
    case LESS:      root = &node->llink;
		    break;
    case EQUAL:     return node;
    case EXTENSION: node->spelling = save_string(spelling);
		    return node;
    case PREFIX:    @<Check for ambiguous prefix@>
		    return node;
    }
    node = *root;
  }
  @<Create new name entry@>
}
@| prefix_add @}

Since a very short prefix might match more than one macro name, I need
to check for other matches to avoid mistakes. Basically, I simply
continue the search down {\em both\/} branches of the tree.

@d Check for ambiguous prefix
@{{
  if (ambiguous_prefix(node->llink, spelling) ||
      ambiguous_prefix(node->rlink, spelling))
    fprintf(stderr,
	    "%s: ambiguous prefix @@<%s...@@> (%s, line %d)\n",
	    command_name, spelling, source_name, source_line);
}@}

@o names.c
@{static int ambiguous_prefix(node, spelling)
     Name *node;
     char *spelling;
{
  while (node) {
    switch (compare(node->spelling, spelling)) {
    case GREATER:   node = node->rlink;
		    break;
    case LESS:      node = node->llink;
		    break;
    case EQUAL:
    case EXTENSION:
    case PREFIX:    return TRUE;
    }
  }
  return FALSE;
}
@}

Rob Shillingsburg suggested that I organize the index of
user-specified identifiers more traditionally; that is, not relying on
strict {\small ASCII} comparisons via {\tt strcmp}. Ideally, we'd like
to see the index ordered like this:
\begin{quote}
\begin{flushleft}
aardvark \\
Adam \\
atom \\
Atomic \\
atoms
\end{flushleft}
\end{quote}
The function {\tt robs\_strcmp} implements the desired predicate.

@o names.c
@{static int robs_strcmp(x, y)
     char *x;
     char *y;
{
  char *xx = x;
  char *yy = y;
  int xc = toupper(*xx);
  int yc = toupper(*yy);
  while (xc == yc && xc) {
    xx++;
    yy++;
    xc = toupper(*xx);
    yc = toupper(*yy);
  }
  if (xc != yc) return xc - yc;
  xc = *x;
  yc = *y;
  while (xc == yc && xc) {
    x++;
    y++;
    xc = *x;
    yc = *y;
  }
  if (isupper(xc) && islower(yc))
    return xc * 2 - (toupper(yc) * 2 + 1);
  if (islower(xc) && isupper(yc))
    return toupper(xc) * 2 + 1 - yc * 2;
  return xc - yc;
}
@| robs_strcmp @}

@o names.c
@{Name *name_add(root, spelling)
     Name **root;
     char *spelling;
{
  Name *node = *root;
  while (node) {
    int result = robs_strcmp(node->spelling, spelling);
    if (result > 0)
      root = &node->llink;
    else if (result < 0)
      root = &node->rlink;
    else
      return node;
    node = *root;
  }
  @<Create new name entry@>
}
@| name_add @}


@d Create new name...
@{{
  node = (Name *) arena_getmem(sizeof(Name));
  node->spelling = save_string(spelling);
  node->mark = FALSE;
  node->llink = NULL;
  node->rlink = NULL;
  node->uses = NULL;
  node->defs = NULL;
  node->tab_flag = TRUE;
  node->indent_flag = TRUE;
  node->debug_flag = FALSE;
  *root = node;
  return node;
}@}


Name terminated by whitespace.  Also check for ``per-file'' flags. Keep
skipping white space until we reach scrap.
@o names.c
@{Name *collect_file_name()
{
  Name *new_name;
  char name[100];
  char *p = name;
  int start_line = source_line;
  int c = source_get();
  while (isspace(c))
    c = source_get();
  while (isgraph(c)) {
    *p++ = c;
    c = source_get();
  }
  if (p == name) {
    fprintf(stderr, "%s: expected file name (%s, %d)\n",
	    command_name, source_name, start_line);
    exit(-1);
  }
  *p = '\0';
  new_name = name_add(&file_names, name);
  @<Handle optional per-file flags@>
  if (c == '@@' && source_get() == '{')
    return new_name;
  else {
    fprintf(stderr, "%s: expected @@{ after file name (%s, %d)\n",
	    command_name, source_name, start_line);
    exit(-1);
  }
}
@| collect_file_name @}

@d Handle optional per-file flags
@{{
  while (1) {
    while (isspace(c))
      c = source_get();
    if (c == '-') {
      c = source_get();
      do {
	switch (c) {
	  case 't': new_name->tab_flag = FALSE;
		    break;
	  case 'd': new_name->debug_flag = TRUE;
		    break;
	  case 'i': new_name->indent_flag = FALSE;
		    break;
	  default : fprintf(stderr, "%s: unexpected per-file flag (%s, %d)\n",
			    command_name, source_name, source_line);
		    break;
	}
	c = source_get();
      } while (!isspace(c));
    }
    else break;
  }
}@}



Name terminated by \verb+\n+ or \verb+@@{+; but keep skipping until \verb+@@{+
@o names.c
@{Name *collect_macro_name()
{
  char name[100];
  char *p = name;
  int start_line = source_line;
  int c = source_get();
  while (isspace(c))
    c = source_get();
  while (c != EOF) {
    switch (c) {
      case '@@':  @<Check for terminating at-sequence and return name@>
		 break;
      case '\t':
      case ' ':  *p++ = ' ';
		 do
		   c = source_get();
		 while (c == ' ' || c == '\t');
		 break;
      case '\n': @<Skip until scrap begins, then return name@>
      default:	 *p++ = c;
		 c = source_get();
		 break;
    }
  }
  fprintf(stderr, "%s: expected macro name (%s, %d)\n",
	  command_name, source_name, start_line);
  exit(-1);
}
@| collect_macro_name @}


@d Check for termina...
@{{
  c = source_get();
  switch (c) {
    case '@@': *p++ = c;
	      break;
    case '{': @<Cleanup and install name@>
    default:  fprintf(stderr,
		      "%s: unexpected @@%c in macro name (%s, %d)\n",
		      command_name, c, source_name, start_line);
	      exit(-1);
  }
}@}


@d Cleanup and install name
@{{
  if (p > name && p[-1] == ' ')
    p--;
  if (p - name > 3 && p[-1] == '.' && p[-2] == '.' && p[-3] == '.') {
    p[-3] = ' ';
    p -= 2;
  }
  if (p == name || name[0] == ' ') {
    fprintf(stderr, "%s: empty scrap name (%s, %d)\n",
	    command_name, source_name, source_line);
    exit(-1);
  }
  *p = '\0';
  return prefix_add(&macro_names, name);
}@}


@d Skip until scrap...
@{{
  do
    c = source_get();
  while (isspace(c));
  if (c == '@@' && source_get() == '{')
    @<Cleanup and install name@>
  else {
    fprintf(stderr, "%s: expected @@{ after macro name (%s, %d)\n",
	    command_name, source_name, start_line);
    exit(-1);
  }
}@}


Terminated by \verb+@@>+
@o names.c
@{Name *collect_scrap_name()
{
  char name[100];
  char *p = name;
  int c = source_get();
  while (c == ' ' || c == '\t')
    c = source_get();
  while (c != EOF) {
    switch (c) {
      case '@@':  @<Look for end of scrap name and return@>
		 break;
      case '\t':
      case ' ':  *p++ = ' ';
		 do
		   c = source_get();
		 while (c == ' ' || c == '\t');
		 break;
      default:	 if (isgraph(c)) {
		   *p++ = c;
		   c = source_get();
		 }
		 else {
		   fprintf(stderr,
			   "%s: unexpected character in macro name (%s, %d)\n",
			   command_name, source_name, source_line);
		   exit(-1);
		 }
		 break;
    }
  }
  fprintf(stderr, "%s: unexpected end of file (%s, %d)\n",
	  command_name, source_name, source_line);
  exit(-1);
}
@| collect_scrap_name @}


@d Look for end of scrap name...
@{{
  c = source_get();
  switch (c) {
    case '@@': *p++ = c;
	      c = source_get();
	      break;
    case '>': @<Cleanup and install name@>
    default:  fprintf(stderr,
		      "%s: unexpected @@%c in macro name (%s, %d)\n",
		      command_name, c, source_name, source_line);
	      exit(-1);
  }
}@}


@o names.c
@{static Scrap_Node *reverse();	/* a forward declaration */

void reverse_lists(names)
     Name *names;
{
  while (names) {
    reverse_lists(names->llink);
    names->defs = reverse(names->defs);
    names->uses = reverse(names->uses);
    names = names->rlink;
  }
}
@| reverse_lists @}

Just for fun, here's a non-recursive version of the traditional list
reversal code. Note that it reverses the list in place; that is, it
does no new allocations.
@o names.c
@{static Scrap_Node *reverse(a)
     Scrap_Node *a;
{
  if (a) {
    Scrap_Node *b = a->next;
    a->next = NULL;
    while (b) {
      Scrap_Node *c = b->next;
      b->next = a;
      a = b;
      b = c;
    }
  }
  return a;
}
@| reverse @}


\section{Searching for Index Entries} \label{search}

Given the array of scraps and a set of index entries, we need to
search all the scraps for occurences of each entry. The obvious
approach to this problem would be quite expensive for large documents;
however, there is an interesting  paper describing an efficient
solution~\cite{aho:75}.


@o scraps.c
@{
typedef struct name_node {
  struct name_node *next;
  Name *name;
} Name_Node;

typedef struct goto_node {
  struct goto_node *next;	/* next goto node with same depth */
  struct goto_node *fail;
  struct move_node *moves;
  Name_Node *output;
} Goto_Node;

typedef struct move_node {
  struct move_node *next;
  Goto_Node *state;
  char c;
} Move_Node;
@| Goto_Node Move_Node Name_Node @}

@o scraps.c
@{
static Goto_Node *root[128];
static int max_depth;
static Goto_Node **depths;
@| root max_depth @}


@o scraps.c
@{static Goto_Node *goto_lookup(c, g)
     char c;
     Goto_Node *g;
{
  Move_Node *m = g->moves;
  while (m && m->c != c)
    m = m->next;
  if (m)
    return m->state;
  else
    return NULL;
}
@| goto_lookup @}


\subsection{Building the Automata}


@o scraps.c
@{
static void build_gotos();

void search()
{
  int i;
  for (i=0; i<128; i++)
    root[i] = NULL;
  max_depth = 10;
  depths = (Goto_Node **) arena_getmem(max_depth * sizeof(Goto_Node *));
  for (i=0; i<max_depth; i++)
    depths[i] = NULL;
  build_gotos(user_names);
  @<Build failure functions@>
  @<Search scraps@>
}
@| search @}



@o scraps.c
@{static void build_gotos(tree)
     Name *tree;
{
  while (tree) {
    @<Extend goto graph with {\tt tree->spelling}@>
    build_gotos(tree->rlink);
    tree = tree->llink;
  }
}
@| build_gotos @}

@d Extend goto...
@{{
  int depth = 2;
  char *p = tree->spelling;
  char c = *p++;
  Goto_Node *q = root[c];
  if (!q) {
    q = (Goto_Node *) arena_getmem(sizeof(Goto_Node));
    root[c] = q;
    q->moves = NULL;
    q->fail = NULL;
    q->moves = NULL;
    q->output = NULL;
    q->next = depths[1];
    depths[1] = q;
  }
  while (c = *p++) {
    Goto_Node *new = goto_lookup(c, q);
    if (!new) {
      Move_Node *new_move = (Move_Node *) arena_getmem(sizeof(Move_Node));
      new = (Goto_Node *) arena_getmem(sizeof(Goto_Node));
      new->moves = NULL;
      new->fail = NULL;
      new->moves = NULL;
      new->output = NULL;
      new_move->state = new;
      new_move->c = c;
      new_move->next = q->moves;
      q->moves = new_move;
      if (depth == max_depth) {
	int i;
	Goto_Node **new_depths =
	    (Goto_Node **) arena_getmem(2*depth*sizeof(Goto_Node *));
	max_depth = 2 * depth;
	for (i=0; i<depth; i++)
	  new_depths[i] = depths[i];
	depths = new_depths;
	for (i=depth; i<max_depth; i++)
	  depths[i] = NULL;
      }
      new->next = depths[depth];
      depths[depth] = new;
    }
    q = new;
    depth++;
  }
  q->output = (Name_Node *) arena_getmem(sizeof(Name_Node));
  q->output->next = NULL;
  q->output->name = tree;
}@}


@d Build failure functions
@{{
  int depth;
  for (depth=1; depth<max_depth; depth++) {
    Goto_Node *r = depths[depth];
    while (r) {
      Move_Node *m = r->moves;
      while (m) {
	char a = m->c;
	Goto_Node *s = m->state;
	Goto_Node *state = r->fail;
	while (state && !goto_lookup(a, state))
	  state = state->fail;
	if (state)
	  s->fail = goto_lookup(a, state);
	else
	  s->fail = root[a];
	if (s->fail) {
	  Name_Node *p = s->fail->output;
	  while (p) {
	    Name_Node *q = (Name_Node *) arena_getmem(sizeof(Name_Node));
	    q->name = p->name;
	    q->next = s->output;
	    s->output = q;
	    p = p->next;
	  }
	}
	m = m->next;
      }
      r = r->next;
    }
  }
}@}


\subsection{Searching the Scraps}

@d Search scraps
@{{
  for (i=1; i<scraps; i++) {
    char c;
    Manager reader;
    Goto_Node *state = NULL;
    reader.scrap = scrap_array(i).slab;
    reader.index = 0;
    c = pop(&reader);
    while (c) {
      while (state && !goto_lookup(c, state))
	state = state->fail;
      if (state)
	state = goto_lookup(c, state);
      else
	state = root[c];
      if (state && state->output) {
	Name_Node *p = state->output;
	do {
	  Name *name = p->name;
	  if (!name->uses || name->uses->scrap != i) {
	    Scrap_Node *new_use =
		(Scrap_Node *) arena_getmem(sizeof(Scrap_Node));
	    new_use->scrap = i;
	    new_use->next = name->uses;
	    name->uses = new_use;
	  }
	  p = p->next;
	} while (p);
      }
      c = pop(&reader);
    }
  }
}@}


\section{Memory Management} \label{memory-management}

I manage memory using a simple scheme inspired by Hanson's idea of
{\em arenas\/}~\cite{hanson:90}.
Basically, I allocate all the storage required when processing a
source file (primarily for names and scraps) using calls to 
{\tt arena\_getmem(n)}, where {\tt n} specifies the number of bytes to
be allocated. When the storage is no longer required, the entire arena
is freed with a single call to  {\tt arena\_free()}. Both operations
are quite fast.
@d Function p...
@{void *arena_getmem();
void arena_free();
@}


@o arena.c
@{typedef struct chunk {
  struct chunk *next;
  char *limit;
  char *avail;
} Chunk;
@| Chunk @}


We define an empty chunk called {\tt first}. The variable {\tt arena} points
at the current chunk of memory; it's initially pointed at {\tt first}.
As soon as some storage is required, a ``real'' chunk of memory will
be allocated and attached to {\tt first->next}; storage will be
allocated from the new chunk (and later chunks if necessary).
@o arena.c
@{Chunk first = { NULL, NULL, NULL };
Chunk *arena = &first;
@| first arena @}


\subsection{Allocating Memory}

The routine {\tt arena\_getmem(n)} returns a pointer to (at least) 
{\tt n} bytes of memory. Note that {\tt n} is rounded up to ensure
that returned pointers are always aligned.
@o arena.c
@{void *arena_getmem(n)
     int n;
{
  char *q;
  char *p = arena->avail;
  n = (n + 3) & ~3;		/* ensuring alignment to 4 bytes */
  q = p + n;
  if (q <= arena->limit) {
    arena->avail = q;
    return p;
  }
  @<Find a new chunk of memory@>
}
@| arena_getmem @}


If the current chunk doesn't have adequate space (at least {\tt n}
bytes) we examine the rest of the list of chunks (starting at 
{\tt arena->next}) looking for a chunk with adequate space. If {\tt n}
is very large, we may not find it right away or we may not find a
suitable chunk at all.
@d Find a new chunk...
@{{
  Chunk *ap = arena;
  Chunk *np = ap->next;
  while (np) {
    char *v = sizeof(Chunk) + (char *) np;
    if (v + n <= np->limit) {
      np->avail = v + n;
      arena = np;
      return v;
    }
    ap = np;
    np = ap->next;
  }
  @<Allocate a new chunk of memory@>
}@}


If there isn't a suitable chunk of memory on the free list, then we
need to allocate a new one.
@d Allocate a new ch...
@{{
  int m = n + 10000;
  np = (Chunk *) malloc(m);
  np->limit = m + (char *) np;
  np->avail = n + sizeof(Chunk) + (char *) np;
  np->next = NULL;
  ap->next = np;
  arena = np;
  return sizeof(Chunk) + (char *) np;
}@}


\subsection{Freeing Memory}

To free all the memory in the arena, we need only point {\tt arena}
back to the first empty chunk.
@o arena.c
@{void arena_free()
{
  arena = &first;
}
@| arena_free @}


\chapter{Indices} \label{indices}

Three sets of indices can be created automatically: an index of file
names, an index of macro names, and an index of user-specified
identifiers. An index entry includes the name of the entry, where it
was defined, and where it was referenced.



\section{Files}

@f

\section{Macros}

@m

\section{Identifiers}

Knuth prints his index of indentifiers in a two-column format.
I could force this automatically by emitting the \verb|\twocolumn|
command; but this has the side effect of forcing a new page.
Therefore, it seems better to leave it this up to the user.

@u

\bibliographystyle{plain}
\bibliography{literate}

\end{document}