% Copyright (C) 1988, all rights reserved.

\def\BibTeX{{\rm B\kern-.05em{\sc i\kern-.025em b}\kern-.08em
    T\kern-.1667em\lower.7ex\hbox{E}\kern-.125emX}}

\title{Designing B\kern-.05em{\large I}\kern-.025em{\large B}\kern-.08em\TeX\
									Styles}
\author{Oren Patashnik}
\date{February 8, 1988}

\documentstyle{article}
\begin{document}

\maketitle

\setcounter{section}{4}
\section{Bibliography-style hacking}
\label{style}

This document starts (and ends) with Section~\ref{style},
because in reality it is the final section of ``\BibTeX ing''~\cite{btxdoc},
the general documentation for \BibTeX.
But that document was meant for all \BibTeX\ users,
while this one is just for style designers,
so the two are physically separate.
Still, you should be completely familiar with ``\BibTeX ing''$\!$,
and all references in this document
to sections and section numbers
assume that the two documents are one.

This section,
along with the standard-style documentation file \hbox{\tt btxbst.doc},
should explain how to modify
existing style files and to produce new ones.
If you're a serious style hacker you should be familiar
with van~Leunen~\cite{van-leunen} for points of style,
with Lamport~\cite{latex} and Knuth~\cite{texbook} for formatting matters,
and perhaps with {\em Scribe\/}~\cite{scribe} for compatibility details.
And while you're at it, if you don't read the great little book by Strunk and
White~\cite{strunk-and-white}, you should at least look at its
entries in the database and the reference list
to see how \BibTeX\ handles multiple names.

To create a new style,
it's best to start with an existing style that's close to yours,
and then modify that.
This is true even if you're simply updating an old style
for \BibTeX\ version 0.99
(I've updated four nonstandard styles,
so I say this with some experience).
If you want to insert into a new style
some function you'd written for an old (version 0.98i) style,
keep in mind that the order of the arguments to
the assignment ({\tt :=}) function has been reversed.
When you're finished with your style,
you may want to try running it on the entire \hbox{\tt XAMPL.BIB} database
to make sure it handles all the standard entry types.

If you find any bugs in the standard styles,
or if there are things you'd like to do
with bibliography-style files but can't,
please complain to Oren Patashnik.


\subsection{General description}

You write bibliography styles in a postfix stack language.  It's
not too hard to figure out how by looking at the standard-style documentation,
but this description fills in a few details (it will fill in more
details if there's a demand for it).

Basically the style file is a program, written in an unnamed language, that
tells \BibTeX\ how to format the entries that will go in the reference list
(henceforth ``the entries'' will be ``the entry list''
or simply ``the list''$\!$, context permitting).
This programming language has ten commands, described in the next subsection.
These commands manipulate the language's objects:
constants, variables, functions, the stack, and the entry list.
(Warning: The terminology in this documentation,
chosen for ease of explanation, is slightly different from \BibTeX's.
For example, this documentation's ``variables'' and ``functions''
are both ``functions'' to \BibTeX.
Keep this in mind when interpreting \BibTeX's error messages.)

There are two types of functions: {\it built-in\/} ones that \BibTeX\ provides
(these are described in Section~\ref{built-in-fns}), and ones you define
using either the \hbox{\tt MACRO} or \hbox{\tt FUNCTION} command.

Your most time-consuming task, as a style designer,
will be creating or modifying functions
using the \hbox{\tt FUNCTION} command
(actually, becoming familiar with the references listed above will be
more time consuming, but assume for the moment that that's done).

Let's look at a sample function fragment.
Suppose you have a string variable named \hbox{\tt label}
and an integer variable named \hbox{\tt lab.width},
and suppose you want to append the character `{\tt a}' to \hbox{\tt label}
and to increment \hbox{\tt lab.width}:
\begin{verbatim}
    .  .  .
    label "a" * 'label :=          % label := label * "a"
    lab.width #1 + 'lab.width :=   % lab.width := lab.width + 1
    .  .  .
\end{verbatim}
In the first line,
\hbox{\tt label} pushes that variable's value onto the stack.
Next, the {\tt "a"} pushes the string constant `{\tt a}' onto the stack.
Then the built-in function {\tt *} pops the top two strings and
pushes their concatenation.
The \hbox{\tt 'label} pushes that variable's name onto the stack.
And finally, the built-in function {\tt :=} pops
the variable name and the concatenation and performs the assignment.
\BibTeX\ treats the stuff following the {\tt \%} as a comment
in the style file.
The second line is similar except that it uses {\tt \#1},
with no spaces intervening between the `{\tt \#}' and the `{\tt 1}'$\!$,
to push this integer constant.

The nonnull spacing here is arbitrary: multiple spaces, tabs, or newlines
are equivalent to a single one (except that you're probably better off
not having blank lines within commands, as explained shortly).

For string constants, absolutely any printing character
is legal between two consecutive double quotes, but \BibTeX\ here
(and only here) treats upper- and lower-case equivalents as different.
Furthermore, spacing {\em is\/} relevant within a string constant,
and you mustn't split a string constant across lines
(that is, the beginning and ending double quotes must be on the same line).

Variable and function names may not begin with a numeral and
may not contain any of the ten restricted characters
on page~143 of the \LaTeX\ book,
but may otherwise contain any printing characters.
Also, \BibTeX\ considers upper- and lower-case equivalents to be the same.

Integers and strings are the only value types for constants and variables
(booleans are implemented simply as 0-or-1 integers).
There are three kinds of variables:
\begin{description}

\item[global variables\hfill] These are either integer- or string-valued,
declared using an \hbox{\tt INTEGERS} or \hbox{\tt STRINGS} command.

\item[entry variables\hfill] These are either integer- or string-valued,
declared using the \hbox{\tt ENTRY} command.
Each has a value for each entry on the list
(example: a variable \hbox{\tt label} might store
the label string you'll use for the entry).

\item[fields\hfill] These are string-valued, read-only variables
that store the information from the database file;
their values are set by the \hbox{\tt READ} command.
As with entry variables, each has a value for each entry.
\end{description}


\subsection{Commands}

There are ten style-file commands:
Five (\hbox{\tt ENTRY}, \hbox{\tt FUNCTION}, \hbox{\tt INTEGERS},
\hbox{\tt MACRO}, and \hbox{\tt STRINGS})
declare and define variables and functions;
one (\hbox{\tt READ}) reads in the database information;
and four (\hbox{\tt EXECUTE}, \hbox{\tt ITERATE}, \hbox{\tt REVERSE},
and \hbox{\tt SORT}) manipulate the entries and produce output.
Although the command names appear here in upper case,
\BibTeX\ ignores case differences.

Some restrictions:
There must be exactly one \hbox{\tt ENTRY} and one \hbox{\tt READ} command;
the \hbox{\tt ENTRY} command, all \hbox{\tt MACRO} commands,
and certain \hbox{\tt FUNCTION} commands
(see next subsection's description of \hbox{\tt call.type\$})
must precede the \hbox{\tt READ} command;
and the \hbox{\tt READ} command must precede the four that
manipulate the entries and produce output.

Also it's best (but not essential) to leave at least one blank line
between commands and to leave no blank lines within a command;
this helps \BibTeX\ recover from any syntax errors you make.

You must enclose each argument of every command in braces.
Look at the standard-style documentation
for syntactic issues not described in this section.
Here are the ten commands:
\begin{description}

\item[\hbox{\tt ENTRY}\hfill]
Declares the fields and entry variables.
It has three arguments, each a (possibly empty) list of variable names.
The three lists are of:
fields, integer entry variables, and string entry variables.
There is an additional field that \BibTeX\ automatically
declares, \hbox{\tt crossref}, used for cross referencing.
And there is an additional string entry variable automatically declared,
\hbox{\tt sort.key\$}, used by the \hbox{\tt SORT} command.
Each of these variables has a value for each entry on the list.

\item[\hbox{\tt EXECUTE}\hfill]
Executes a single function.
It has one argument, the function name.

\item[\hbox{\tt FUNCTION}\hfill]
Defines a new function.
It has two arguments; the first is the function's name and the
second is its definition.
You must define a function before using it;
recursive functions are thus illegal.

\item[\hbox{\tt INTEGERS}\hfill]
Declares global integer variables.
It has one argument, a list of variable names.
There are two such automatically-declared variables,
\hbox{\tt entry.max\$} and \hbox{\tt global.max\$},
used for limiting the lengths of string variables.
You may have any number of these commands, but a variable's declaration
must precede its use.

\item[\hbox{\tt ITERATE}\hfill]
Executes a single function, once
for each entry in the list, in the list's current order
(initially the list is in citation order, but the \hbox{\tt SORT}
command may change this).
It has one argument, the function name.

\item[\hbox{\tt MACRO}\hfill]
Defines a string macro.
It has two arguments; the first is the macro's name, which is treated like
any other variable or function name,
and the second is its definition, which must be double-quote-delimited.
You must have one for each three-letter month abbreviation;
in addition, you should have one for common journal names.
The user's database may override any definition you define using this command.
If you want to define a string the user can't touch,
use the \hbox{\tt FUNCTION} command, which has a compatible syntax.

\item[\hbox{\tt READ}\hfill]
Dredges up from the database file
the field values for each entry in the list.
It has no arguments.
If a database entry doesn't have a value for a field
(and probably no database entry will have a value for every field),
that field variable is marked as missing for the entry.

\item[\hbox{\tt REVERSE}\hfill]
Exactly the same as the
\hbox{\tt ITERATE} command except that it executes the function
on the entry list in reverse order.

\item[\hbox{\tt SORT}\hfill]
Sorts the entry list using
the values of the string entry variable \hbox{\tt sort.key\$}.
It has no arguments.

\item[\hbox{\tt STRINGS}\hfill]
Declares global string variables.
It has one argument, a list of variable names.
You may have any number of these commands, but a variable's declaration
must precede its use.
\end{description}


\subsection{The built-in functions}
\label{built-in-fns}

Before we get to the built-in functions,
a few words about some other built-in objects.
There is one built-in string entry variable, \hbox{\tt sort.key\$},
which the style program must set if the style is to do sorting.
There is one built-in field, \hbox{\tt crossref},
used for the cross referencing feature
described in Section~4.
And there are two built-in integer global variables,
\hbox{\tt entry.max\$} and \hbox{\tt global.max\$},
which are set by default to some internal \BibTeX\ constants;
you should truncate strings to these lengths before
you assign to string variables,
so as to not generate any \BibTeX\ warning messages.

There are currently 37 built-in functions.
Every built-in function with a letter in its name ends with a `{\tt \$}'$\!$.
In what follows, ``first''$\!$, ``second''$\!$,
and so on refer to the order popped.
A ``literal'' is an element on the stack, and it will be either
an integer value, a string value, a variable or function name,
or a special value denoting a missing field.
If any popped literal has an incorrect type, \BibTeX\ complains and pushes
the integer 0 or the null string, depending on whether the function
was supposed to push an integer or string.
\begin{description}

\item[\hbox{\tt >}\hfill]
Pops the top two (integer) literals,
compares them, and pushes the integer 1 if the second is greater than
the first, 0 otherwise.

\item[\hbox{\tt <}\hfill]
Analogous.

\item[\hbox{\tt =}\hfill]
Pops the top two (both integer or both string) literals,
compares them,
and pushes the integer 1 if they're equal, 0 otherwise.

\item[\hbox{\tt +}\hfill]
Pops the top two (integer) literals and pushes their sum.

\item[\hbox{\tt -}\hfill]
Pops the top two (integer) literals and pushes their difference
(the first subtracted from the second).

\item[\hbox{\tt *}\hfill]
Pops the top two (string) literals,
concatenates them (in reverse order, that is, the order in which
pushed), and pushes the resulting string.

\item[\hbox{\tt :=}\hfill]
Pops the top two literals and assigns
to the first (which must be a global or entry variable)
the value of the second.

\item[\hbox{\tt add.period\$}\hfill]
Pops the top (string) literal,
adds a `{\tt .}' to it if the last non`{\tt \}}' character
isn't a `{\tt .}'$\!$, `{\tt ?}', or `{\tt !}'$\!$,
and pushes this resulting string.

\item[\hbox{\tt call.type\$}\hfill]
Executes the function whose name is the entry type of an entry.
For example if an entry is of type {\tt book}, this function executes
the {\tt book} function.
When given as an argument to the \hbox{\tt ITERATE} command,
\hbox{\tt call.type\$} actually produces the output for the entries.
For an entry with an unknown type,
it executes the function \hbox{\tt default.type}.
Thus you should define (before the \hbox{\tt READ} command) one function
for each standard entry type as well as a \hbox{\tt default.type} function.

\item[\hbox{\tt change.case\$}\hfill]
Pops the top two (string) literals;
it changes the case of the second according to the
specifications of the first, as follows.  (Note: The word `letters' in
the next sentence refers only to those at brace-level~0, the top-most
brace level; no other characters are changed, except perhaps for
``special characters''$\!$, described in Section~4.)
If the first literal is the
string~`{\tt t}'$\!$, it converts to lower case all letters except the very
first character in the string, which it leaves alone, and except the
first character following any colon and then nonnull white space,
which it also leaves alone; if it's the string~`{\tt l}'$\!$, it converts all
letters to lower case; and if it's the string~`{\tt u}'$\!$, it converts all
letters to upper case.
It then pushes this resulting string.  If either
type is incorrect, it complains and pushes the null string; however,
if both types are correct but the specification string (i.e., the
first string) isn't one of the legal ones, it merely pushes the second
back onto the stack, after complaining.  (Another note: It ignores
case differences in the specification string; for example, the strings
{\tt t} and {\tt T} are equivalent for the purposes of this built-in
function.)

\item[\hbox{\tt chr.to.int\$}\hfill]
Pops the top (string) literal,
makes sure it's a single character, converts it to the
corresponding ASCII integer, and pushes this integer.

\item[\hbox{\tt cite\$}\hfill]
Pushes the string that was the
\hbox{\verb|\cite|}-command argument for this entry.

\item[\hbox{\tt duplicate\$}\hfill]
Pops the top literal from the stack and pushes two copies of it.

\item[\hbox{\tt empty\$}\hfill]
Pops the top literal and pushes
the integer 1 if it's a missing field or a string having no
non-white-space characters, 0 otherwise.

\item[\hbox{\tt format.name\$}\hfill]
Pops the top three literals
(they are a string, an integer, and a string literal).
The last string literal represents a name list (each name
corresponding to a person), the integer literal specifies which name
to pick from this list, and the first string literal specifies how to
format this name, as explained in the next subsection.
Finally, this function pushes the formatted name.

\item[\hbox{\tt if\$}\hfill]
Pops the top three literals (they
are two function literals and an integer literal, in that order);
if the integer is greater than 0, it executes the second literal,
else it executes the first.

\item[\hbox{\tt int.to.chr\$}\hfill]
Pops the top (integer) literal,
interpreted as the ASCII integer value of a single character,
converts it to the corresponding single-character string, and pushes
this string.

\item[\hbox{\tt int.to.str\$}\hfill]
Pops the top (integer) literal,
converts it to its (unique) string equivalent, and pushes this string.

\item[\hbox{\tt missing\$}\hfill]
Pops the top literal and
pushes the integer 1 if it's a missing field, 0~otherwise.

\item[\hbox{\tt newline\$}\hfill]
Writes onto the {\tt bbl} file
what's accumulated in the output buffer.
It writes a blank line if and only if the output buffer is empty.
Since \hbox{\tt write\$} does reasonable line breaking, you should use
this function only when you want a blank line or an explicit line break.

\item[\hbox{\tt num.names\$}\hfill]
Pops the top (string) literal
and pushes the number of names the string represents---one plus
the number of occurrences of the substring ``and'' (ignoring case differences)
surrounded by nonnull white-space at the top brace level.

\item[\hbox{\tt pop\$}\hfill]
Pops the top of the stack but
doesn't print it; this gets rid of an unwanted stack literal.

\item[\hbox{\tt preamble\$}\hfill]
Pushes onto the stack the concatenation of all the
\hbox{\tt @PREAMBLE} strings read from the database files.

\item[\hbox{\tt purify\$}\hfill]
Pops the top (string) literal,
removes nonalphanumeric characters except for white-space characters and
hyphens and ties (these all get converted to a space), removes
certain alphabetic characters contained in the control sequences
associated with a ``special character''$\!$, and pushes the resulting string.

\item[\hbox{\tt quote\$}\hfill]
Pushes the string consisting of the double-quote character.

\item[\hbox{\tt skip\$}\hfill]
Is a no-op.

\item[\hbox{\tt stack\$}\hfill]
Pops and prints the whole stack;
it's meant to be used for style designers while debugging.

\item[\hbox{\tt substring\$}\hfill]
Pops the top three literals
(they are the two integers literals {\it len\/} and {\it start}, and a
string literal, in that order).
It pushes the substring of the (at most) {\it len\/} consecutive characters
starting at the {\it start\/}th character (assuming 1-based indexing)
if {\it start\/} is positive, and ending at the $-${\it start\/}th character
from the end if {\it start\/} is negative
(where the first character from the end is the last character).

\item[\hbox{\tt swap\$}\hfill]
Swaps the top two literals on the stack.

\item[\hbox{\tt text.length\$}\hfill]
Pops the top (string) literal,
and pushes the number of text characters it contains, where an
accented character (more precisely, a ``special character''$\!$,
defined in Section~4)
counts as a single text character, even if it's missing
its matching right brace, and where braces don't count as
text characters.

\item[\hbox{\tt text.prefix\$}\hfill]
Pops the top two literals
(the integer literal {\it len\/} and a string literal, in that order).
It pushes the substring of the (at most) {\it len\/} consecutive text
characters starting from the beginning of the string.  This function
is similar to \hbox{\tt substring\$}, but this one considers
a ``special character''$\!$, even if
it's missing its matching right brace, to be a single text character
(rather than however many ASCII characters it actually comprises),
and this function doesn't consider braces to be text characters;
furthermore, this function appends any needed matching right braces.

\item[\hbox{\tt top\$}\hfill]
Pops and prints the top of the stack on the terminal and log file.
It's useful for debugging.

\item[\hbox{\tt type\$}\hfill]
Pushes the current entry's type (book, article, etc.),
but pushes the null string
if the type is either unknown or undefined.

\item[\hbox{\tt warning\$}\hfill]
Pops the top (string) literal
and prints it following a warning message.
This also increments a count of the number of warning messages issued.

\item[\hbox{\tt while\$}\hfill]
Pops the top two (function) literals,
and keeps executing the second as long as the (integer)
literal left on the stack by executing the first is greater than 0.

\item[\hbox{\tt width\$}\hfill]
Pops the top (string) literal
and pushes the integer that represents its width in some relative units
(currently, hundredths of a point, as specified by the June 1987 version
of the $cmr10$ font; the only white-space character with nonzero width
is the space).
This function takes the literal literally;
that is, it assumes each character in the string is to be printed as
is, regardless of whether the character has a special meaning to \TeX,
except that ``special characters'' (even without their right braces) are
handled specially.
This is meant to be used for comparing widths of label strings.

\item[\hbox{\tt write\$}\hfill]
Pops the top (string) literal
and writes it on the output buffer (which will result in
stuff being written onto the {\tt bbl} file when the buffer fills up).

\end{description}

Note that the built-in functions \hbox{\tt while\$} and \hbox{\tt if\$}
require two function literals on the stack.
You get them there either by immediately preceding the name of a function
by a single quote, or, if you don't feel like defining a new function with
the \hbox{\tt FUNCTION} command,
by simply giving its definition (that is, giving what would be the second
argument to the \hbox{\tt FUNCTION} command, including the surrounding braces).
For example the following function fragment appends the character `{\tt a}'
if the string variable named \hbox{\tt label} is nonnull:
\begin{verbatim}
    .  .  .
    label "" =
      'skip$
      { label "a" * 'label := }
    if$
    .  .  .
\end{verbatim}
A function whose name you quote needn't be built in
like \hbox{\tt skip\$} above---it may, for example,
be a field name or a function you've defined earlier.


\subsection{Name formatting}

What's in a name?
Section~4 pretty much describes this.
Each name consists of four parts: First, von, Last, and Jr;
each consists of a list of name-tokens,
and any list but Last's may be empty for a nonnull name.
This subsection describes the format string you must supply to
the built-in function \hbox{\tt format.name\$}.

Let's look at an example of a very long name.
Suppose a database entry~\cite{prime-number-theorem} has the field
\begin{verbatim}
  author = "Charles Louis Xavier Joseph de la Vall{\'e}e Poussin"
\end{verbatim}
and suppose you want this formatted ``last name comma initials''$\!$.
If you use the format string
\begin{verbatim}
    "{vv~}{ll}{, jj}{, f}?"
\end{verbatim}
\BibTeX\ will produce
\begin{verbatim}
    de~la Vall{\'e}e~Poussin, C.~L. X.~J?
\end{verbatim}
as the formatted string.

Let's look at this example in detail.
There are four brace-level~1 {\em pieces\/} to this format string,
one for each part of a name.
If the corresponding part of a name isn't present (the Jr part for this name),
everything in that piece is ignored.
Anything at brace-level~0 is output verbatim
(the presumed typo `{\tt ?}' for this name is at brace-level~0),
but you probably won't use this feature much.

Within each piece a double letter tells \BibTeX\ to use whole tokens, and
a single letter, to abbreviate them (these letters must be at brace-level~1);
everything else within the piece is used verbatim
(well, almost everything---read on).
The tie at the end of the von part (in \hbox{\verb|{vv~}|})
is a discretionary tie---\BibTeX\ will output a tie at that point
if it thinks there's a need for one;
otherwise it will output a space.
If you really, really, want a tie there,
regardless of what \BibTeX\ thinks, use two of them
(only one will be output); that is, use \hbox{\verb|{vv~~}|}.
A tie is discretionary only if it's the last character of the piece;
anywhere else it's treated as an ordinary character.

\BibTeX\ puts default strings {\em between\/} tokens of a name part:
For whole tokens it uses either a space or a tie,
depending on which one it thinks is best,
and for abbreviated tokens it uses a period followed by
either a space or a tie.
However it doesn't use this default string after the last token in a list;
hence there's no period following the `J' for our example.
You should have used
\begin{verbatim}
    "{vv~}{ll}{, jj}{, f.}"
\end{verbatim}
to get \BibTeX\ to produce the same formatted string but with the question
mark replaced by a period.
Note that the period should go inside the First-name piece,
rather than where the question mark was, in case a name has no First part.

If you want to override \BibTeX's default between-token strings, you
must explicitly specify a string.
For example suppose you want a label to contain the first letter from each
token in the von and Last parts, with no spaces;
you should use the format string
\begin{verbatim}
    "{v{}}{l{}}"
\end{verbatim}
so that \BibTeX\ will produce `{\tt dlVP}' as the formatted string.
You must give a string for each piece whose default you want overridden
(the example here uses the null string for both pieces), and this string
must immediately follow either the single or double letter for the piece.
You may not have any other letters at brace-level~1 in the format string.

\bibliography{btxdoc}
\bibliographystyle{plain}
\end{document}