% \iffalse meta-comment
% Copyright (C) 2015 by Edgar Olthof
%
% This file may be distributed and/or modified under the
% conditions of the LaTeX Project Public License, either
% version 1.3 of this license or (at your option) any later
% version.  The latest version of this license is in
%
%    http://www.latex-project.org/lppl.txt
%
% and version 1.3 or later is part of all distributions of
% LaTeX version 1999/12/01 or later.
%
% \fi
% \iffalse
%<package>\NeedsTeXFormat{LaTeX2e}
%<package>\ProvidesPackage{fcolumn}
%<package>          [2015/09/20 v1.1.1 Automatic Currency Package (EO)]
%<package>\RequirePackage{array}
%<*driver>
\documentclass{ltxdoc}
\usepackage{fcolumn}
\usepackage{booktabs}
\usepackage{paralist}
\EnableCrossrefs
\CodelineIndex
\RecordChanges
\begin{document}
   \DocInput{fcolumn.dtx}
\end{document}
%</driver>
% \fi
% \CheckSum{618}
% \CharacterTable
%  {Upper-case    \A\B\C\D\E\F\G\H\I\J\K\L\M\N\O\P\Q\R\S\T\U\V\W\X\Y\Z
%   Lower-case    \a\b\c\d\e\f\g\h\i\j\k\l\m\n\o\p\q\r\s\t\u\v\w\x\y\z
%   Digits        \0\1\2\3\4\5\6\7\8\9
%   Exclamation   \!     Double quote  \"     Hash (number) \#
%   Dollar        \$     Percent       \%     Ampersand     \&
%   Acute accent  \'     Left paren    \(     Right paren   \)
%   Asterisk      \*     Plus          \+     Comma         \,
%   Minus         \-     Point         \.     Solidus       \/
%   Colon         \:     Semicolon     \;     Less than     \<
%   Equals        \=     Greater than  \>     Question mark \?
%   Commercial at \@     Left bracket  \[     Backslash     \\
%   Right bracket \]     Circumflex    \^     Underscore    \_
%   Grave accent  \`     Left brace    \{     Vertical bar  \|
%   Right brace   \}     Tilde         \~}
%
% \changes{v0.1}{2015/02/06}{First working version}
% \changes{v1.0}{2015/03/07}{Three-argument version is working properly}
% \changes{v1.1}{2015/09/14}{Automatic checking of column balance performed when number of
%   F-columns is even (behaviour can be overridden).  Empty entries are now recognised and 
%   correctly treated as such, except for the one ended by the double backslash. Not serious;
%   workaround possible. Furthermore optimisation of code: minimised the number of private 
%   counts and resetting of column counter done in a nicer way.}
% \changes{v1.1.1}{2015/09/27}{Installation procedure changed from .ins-in-.dtx to separate
% .ins and .dtx after discussion with Karl Berry as well as some minor code improvements.}
% \GetFileInfo{fcolumn.sty}
% \DoNotIndex{\@arstrut,\@arstrutbox,\@depth,\@firstampfalse,\@firstamptrue,\@gobble,\@halignto}
% \DoNotIndex{\@height,\@mkpream,\@ne,\@nextchar,\@preamble,\@sharp,\@tempdima,\@tempswafalse}
% \DoNotIndex{\@tempswatrue,\@\temptokena,\@tfor,\@whilesw,\@width,\advance,\arraystretch,\baselineskip}
% \DoNotIndex{\begingroup,\bgroup,\count,\count@,\cr,\csname,\def,\divide,\do,\dp,\edef,\egroup}
% \DoNotIndex{\else,\endcsname,\endgroup,\expandafter,\extrarowheight,\fi,\gdef,\halignto,\hbox}
% \DoNotIndex{\hfil,\ht,\if,\if@firstamp,\if@tempswa,\ifcase,\lineskip,\loop,\m@ne,\m@th,\message}
% \DoNotIndex{\multiply,\or,\par,\prepnext@tok,\protect,\repeat,\setbox,\space,\string,\strutbox}
% \DoNotIndex{\tabskip,\tabularnewline,\the,\the@toks,\thr@@,\tw@,\unhbox,\vbox,\vcenter,\vrule}
% \DoNotIndex{\vskip,\vtop,\wd,\xdef,\z@,\z@skip}
% \title{The \textsf{fcolumn} package\thanks{This file has version
% number \fileversion, last revised \filedate.}}
% \author{Edgar Olthof\\
% \texttt{edgar <dot> olthof <at> inter <dot> nl <dot> net}}
% \date{Printed \today}
% \maketitle
% \begin{abstract}
% In financial reports, text and currency amounts are regularly put in
% one table, e.g., a year balance or a profit-and-loss overview.
% This package provides the settings for automatically typesetting and checking
% such columns, including the sum line (preceded by a rule of the correct
% width), using the specifier |f|.
% \end{abstract}
%
% \section{Introduction}\label{intro}
% The package |fcolumn| provides the macros for an extra tabular specifier that makes creating 
% financial tables easy.  The column specifier |f| itself is rather simple.  It is the
% predefined version of a generic column |F|.  The generic version expects three arguments:
% |#1| is the group separator, |#2| is the decimal mark, and |#3| the coding used for
% grouping digits of the integer part and decimal part.
%
% The f-column in the current version of the package is defined for the continental European
% standard: |\newcolumntype{f}{F{.}{,}{3,2}}|.  This means that a number like 1234567 will be 
% typeset as $12{.}345{,}67$.  People in the Anglo-saxon world would rather
% code |\newcolumntype{f}{F{,}{.}{3,2}}|, yielding $12{,}345{.}67$ for the number given above.
% The default value for |#3| is 3,2, indicating that grouping of the integer part
% is by three digits and that the decimal part consists of two digits. 
% If however, in your country or company grouping is done with a thinspace every 
% four digits and there are three digits after the decimal mark---that happens to 
% be a |\cdot|---, then simply specify |\newcolumntype{f}{F{\,}{\cdot}{4,3}}| in 
% that case.  
%
% By default two digits are used for the decimal part, so if you really
% want no decimal digits (in that case of course also skipping the decimal mark) 
% you have to explicitly specify |x,0|.
%
% This package requires and loads the |array| package~\cite{array}.
% To show where and how the F-column is used, let's look at some typical financial information 
% as shown in Table~\ref{tab:ex1}
% \begin{table}[htb]
% \caption{Example Table}
% \label{tab:ex1}
% \begin{tabular}{@{}lflf@{}}
% \multicolumn4c{\bfseries Balance sheet}\\
% \toprule
% properties&\multicolumn1r{31 dec 2014}&
% debts&\multicolumn1{l@{}}{31 dec 2014}\\
% \midrule
% house &     20000000 & equity capital&  5000000\\
% bank account& -60323 & mortgage &      15000000\\
% savings&     2800000 \\
% cash&          14585 & profit  &        2754262\\
% \sumline
% \bottomrule
% \end{tabular}
% \end{table}
% and how this is entered in \LaTeX\ (Table~\ref{tab:ex2}).
% \begin{table}[htb]
% \caption{Verbatim version of Example Table}
% \label{tab:ex2}
% \begin{verbatim}
% \begin{table}[htb]
% \caption{Example Table}
% \label{tab:ex1}
% \begin{tabular}{@{}lflf@{}}
% \multicolumn4c{\bfseries Balance sheet}\\
% \toprule
% properties&\multicolumn1r{31 dec 2014}&
% debts&\multicolumn1{l@{}}{31 dec 2014}\\
% \midrule
% house &     20000000 & equity capital&  5000000\\
% bank account& -60323 & mortgage &      15000000\\
% savings&     2800000 \\
% cash&          14585 & profit  &        2754262\\
% \sumline
% \bottomrule
% \end{tabular}
% \end{table}
% \end{verbatim}
% \end{table}
% All the work was done by the column specifier ``f'' (for ``finance'').  In this case
% it constructs the |\sumline|, typesets the numbers, calculates the totals, determines
% the widths of the sumrules, and checks whether the two columns are in balance; if not,
% the user is warned via a |\message|.  Of course for nice settings the
% |booktabs| package~\cite{booktabs} was used, but that is not the point here.
%
% This package is heavily inspired by the |dcolumn| package by David Carlisle~\cite{dcolumn},
% some constructions are more or less copied from that package.
%
% \section{Commands}
% The user only needs to know six commands or constructions.  These six are given here.
% \begin{macro}{F}
% In the tabular the column specifier |F| can be given with arguments, or the predefined
% version |f|, where the three arguments of |F| are |{.}|, |{,}|, and |3,2|.  If you
% want |g| to be your own definition like the curious one given in Section~\ref{intro}, then
% specify |\newcolumntype{g}{F{\,}{\cdot}{4,3}}| prior to using |g| in a tabular.
% 
% Entries in an F-column are from that moment on, treated as (integer) numbers unless
% explicitly escaped by |\leeg|, see below.  The numbers are typeset according to the template
% the user gives with his/her |F|-column.
% \end{macro}
% \begin{macro}{\sumline}
% The numbers in an |F|-column are typeset as a financial amount, but the real benefit comes
% with the |\sumline|. It does three things:
% \begin{compactitem}
% \item It calculates the total of the column so far and the maximum width encountered 
% so far, including the width of the total;
% \item It generates a rule with width calculated in the first item;
% \item It checks the columns that are supposed to balance whether or not they actually do.
% If so, nothing happens.  If not, a |\message| is given that column $i$ and $j$ do not balance,
% where $i$ and $j$ are the relevant columns.  This is only done if the total number of 
% |F|-columns is even, e.g., if there are six |F|-columns, then~1 is checked against~4, 2 against~5, 
% and 3 against~6.  If the number of |F|-columns is odd then anything could be possible in that table
% and nothing is assumed about structure within the table.  This behaviour can be overridden, see below.
% \end{compactitem}
% \end{macro}
% \begin{macro}{\resetsumline}
% Suppose you want to typeset one tabular with the profit-and-loss of many projects individually.
% The layout of those tabulars is the same and it were nice if all columns were aligned.  This can 
% be done by making it one big tabular with a fresh start for each project.  The macro |\resetsumlines|
% is used for that: it resets all totals and all column widths, see for example Table~\ref{tab:ex3}.
% \begin{table}[htb]
% \caption{Example: multiple projects}
% \label{tab:ex3}
% \begin{tabular}{@{}lfflff@{}}
% \multicolumn6c{\bfseries Project~1}\\
% \toprule
% expense&\multicolumn1r{actual}&\multicolumn1r{budget}&
% income&\multicolumn1r{actual}&\multicolumn1{r@{}}{budget}\\
% \midrule
% food &         45000 & 50000 & tickets&  100000 &  100000\\
% drinks &       35000 & 40000 \\
% music &        18000 & 10000 \\
% profit &        2000 \\
% \sumline
% \resetsumline
% \multicolumn6c{\bfseries Project~2}\\
% \toprule
% expense&\multicolumn1r{actual}&\multicolumn1r{budget}&
% income&\multicolumn1{r}{actual}&\multicolumn1{r@{}}{budget}\\
% \midrule
% food &         25000 & 30000 & tickets&  40000 &  45000\\
% drinks &       10000 &  8000 \\
% music &         8000 &  7000 & loss &     3000 \\
% \sumline
% \bottomrule
% \end{tabular}
% \end{table}
% Note that the rules in the first and third F-columns of project~1 cover $1{.}000{,}00$ whereas
% in project~2 those rules are shorter since they only cover $430{,}00$; still the columns are 
% aligned.  The verbatim way of setting up Table~\ref{tab:ex3} is given in Table~\ref{tab:ex4}.
% \begin{table}[htb]
% \caption{Verbatim version of Table~\ref{tab:ex3}}
% \label{tab:ex4}
% \begin{verbatim}
% \begin{table}[htb]
% \caption{Example: multiple projects}
% \label{tab:ex3}
% \begin{tabular}{@{}lfflff@{}}
% \multicolumn6c{\bfseries Project~1}\\
% \toprule
% expense&\multicolumn1r{actual}&\multicolumn1r{budget}&
% income&\multicolumn1r{actual}&\multicolumn1{r@{}}{budget}\\
% \midrule
% food &         45000 & 50000 & tickets&  100000 &  100000\\
% drinks &       35000 & 40000 \\
% music &        18000 & 10000 \\
% profit &        2000 \\
% \sumline
% \resetsumline
% \multicolumn6c{\bfseries Project~2}\\
% \toprule
% expense&\multicolumn1r{actual}&\multicolumn1r{budget}&
% income&\multicolumn1r{actual}&\multicolumn1{r@{}}{budget}\\
% \midrule
% food &         25000 & 30000 & tickets&  40000 &  45000\\
% drinks &       10000 &  8000 \\
% music &         8000 &  7000 & loss &     3000 \\
% \sumline
% \bottomrule
% \end{tabular}
% \end{table}
% \end{verbatim}
% \end{table}
% \end{macro}
% \begin{macro}{\leeg}
% If an |F|-column should be empty then simply leave it empty.  If however it should not be empty
% but the entry should be treated as text---even it is a number---this can be done with |\leeg|.
% It expects an argument and this argument is typeset in the column, affecting the maximum column
% width so far.  The common case is where |p.m.| (\textit{pro memoria}) is entered. An empty |F|-column
% followed by |\\| does not work: either remove the |&| or specify |\leeg{}|.
% \end{macro}
% \begin{macro}{\checkfcolumns}
% The automatic column balance check can also be done manually.  If |F|-columns~1 and~4 should balance
% and you want them to be checked, then simply say |\checkfcolumns14|.  With more than nine |F|-columns
% you may be forced to say something like |\checkfcolumns{10}{12}|.
% \end{macro}
% \begin{macro}{\strictaccounting}
% In the rare occasion that a negative number occurs in a financial table, the sign of that number
% can be an explicit minus sign ($-$) or the number is coloured red, or it is typeset between parentheses,
% and there may be even other ways.  By default (for aesthetic reasons) |fcolumn| typesets it with 
% a minus~sign, but strict accounting prescibes that the number should be put between parentheses.
% The latter can be accomplished by setting |\strictaccountingtrue|. 
% \end{macro}
% \StopEventually{\PrintChanges\PrintIndex}
% \section{The macros}
% \begin{macro}{column F}
% \begin{macro}{column f}
% The column specifier |F| is the generic one, and |f| is the default (European) one
% for easy use. Note that the definition of the column type |f| does not use private macros 
% (no |@|), so overriding its definition is easy for a user.
%    \begin{macrocode}
\newcolumntype{F}[3]{>{\b@fi{#1}{#2}{#3}}r<{\e@fi}}
\newcolumntype{f}{F{.}{,}{3,2}}
%    \end{macrocode}
% \end{macro}
% \end{macro}
% \begin{macro}{\FCsc@l}
% \begin{macro}{\FCtc@l}
% Two \meta{count}s are defined, that both start at zero: the \meta{count} 
% |\FCsc@l|, that keeps track at which F-column the tabular is working on and 
% the \meta{count} |\FCtc@l|, that records the number of F-columns that were 
% encountered so far.  Later in the package the code can be found for generating
% a new \meta{count} and a new \meta{dimen} if the number of requested F-columns
% is larger than currently available.  This is of course the case when an F-column
% is used for the first time.
%    \begin{macrocode}
\newcount\FCsc@l \FCsc@l=0
\newcount\FCtc@l \FCtc@l=0
%    \end{macrocode}
% \end{macro}
% \end{macro}
% \begin{macro}{\geldm@cro}
% The macro |\geldm@cro| takes a number and by default interprets this as
% an amount expressed in cents (dollar cents, euro cents, centen, 
% Pfennige, kopecks, groszy) and typesets it as the amount in entire 
% currency units (dollars, euros, guldens, Marke, ruble, z\l oty)
% with comma as decimal separator and the dot as thousand separator.
% As explained, this can be changed. It uses a private boolean |\withs@p|
% and an accessable, i.e., without |@| boolean: |\strictaccounting|.  The latter  
% is used to typeset negative numbers between parentheses.  By default it doesn't
% do this: a minus sign is used.
%    \begin{macrocode}
\newif\ifwiths@p
\newif\ifstrictaccounting \strictaccountingfalse
%    \end{macrocode}
% Actually |\geldm@cro| is only a wrapper around |\g@ldm@cro|.
%    \begin{macrocode}
\def\geldm@cro#1#2{\withs@pfalse 
\afterassignment\g@ldm@cro\count@#2\relax{#1}}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\g@ldm@cro}
% This macro starts by looking at the sign of |#2|: if it is negative, it prints the correct
% indicator (a parenthesis or a minus sign), assigns the absolute value of |#2| to |\count2| 
% and goes on.  Note that |\geldm@cro| and therefore |\g@ldm@cro| are always used within |$|s, so 
% it is really a minus sign that is printed, not a hyphen.  All calculations are done with
% |\count0|, |\count1|, etc.\ i.e., without fcolumn-specific \meta{count}s because it is all
% done locally. Leaving the tabular environment will restore their values.
%    \begin{macrocode}
\def\g@ldm@cro#1\relax#2{%
\ifnum#2<0 \ifstrictaccounting(\else-\fi\count2=-#2 \else\count2=#2 \fi
%    \end{macrocode}
% Calculate the entire currency units: this is the result of $x/a$ as integer
% division, with $a=10^n$ and $n$ the part of |#1| after the separator (if any).
% Here the first character of |#1| is discarded, so the separator in |#1| is not strict:
% you could also specify |3.2| instead of |3,2| (or even |3p2|).
%   \begin{macrocode}
\count3=\ifx\relax#1\relax2 \else \@gobble#1\relax\fi
\count4=\count3
\loop
 \ifnum\count3>0 \divide\count2 by 10 \advance\count3 by \m@ne
\repeat
%    \end{macrocode}
% The value in |\count2| is then output by |\g@ldens| using the separation given.
%    \begin{macrocode}
\g@ldens{\the\count@}%
%    \end{macrocode}
% If there is a decimal part\dots
%    \begin{macrocode}
\ifnum\count4>0\decim@lmark
%    \end{macrocode}
% Next the decimal part is dealt with. Now $x\bmod a$ is calculated in the 
% usual way: $x-(x/a)*a$ with integer division.  The minus sign necessary 
% for this calculation is introduced in the next line by changing the 
% comparison from |<| to |>|.
%    \begin{macrocode}
 \ifnum#2>0 \count2=-#2\else\count2=#2 \fi
 \count3=\count4
 \loop
  \ifnum\count3>0 \divide\count2 by 10 \advance\count3 by \m@ne
 \repeat
%    \end{macrocode}
% The value of |\count3| is~0 now. Now counting up again, this saved an assignment.
%    \begin{macrocode}
 \loop
  \ifnum\count3<\count4 \multiply\count2 by 10 \advance\count3 by \@ne
 \repeat
 \ifnum#2>0 \advance\count2 by #2
 \else \advance\count2 by -#2
 \fi
 \zerop@d{\number\count3}{\number\count2}%
\fi
%    \end{macrocode}
% If the negative number is indicated by putting it between parentheses, then the
% closing parenthesis should stick out of the column, otherwise the alignment of 
% this entry in the column is wrong.  This is done by an |\rlap| and therefore does
% not influence the column width.  For the last column this means that this parenthesis
% may even stick out of the table.  I don't like this, therefore I chose to put 
% |\strictaccountingfalse|.  Change if you like.
%    \begin{macrocode}
\ifnum#2<0 \ifstrictaccounting\rlap{)}\fi\fi}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\g@ldens}
% Here the whole currency units are dealt with.  The macro |\g@ldens| is used
% recursively, therefore the double braces; this allows to use |\count0| locally.
% Tail recursion is not possible here (at least I don't know of a way), but that 
% is not very important, as the largest number (which is $2^{31}-1$) will only cause a 
% threefold recursion.  This also implies that the largest amount this package
% can deal with is 2.147.483.647 (using |x.0|). For most people this is probably
% more than enough if the currency is euros or dollars.  If not, then it is 
% more likely that you are spending your time on the beach than studying this package.
% And otherwise make clear that you use a currency unit of k\$.
%    \begin{macrocode}
\def\g@ldens#1{{\count3=\count2 \count0=#1
%    \end{macrocode}
% First divide by $10^n$, where $n$ is |#1|.
%    \begin{macrocode}
\loop
 \ifnum\count0>0
 \divide\count2 by 10
 \advance\count0 by \m@ne
\repeat
%    \end{macrocode}
% Here is the recursive part,
%    \begin{macrocode}
\ifnum\count2>0 \g@ldens{#1}\fi
%    \end{macrocode}
% and then reconstruct the rest of the number.
%    \begin{macrocode}
\count0=#1
\loop
 \ifnum\count0>0
 \multiply\count2 by 10
 \advance\count0 by\m@ne
\repeat
\count2=-\count2
\advance\count2 by \count3 \du@zendprint{#1}}}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\du@zendprint}
% The macro |\du@zendprint| takes care for correctly printing the separator
% and possible trailing zeros.
%    \begin{macrocode}
\def\du@zendprint#1{\ifwiths@p\sep@rator\zerop@d{#1}{\number\count2}%
 \else\zerop@d{1}{\number\count2}\fi \global\withs@ptrue}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\zerop@d}
% The macro |\zerop@d| uses at least |#1| digits for printing the 
% number |#2|, padding with zeros when necessary.
% Initially this was a nice macro using tail recursion until it was
% found that the running time of that macro was proportional to $n^2$,
% where $n$ is roughly the number of zeros to be padded. The worst
% case situation is when printing ``0''.  The running time of the current version
% is only linear in $n$.
%
% It is done within an extra pair of braces, so that |\count0| 
% and |\count1| can be used without disturbing their values in other macros.
%    \begin{macrocode}
\def\zerop@d#1#2{{\count0=1 \count1=#2
%    \end{macrocode}
% First determine the number of digits of |#2| (expressed in the decimal
% system).  This number is in |\count0| and is at least~1.
%    \begin{macrocode}
\loop
 \divide \count1 by 10
 \ifnum\count1>0 \advance\count0 by\@ne
\repeat
%    \end{macrocode}
% The number of zeros to be padded is $\max(0,|#1-\count0|)$ (the second
% argument can be negative), so a simple loop suffices.
%    \begin{macrocode}
\loop
 \ifnum\count0<#1\relax 0\advance\count0 by\@ne
\repeat 
\number#2}}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\zetg@ld}
% This macro takes care for several things: it increases the total for a given
% F-column, it records the largest width of the entries in that column and it
% typesets |#1| via |\geldm@cro|.
%    \begin{macrocode}
\def\zetg@ld#1#2{\global\advance\csname 
FCtot@\romannumeral\FCsc@l\endcsname by #1
\setbox0=\hbox{$\geldm@cro{#1}{#2}$}%
\ifdim\wd0>\csname FCwd@\romannumeral\FCsc@l\endcsname
 \global\csname FCwd@\romannumeral\FCsc@l\endcsname=\wd0
\fi\unhbox0}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\b@fi}
% The macro |\b@fi| provides the beginning of the financial column.
% It will be inserted in the column to capture the number
% entered by the user. The |\let| is only local to the column and is necessary because
% |\ignorespaces| is automatically inserted by the preamble-generating macro |\@mkpream|.
% Its effect should be annihilated, otherwise the assignment to |\bedr@g| goes wrong.
% The plain \LaTeX\ command |\@empty| is used for that.  The separator and decimal mark
% are within a math environment, so you can indeed specify |\,| instead of |\thinspace|,
% but there is an extra brace around, so it doesn't affect
% the spacing between the digits (trick copied from |dcolumn|, Ref.~\cite{dcolumn}).
%    \begin{macrocode}
\newcount\bedr@g
\def\b@fi#1#2#3{\def\sep@rator{{#1}}\def\decim@lmark{{#2}}%
\def\sp@l{#3}\let\ignorespaces=\@empty \let\unskip=\@empty
\global\advance\FCsc@l by \@ne
%    \end{macrocode}
% The value specified by the user is then captured by |\bedr@g| and this is done in a
% special way: |\bedr@g| is assigned globally within |\box0|.  Why? To later check
% the width of this box, see |\e@fi|.
%    \begin{macrocode}
\setbox0=\hbox\bgroup\global\bedr@g=}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\e@fi}
% If the user enters a number, say 10, \LaTeX\ will read
% |\setbox| |0| |=| |\hbox| |\bgroup| |\global| |\bedr@g| |=| |10| |123| |\relax| |\egroup|, which
% is perfectly valid.  The count |\bedr@g| is~10 now and |\box0| is non-empty, in fact containing
% the number 123 (although any number would be good: it is discarded anyway).  If however nothing 
% was entered by the user, \LaTeX\ will read |\setbox| |0| |=| |\hbox| |\bgroup| |\global| |\bedr@g|
% |=| |123| |\relax| |\egroup|, which is also perfectly valid, but now |\bedr@g| is~123 and |\box0| 
% is empty.  So the width of |\box0| can be used to discriminate without actually using its contents
% in the horizontal list.  Later, with the code for |\leeg| a related trick will be used.  The space
% at the beginning of |\e@fi| is important, otherwise the 123 might be concatenated with the user 
% entry, leading to wrong numbers.
%    \begin{macrocode}
\def\e@fi{ 123\relax\egroup\ifdim\wd0>\z@
 \zetg@ld{\number\bedr@g}{\sp@l}%
\fi}
%    \end{macrocode}
% Please note that this trick does not work for an empty F-column that is ended by |\\|.  The reason 
% is that the first thing |\\| does---even before providing the |\cr| ---is to issue a |\relax|.  
% This stops the assignment to |\bedr@g| and causes a \TeX\ error, complaining about a missing number.
% This is then repaired by \TeX\ by inserting a zero, so it doesn't affect the total of that column, 
% but in particular this zero is then typeset, which it is not supposed to.  I decided to leave it 
% this way because in practice this is never a problem: simply discard the last |&| and the column
% before the offending column will now be ended by |\\|.  Iterate for this line until a filled F-column
% or a non-F-column is encountered.  This is assuming that you did not specify \verb+|+s in the tabular
% description: for them to work for the whole table you should have all |&|s in place (this is true in 
% general, not only for |fcolumn|).  On the other hand you should never, ever use vertical rules in 
% tables, according to Ref.~\cite{booktabs}.  And if you really must, then fill the empty F-columns 
% with |\leeg{}|.
% \end{macro}
%
% Here are adaptations to existing macros.
% \begin{macro}{\@array}
% The definition of |\@array| had to be extended slightly because
% it should also include |\@mksumline| (acting on the same |#2|
% as |\@mkpream| gets).  This change is transparant: it only adds 
% functionality and if you don't use that, you won't notice the difference.
% It starts by just copying the original definition from the |array| package~\cite{array}.
%    \begin{macrocode}
\def\@array[#1]#2{%
\@tempdima \ht \strutbox
\advance \@tempdima by\extrarowheight
\setbox \@arstrutbox \hbox{\vrule
           \@height \arraystretch \@tempdima
           \@depth \arraystretch \dp \strutbox
           \@width \z@}%
%    \end{macrocode}
% Here comes the first change: after each |\\| (or |\cr| for that matter) the \meta{count}
% |\FCsc@l| should be reset.  This is easiest done with |\everycr|, but |\everycr| is
% put to |{}| by |\ialign|, so that definition should change.  The resetting should 
% be done globally.
%    \begin{macrocode}
\def\ialign{\everycr{\noalign{\global\FCsc@l=0 }}%
 \tabskip\z@skip\halign}
%    \end{macrocode}
% Then the definition is picked up again.
%    \begin{macrocode}
\begingroup
\@mkpream{#2}%
\xdef\@preamble{\noexpand \ialign \@halignto
\bgroup\@arstrut\@preamble\tabskip\z@\cr}%
\endgroup 
%    \end{macrocode}
% The combination |\endgroup| followed by |\begingroup| seems redundant, but that is
% not the case: the |\endgroup| restores everything that was not |\global|.  With the
% following |\begingroup| it is ensured that |\@mksumline| experiences the same
% settings as |\@mkpream| did.
%    \begin{macrocode}
\begingroup
\@mksumline{#2}%
\endgroup
%    \end{macrocode}
% As a side product of |\@mksumline| also the \meta{count}s for the totals and 
% \meta{dimen}s for the widths of the colums are created.  The columns should start
% fresh, i.e., totals are~0 and widths are 0\thinspace pt.
%    \begin{macrocode}
\res@tsumline
%    \end{macrocode}
% From here on it is just the old definition of |array.sty|.
%    \begin{macrocode}
\@arrayleft
\if #1t\vtop \else \if#1b\vbox \else \vcenter \fi \fi
\bgroup
\let \@sharp ##\let \protect \relax
\lineskip \z@
\baselineskip \z@
\m@th
\let\\\@arraycr \let\tabularnewline\\\let\par\@empty \@preamble} 
%    \end{macrocode}
% Because |\@array| was changed here and it is this version that should be used,
% |\@@array| should be |\let| equal to |\@array| again.
%    \begin{macrocode}
\let\@@array=\@array
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\@mksumline}
% The construction of the sumline is much easier than that of the preamble for
% several reasons.  It may be safely assumed that the preamble specifier is
% grammatically correct because it has already been screened by |\@mkpream|.
% Furthermore most entries will simply add nothing to |\sumline|, e.g., |@|, |!|,
% and \verb+|+ can be fully ignored. Ampersands are only inserted by |c|, |l|, |r|,  
% |p|, |m|, and |b|.  So a specifier like |@{}lflf@{}| will yield the sumline
% |&\a&&\a\\|, (where |\a| is a macro that prints the desired result of the column, 
% see later).  Had the specifier been \verb+l|f||@{   }l|f+, then the same sumline must
% be constructed: all difficulties are already picked up and solved in the creation
% of the preamble.
%
% In reality the sumline must be constructed from the expanded form of the specifier,
% so |@{}lf@{}| will expand as |@{}l>{\b@fi{.}{,}{3,2}}r<{\e@fi}@{}|. The
% rules for constructing the sumline are now very simple:
% \begin{compactitem}
% \item add an ampersand when |c|, |l|, |r|, |p|, |m|, or |b| is found, unless it is
% the first one (this is the same as in the preamble);
% \item add a |\a| when |<{\e@fi}| is found;
% \item ignore everything else;
% \item close with a |\\|.
% \end{compactitem}
% (In reality also the column check is inserted just before the |\\|, see |\aut@check|.)
% To discriminate, a special version of |\testpach| could be written, but that is 
% not necessary: |\testpach| can do all the work, although much of it will be discarded.
% Here speed is sacrificed for space and this can be afforded because the creation
% of the sumline is only done once per |\tabular|.
% 
% The start is copied from |\@mkpream|.
%    \begin{macrocode}
\def\@mksumline#1{\gdef\sumline{}\@lastchclass 4 \@firstamptrue
%    \end{macrocode}
% At first the column number is reset and the actual code for what was called |\a| 
% above is made inactive.
%    \begin{macrocode}
\global\FCsc@l=0
\let\prr@sult=\relax
%    \end{macrocode}
% Then |\@mkpream| is picked up again.
%    \begin{macrocode}
\@temptokena{#1}
\@tempswatrue
\@whilesw\if@tempswa\fi{\@tempswafalse\the\NC@list}%
\count0\m@ne
\let\the@toks\relax
\prepnext@tok
%    \end{macrocode}
% Next is the loop over all tokens in the expanded form of the specifier.  The change
% with respect to |\@mkpream| is that the body of the loop is now only dealing
% with F-classes~0, 2, and 10.  What to do in those cases is of course different from
% what to do when constructing the preamble, so special definitions are created, see
% below.
%    \begin{macrocode}
\expandafter \@tfor \expandafter \@nextchar
 \expandafter :\expandafter =\the\@temptokena \do
 {\@testpach
 \ifcase \@chclass \@classfz
  \or \or \@classfii \or
  \or \or \or \or \or \or \or \@classfx \fi
 \@lastchclass\@chclass}%
%    \end{macrocode}
% And the macro is finished by applying the |\aut@check| and appending the |\\| to the sumline.
% Note that the |\aut@check| is performed {\itshape in\/} the last column, but since it does not
% put anything in the horizontal list---it only writes to screen and transcript file---, this 
% is harmless.
%    \begin{macrocode}
\xdef\sumline{\sumline\noexpand\aut@check\noexpand\\}}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\@addtosumline}
% Macro |\@addtosumline|, as its name already suggests, adds something to the 
% sumline, like its counterpart |\@addtopreamble| did to the preamble.
%    \begin{macrocode}
\def\@addtosumline#1{\xdef\sumline{\sumline #1}}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\@classfx}
% Class~f10 for the sumline creation is a stripped down version of |\@classx|: add an ampersand
% unless it is the first.  It deals with the specifiers |b|, |m|, |p|, |c|, |l|, and |r|.
%    \begin{macrocode}
\def\@classfx{\if@firstamp \@firstampfalse \else \@addtosumline &\fi} 
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\@classfz}
% Class~f0 is applicable for specifiers |c|, |l|, and |r|, and if the arguments of |p|, |m|,
% or |b| are given.  The latter three cases, with |\@chnum| is 0, 1, or~2 should be ignored
% and the first three cases are now similar to class~f10.
%    \begin{macrocode}
\def\@classfz{\ifnum\@chnum<\thr@@ \@classfx\fi}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\@classfii}
% Here comes the nice and nasty part. Class~f2 is applicable if a~|<| is specified.  This is
% tested by checking |\@lastchclass|, which should be equal to~8.  Then it is checked that
% the argument to |<| is indeed |\e@fi|.  This check is rather clumsy but this was the
% first way, after many attempts, that worked.  It is necessary because the usage of |<|
% is not restricted to |\e@fi|: the user may have specified other \LaTeX-code using~|<|.
%    \begin{macrocode}
\def\@classfii{\ifnum\@lastchclass=8 
 \edef\t@stm{\expandafter\string\@nextchar}
 \edef\t@stn{\string\e@fi}
 \ifx\t@stm\t@stn
%    \end{macrocode}
% If both tests yield |true| then add the macro to typeset everything.
%    \begin{macrocode}
  \@addtosumline{\prr@sult}
%    \end{macrocode}
% But we're not done yet: in the following lines of code the appropriate \meta{count}s and
% \meta{dimen}s are created, if necessary.  Note that |\FCsc@l| was set to~0 in the beginning
% of |\@mksumline|, so it is well-defined when |\@classfii| is used.
%    \begin{macrocode}
  \global\advance\FCsc@l by \@ne 
  \ifnum\FCsc@l>\FCtc@l
%    \end{macrocode}
% Apparently the number of requested columns is larger than the currently available number of
% relevant \meta{count}s and \meta{dimen}s, so new ones should be created.  What is checked 
% here is merely the existence of |\FCtot@<some romannumeral>|.  If it already exists---although
% it may not even be a \meta{count}; that is not checked---it is not created by |fcolumn| and a 
% warning is given.  In case it is a \meta{count} you're just lucky, although any change to this
% \meta{count} is global anyway, so things will be overwritten.  In the case it is not a 
% \meta{count}, things will go haywire and you'll soon find out.  The remedy then is to rename 
% your \meta{count} prior to |fcolumn| to avoid this name clash.
%    \begin{macrocode}
   \expandafter\ifx\csname FCtot@\romannumeral\FCsc@l\endcsname\relax
    \expandafter\newcount\csname FCtot@\romannumeral\FCsc@l\endcsname
   \else
    \message{^^JWarning: FCtot@\romannumeral\FCsc@l \space is already 
    defined and it may not even be a <count>. I'll proceed,
    but with fingers crossed. }
   \fi
%    \end{macrocode}
% And the same is applicable for the \meta{dimen}: in case of a name clash you have to 
% rename your \meta{dimen} prior to |fcolumn|.
%    \begin{macrocode}
   \expandafter\ifx\csname FCwd@\romannumeral\FCsc@l\endcsname\relax
    \expandafter\newdimen\csname FCwd@\romannumeral\FCsc@l\endcsname
%    \end{macrocode}
% If the creation was successful, the \meta{count} |\FCtc@l| should be increased.
%    \begin{macrocode}
    \global\FCtc@l=\FCsc@l
   \else
    \message{^^JWarning: FCwd@\romannumeral\FCsc@l \space is already 
    defined and it may not even be a <dimen>. I'll proceed,
    but with fingers crossed. }
   \fi
  \fi
 \fi
\fi}
%    \end{macrocode}
% Once created it is not necessary to initialise them here because that is done later in one go.
% \end{macro}
% \begin{macro}{\leeg}
% This macro is used to overrule the default behaviour of the pair |\b@fi| and |\e@fi|.  It
% starts with ending the assignment to |\bedr@g| in the same way that |\e@fi| would normally
% do.  Then the effect of |\e@fi| (that is still in the preamble) is annihilated by |\let|ting 
% it to be |\relax|.  This |\let| is only local to the current column.  Then the argument to 
% |\leeg| is treated in a similar way as |\e@fi| would do with a typeset number.
%
% Since the user may from time to time also need a column entry other than a number in the 
% table, e.g., |\leeg{p.m.}|, this definition is without at-sign.
%    \begin{macrocode}
\def\leeg#1{ 234\relax\egroup \let\e@fi=\relax \setbox0=\hbox{#1}%
\ifdim\wd0>\csname FCwd@\romannumeral\FCsc@l\endcsname
 \global\csname FCwd@\romannumeral\FCsc@l\endcsname=\wd0
\fi\unhbox0}
%    \end{macrocode}
% Note that anything may be given as argument to |\leeg|, so in principle it can also be used
% to cheat: |\leeg{0,03}| will insert |0,03| in the table but it doesn't increase the totals
% of that column by~3 (assuming |3,2| coding for the separations).  But you won't cheat, won't you?
% It may affect the width, so be careful: don't insert the unabridged version of Romeo and Julia here.
% \end{macro}
% \begin{macro}{\prr@sult}
% The macro |\prr@sult| actually puts the information together.  It starts like |\leeg|.
%    \begin{macrocode}
\def\prr@sult{ 345\relax\egroup \let\e@fi=\relax
%    \end{macrocode}
% Then the information for the last line is computed.  It is not sufficient to calculate
% the width of the result (in points) to use that as the width of the rule separating 
% the individual entries and the result.  It may be that the sum is larger (in points) than
% any of the entries, e.g., when the result of $600+600$ is typeset.  The width of the rule
% should be equal to the width of |\hbox{$12{,}00$}| then (using specifier |3,2|).  On the other
% hand the width of the rule when summing $2400$ and $-2400$ should be that of |\hbox{$-24{,}00$}|
% (or |\hbox{$(24{,}00$}|, see above), not the width of the result |\hbox{$0{,}00$}|.  Therefore the 
% maximum of all entry widths, including the result, was calculated.
%    \begin{macrocode}
\setbox0=\hbox{$\geldm@cro{\number\csname
FCtot@\romannumeral\FCsc@l\endcsname}{\sp@l}$}%
\ifdim\wd0>\csname FCwd@\romannumeral\FCsc@l\endcsname
 \global\csname FCwd@\romannumeral\FCsc@l\endcsname=\wd0
\fi
\vbox{\hrule width \csname FCwd@\romannumeral\FCsc@l\endcsname
\vskip2pt
\hbox to \csname FCwd@\romannumeral\FCsc@l\endcsname{\hfil\unhbox0}}}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\aut@check}
% If the number of F-columns is even, it is assumed that they are part of two sets of columns of
% which each column of the first set should balance the appropriate column of the second set.  If
% on the other hand the number of columns is odd, then at least one column has nothing to balance
% against and no checking occurs.  It is correct to check for oddness of |\FCsc@l| since this
% |\aut@check| is only performed in the last column of the tabular: the value of |\FCsc@l| now 
% equals the number of columns used in the current tabular (and may differ from |\FCtc@l|).
%
% The output is only to screen and the transcript file; it doesn't change the appearance of your
% document, so in case the assumption is wrong you can safely ignore the result and go on.
% The \meta{count}s 0 and~1 are used here and this can be done because any content of those 
% \meta{count}s from previous calculations has become irrelevant at this moment.
% \begin{macrocode}
\def\aut@check{\ifodd\FCsc@l\else \count0=\@ne \count1=\FCsc@l 
 \divide\count1 by \tw@ \advance\count1 by \@ne
 \loop 
  \ifnum\csname FCtot@\romannumeral\count0\endcsname=
   \csname FCtot@\romannumeral\count1\endcsname\else
   \message{^^JWarning: F-columns \number\count0 \space
    and \number\count1 \space do not balance! }%
  \fi
  \ifnum\count1=\FCsc@l\else
   \advance\count0 by\@ne \advance\count1 by\@ne
 \repeat
\fi }
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\checkfcolumns}
% But the assumptions for |\aut@check| may be wrong, therefore manual
% control on this checking is also made possible here. The macro |\checkfcolumns| 
% provides a way to the user to check that the appropriate 
% columns are balanced (as it should in a balance).  Arguments |#1| and |#2| are
% the column numbers to compare.  It is the responsibility of the user to provide
% the correct numbers here, otherwise bogus output is generated. 
%    \begin{macrocode}
\def\checkfcolumns#1#2{\noalign{\ifnum\csname FCtot@\romannumeral#1
\endcsname=\csname FCtot@\romannumeral#2\endcsname\else
 \message{^^JWarning: F-columns #1 and #2 do not balance! }%
\fi}}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\res@tsumline}
% Since all changes to the totals and widths of the columns are global,
% they have to be reset actively at the start of a tabular or array.
% That is an action by itself, but it may occur more often, on request
% of the user, therefore a special macro is defined.  A side effect of
% this macro is that |\FCsc@l| is reset to~0. This is an advantage:
% it should be zero at the beginning of a line in the table (for other
% lines this is done by the |\\|).
%    \begin{macrocode}
\def\res@tsumline{\FCsc@l=\FCtc@l\loop\ifnum\FCsc@l>0
 \global\csname FCtot@\romannumeral\FCsc@l\endcsname=0
 \global\csname FCwd@\romannumeral\FCsc@l\endcsname=\z@
 \advance\FCsc@l by \m@ne
\repeat}
%    \end{macrocode}
% \end{macro}
% \begin{macro}{\resetsumline}
% To reset a sumline within a table, it should be done within a |\noalign|.
%    \begin{macrocode}
\def\resetsumline{\noalign{\res@tsumline}}
%    \end{macrocode}   
% \end{macro}
% That's it!
%\section*{Acknowledgement}
% Thanks to Karl Berry for valuable comments regarding the consistency of
% the installation procedure of this version.
%\bibliographystyle{plain}
%\begin{thebibliography}{9}
%\itemsep0pt
%\parsep0pt
%\parskip0pt
%\bibitem{array} Frank Mittelbach and David Carlisle. 
%    A new implementation of \LaTeX's |tabular| and |array| environment.
%\bibitem{booktabs} Simon Fear. Publication quality tables in \LaTeX.
%\bibitem{dcolumn} David Carlisle.  The |dcolumn| package.
%\end{thebibliography}
%\Finale
\endinput