mathalign.sty

\NeedsTeXFormat{LaTeX2e}[1995/12/01]
\ProvidesClass{mathalign}%
  [2005/07/13 Flexible Math Alignment]

%
% This class provides the environment mathalign, which implements a
% slightly more flexible way of math alignment than eqnarray does.
% Well, it's what we need, anyway.
%
% Without \\, & or \label, mathalign behaves like displaymath:
%
%     \begin{mathalign}
%       a^2 + b^2 = c^2
%     \end{mathalign}
%
% Lines are separated by \\ and (by default) will be left aligned.
%
%     \begin{mathalign}
%       (a + b)^2 = a^2 + 2ab + b^2 \\
%       (a - b)^2 = a^2 - 2ab + b^2 \\
%       (a + b)(a - b) = a^2 - b^2
%     \end{mathalign}
%
% Using _one_ & allows aligning the relations -- multiple ones are
% ignored.
%
%     \begin{mathalign}
%        (a + b)^2 &= a^2 + ab + ba + b^2 \\
%                  &= a^2 + 2ab + b^2
%     \end{mathalign}
%
% Using one line without & and a following one with & indents the
% relation by 1cm
%
%     \begin{mathalign}
%       (x - a)(x - b)(x -c)\cdots(x - z) \\
%         &= (x - a)\cdots (x - x)\cdots (x - z) = 0
%     \end{mathalign}
%
% If a line starts with an equivalence relation, it is put left of the
% whole alignment:
%
%    \begin{mathalign}
%           a^2 + b^2 &= c^2       \\
%      \iff       a^2 &= b^2 + b^2 
%    \end{mathalign}
%
% By default, \Leftrightarrow, \Leftarrow, \Rightarrow and \iff are 
% recognised, but every option given to this class which is neither
% fleqn, *reset nor *fakelabels will be added to this list. If *reset
% is among the options, the default list will be replaced instead of
% being added to.
%
% Every line can be given a number by specifying \label:
%
%     \begin{mathalign}
%       a^2 + b^2 = c^2  \label{pyth}
%     \end{mathalign}
%
% is just like the equation-environment. If *fakelabels is among the
% options, we do not display the numbers but instead the labels
% prefixed with `EQ:'.
%
% The environment can be used both in vertical mode and in math mode.
% In vertical mode it behaves like eqnarray, aligning labels to the
% borders of the textarea (it understands the documentwide fleqn
% option). In math mode it packages everything up in a \vcenter,
% typesetting labels a \qquad away from the side of the equation.
%

% First, configuration. Since we want our documents to look
% consistent, we will have to keep track of where the equation
% numbers go.

\newif\ifma@fleqn \ma@fleqnfalse

\DeclareOption{fleqn}%
  {\ma@fleqntrue}

% Then, we sometimes (that is, in the preview) don't want equation
% numbers to show up at all -- we want the label names to take
% their place.

\def\ma@labeltext     #1{\@eqnnum}
\def\ma@fakelabeltext #1{(EQ:#1)}

\DeclareOption{*fakelabels}%
  {\let\ma@labeltext\ma@fakelabeltext}

% Everything else is dedicated to managing the list of equivalence
% relations. This is the default, ...

\def\ma@equivrelations%
  {\do\Leftarrow \do\Rightarrow \do \Leftrightarrow \do\iff}

% ... this is how we clear it out, ...

\DeclareOption{*reset}
  {\let\ma@equivrelations\empty}

% ... and this is how we add to it.

\def\ma@doaddtoequivrels#1\ma@end%
  {\def\ma@equivrelations{#1}}
\def\ma@addtoequivrels#1%
  {\expandafter\ma@doaddtoequivrels\ma@equivrelations\do#1\ma@end}
\DeclareOption*%
  {\expandafter\ma@addtoequivrels\csname\CurrentOption\endcsname}

\ProcessOptions

% Now for the actual implementation. Since we gather all the
% material in the first column and split it up ourselves, we will
% make the alignment character active. Note that the &s in the
% macros below are _not_ active.

\bgroup
  \catcode`\&=\active
  \gdef\ma@activateamp%
    {\catcode`\&=\active
     \let&\ma@column
     \let\begin\ma@begin}%
\egroup

% One major problem with the way things are done here is that
% (since we change the catcode of &) nested alignments just
% won't work the way it should. Unfortunately, I don't know a 
% way to redefine \halign which always correctly restores
% the catcode and undoes the change, even in the presence of
% \edef-trickery as used in array. We can, however postulate
% that everything within an environment ought to be a normal &.
% This still breaks commands which take ampersands as an argument,
% but it is better than nothing ...

% FIXME can we make this work reliably?

\let\ma@begin@orig\begin
\let\ma@begingroup@orig\begingroup

\def\ma@begin@hook%
  {\let\begingroup\ma@begingroup@orig \begingroup
   \catcode`\&=4 \let\begin\ma@begin@orig}
\def\ma@begin%
  {\let\begingroup\ma@begin@hook \ma@begin@orig}

% Communication between the different columns happens via the
% following means:

\newbox\ma@lhs \newbox\ma@rhs

\def\ma@doiflhs%
  {\relax\ifvoid\ma@lhs
     \expandafter\@secondoftwo
   \else
     \expandafter\@firstoftwo
   \fi}

\def\ma@doifrhs%
  {\relax\ifvoid\ma@rhs
     \expandafter\@secondoftwo
   \else
     \expandafter\@firstoftwo
   \fi}

% Note that this assumes that \ma@buildline _always_ empties
% \box\ma@rhs (if possible).

\def\ma@init@communication%
  {\global\let\ma@curlabel\empty
   \global\let\ma@curequiv\empty}

% Every line ends with the following macro (which will be named \\
% in the end). This macro terminates the first column (thus allowing
% \ma@split to do its work, which will ultimately leave the first
% column empty) and then starts rebuilding the columns.

\def\ma@cr%
  {\@ifnextchar\ma@crcr\empty\ma@crcr}
\def\ma@crcr%
   {&\ma@buildline}


% Splitting works by typesetting the tokens in an hbox. We have to
% split this macro in two -- passing the tokens as an argument
% prevents expansion, so TeX will not find the & buried in a \\.

% FIXME: No need to split, we can fake it with \aftergroup. But is
% it worth it?

\def\ma@split@start%
  {\ma@init@communication
   \global\setbox\ma@lhs=\hbox
     \bgroup$\displaystyle\ma@startline}
\def\ma@split@stop%
  {$\egroup}

% This allows for example (global) redefinition of things like the
% current label or the desired equivalence relation.

\let\ma@label@orig\label
\def\ma@label#1%
  {\refstepcounter{equation}\ma@label@orig{#1}%
   \gdef\ma@curlabel{\ma@labeltext{#1}}}

% The latter is a bit special, since it is only supposed to be magical
% at the beginning of a line. We therefore introduce a new signal and
% check for it.

\def\ma@signalamount {42sp }
\def\ma@startline    {\kern\ma@signalamount}

\def\ma@equiv#1%
  {\relax\ifdim\lastkern=\ma@signalamount
     \gdef\ma@curequiv{$#1$\qquad}%
   \else
     #1%
   \fi}

% Actually, for equivalence relations this is supposed to happen
% automatically. So let's activate them:

\def\ma@rel@{ma@rel@}

\def\ma@doactivaterels#1%
  {\expandafter\let\csname\ma@rel@\string#1\endcsname#1%
   \edef#1{\noexpand\ma@equiv
           \expandafter\noexpand\csname\ma@rel@\string#1\endcsname}}

\def\ma@activaterels%
  {\let\do\ma@doactivaterels \ma@equivrelations}

% Switching columns is realized by terminating the first and starting
% another hbox in a different register. We remember the fact that
% we have done so, among other things to ignore more than one &
% in one row.

\def\ma@column%
  {\ma@doiflhs\empty
     {$\egroup
      \global\setbox\ma@rhs=\hbox
        \bgroup$\displaystyle{}}}

% Now that we have \ma@curequiv, \ma@curlabel and the two sides if the
% equation, we can start building the actual alignment. Note that the
% template für these columns will be empty, so we have to insert all
% the \hfils here.

% In case there is no rhs, we fix the left column to be 1cm wide and
% empty and let the right column lap into it. This gives the
% impression of a \lefteqn{...} while still taking the width of the
% line into account.

\def\ma@buildline%
  {\ifma@fleqn
     \ma@curlabel\hfill\ma@labelspacing
   \fi&%
   \hfil\ma@curequiv&%
   \ma@doifrhs
     {\hfil\unhbox\ma@lhs&%
      \unhbox\ma@rhs\hfil&}%
     {\kern1cm&%
      \kern-1cm\unhbox\ma@lhs\hfil&}%
   \ifma@fleqn\else
     \ma@labelspacing\ma@curlabel\hfil
   \fi\cr}

% The actual alignment now only has to activate the commands defined
% above and then get out of the way.

\def\ma@align@start%
  {\let\label\ma@label \let\\\ma@cr
   \ma@activaterels
   \ma@align\bgroup
     \ma@split@start##\ma@split@stop
                        &##%
     \tabskip\@centering&##%
     \tabskip\z@        &##&##%
     \tabskip\@centering&##%
     \tabskip\z@\cr}
\def\ma@align@end%
  {\ma@crcr\egroup}

% Now we only have to make sure that the ampersand is actually active
% when the equation is scanned. Oh, and look if we are used in math
% mode -- in this case we package everything up in a box and adjust
% the spacing at the labels.

\def\mathalign%
  {\ifmmode
     \let\ma@labelspacing\qquad \let\ma@align\ialign
     \vcenter\bgroup\let\ma@mathalign@end\egroup
   \else
     \let\ma@labelspacing\empty \def\ma@align{\ialign to \displaywidth}%
     $$\def\ma@mathalign@end{$$\@ignoretrue}%
   \fi
   \ma@activateamp\ma@align@start}
\def\endmathalign%
  {\ma@align@end\ma@mathalign@end}

\endinput