On Tue, 8 Aug 2006, Taco Hoekwater wrote:
Aditya Mahajan wrote:
2. In amsmath, most of alignment constructs exist in two forms: outer and inner. The mathalignment implemented in core-mat.tex corresponds to outer alignment. The inner alignment is same as outer alignment, but is only as wide as necessay. The most common amsmath inner alignment constructs are aligned and gathered. It is easiest to explain by means of an example. Suppose I want to type
a x + b y = c `\ } (simultaneous equations) d x + e y = f / ,
I want to be able to do
\defineinnermathalignment[aligned][n=2,left={\left.},right={\right\}}]
This sounds very close to \definemathmatrix, yes?
It is, and my first thought was that this can be achieved using \definematrix (that is why the previous requests for location= and style=). However, all my attempts to use a matrix failed because: 1. Matrix does not "see" beyond the \NCs as align. Comapre the output of \startformula \startalign[n=2] \NC a \NC = bx + c \NR \NC \NC + ey \NR \stopalign \stopformula with \startformula \startmatrix[n=2,distance=0pt, style=\displaystyle] \NC a \NC = bx + c \NR \NC \NC + ey \NR \stopmatrix \stopformula The '+' in the second line comes out as a unary operator rather than a binary operator. This can be corrected by using \NC{}+ but should the user really know the ugly implementation details? 2. Matrix does not correct interline space. Compare \startformula \startalign[n=2] \NC f(x) \NC = \int_{-\infty}^{\infty} \phi(y-x) dy \NR \NC \NC = \sum_{i=-\infty}^{\infty} \hat phi(i-x) \NR \stopalign \stopformula with \startformula \startmatrix[n=2,distance=0pt, style=\displaystyle] \NC f(x) \NC = \int_{-\infty}^{\infty} \phi(y-x) dy \NR \NC \NC = \sum_{i=-\infty}^{\infty} \hat phi(i-x) \NR \stopmatrix \stopformula The two lines are too close. I thought that it might be easier to simply wrap the whole align around a hbox as these inner alignments need not break across pages. If matrix can be enhanced to take care of the above two requirements then matrix is fine. However, if I just want a matrix, then both above behaviours of the matrix are correct. I am not sure what kind of interface matrix should have to behave in both ways (the current matrix behaviour and the requested aligned behaviour). Aditya