Fix tex macros in pdflatex build of docs

[chrisjsewell]

Originally posted by @chrisjsewell in https://github.com/executablebooks/jupyter-book/pull/952#issuecomment-689366919

Currently, the equation with the tex macros in is wrapped:

```{only} html
$$
A = \bmat{} 1 & 1 \\ 2 & 1\\ 3 & 2 \emat{},\ b=\bmat{} 2\\ 3 \\ 4\emat{},\ \gamma = 0.5
$$
```

This should be working with LaTeX though, because I added the macros to the preamble. But still I get an error: Package array Error: Empty preamble: 'I' used, which I haven't got time to work out the issue right now.

[mmcky]

@AakashGfude would you mind checking the docs and seeing why latex preamble isn't getting passed through?

[AakashGfude]

@chrisjsewell @mmcky the issue is not showing up now when I check it in jupyter-book. Can anyone double-check. Then we can close this issue.

[mmcky]

@AakashGfude if I take away the only directive which targets html outputs I get the following tex warning on #1167

! Package array Error: Empty preamble: `l' used.

See the array package documentation for explanation.
Type  H <return>  for immediate help.
 ...                                              
                                                  
l.7707 \end{split}

I see this is in the tex

\newcommand{\bmat}{\left[\begin{array}}

and these is the problematic lines

\begin{equation*}
\begin{split}
A = \bmat{} 1 & 1 \\ 2 & 1\\ 3 & 2 \emat{},\ b=\bmat{} 2\\ 3 \\ 4\emat{},\ \gamma = 0.5
\end{split}
\end{equation*}

[alcrene]

I think this issue might be closeable as a red herring: the error reported relates to the matrix preamble, not the document preamble. I.e., one writes

\begin{array}{ll}
  1 & 1
\end{array}

The part {ll}, which specifies the horizontal alignment of each column, is also called a “preamble”. It must contain as many chars as there are columns in the array. Here, once the macros expand you get

\begin{array}{}
  1 & 1
  ...
\end{array}

The message is telling that your preamble ({}) is empty. Since the TeX processor can't look ahead and infer the number of required columns, it just guesses {l}, and then promptly chokes on the rest of the input.

At least with the current Jupyter Book stack (jupyter-book 0.12.3), the macros specified in my preamble file work as expected.

As a side note: amsmath already provides \begin{bmatrix} which does exactly what your \bmat macro purports to do.