Tikz cd

[brucemiller]

This PR provides a binding for tikz-cd, after fixing some infrastructure.

The critical bug was that ALIGN_STATE which tracks {} during scanning to enable or disable certain alignment tokens (&,\cr,...) was being done too late (in Stomach rather than Gullet). It's easy to miss some corner cases, so some sandbox testing is warranted.

This is a WiP since the resulting svg are displayed too low; there would seem to be one too many adjustments involving pgf@pixminy (?) , although other tikz pictures seem mostly correct.

[dginev]

I was just discussing with arXiv three of our existing successful Tikz conversions, and did a quick test with this PR - linking to the significantly older ar5iv editions:

• arXiv:1804.07114 has a few nice pgfplots. Figure 4 now raises errors, and the article hits a Fatal by overflowing the pushback limit. So a regression?
• arXiv:1501.02026 draws some grids. With the PR I seem to be getting an identically looking output, and no changes in the severity, great.
• arXiv:2004.08822 has a couple of geometric drawings. That also come out as expected.

So, focusing on the one pgfplots case that regressed: it indeed works with the latexml master branch. Isolating a minimal regressing example:

\documentclass{article}
\begin{filecontents}{coord.txt}
  nc	I_M_opt	I_J	I
  1.000000	0.000000	0.000000	0.001190
  2.000000	0.004133	0.003861	0.002870
  3.000000	0.005510	0.007181	0.004562
\end{filecontents}

\usepackage{pgfplotstable}

\begin{document}

\begin{tikzpicture}
\begin{loglogaxis}[legend style={cells={align=left}}]
\pgfplotstableread{coord.txt}\mytable
\addplot[solid] table[x=nc,y=I] {\mytable};
\addlegendentry{legend}
\end{loglogaxis}
\end{tikzpicture}

\end{document}

Edit: can confirm this was fixed by 5298bf8 !

[dginev]

Our arXiv sandboxes are looking healthy, and we can continue to improve post-release - tikz has a lot of terrain to test, e.g. #2326 .

I am bookkeeping the sandbox results for the v0.8.8 release here, results of this PR are in the 5th table from the top:
https://gist.github.com/dginev/abb72c989f4474a3a33300ed23ac6c86

I think the PR gets us another step closer to "faithful TeX interpretation", which is excellent. Looks good to merge.