Poked summary; recommend just doing this in PHP...

A few issues that I noted while playing around some at https://github.com/scfc/mediawiki-math-extension (ATM ~ 90/1000 deviations in DVIs):

• Despite being one of its purposes, there are no testcases that "rogue" TeX commands (I/O, "\catcode", etc.) are actually rejected.
• texvc not only validates the input and generates HTML, but also /transforms/ the TeX source:
  • Some functions are aliased ("\uarr" => "\uparrow").
  • White space is added and deleted.
  • "M^{II}_3N_2" gets rearranged to "M_{3}^{{II}}N_{2}".

IMHO the transformations are bad (TM):

• A reader cannot simply copy and paste apparent TeX from a wiki source to other applications.
• An editor cannot simply copy and paste apparent TeX from a TeX source to MediaWiki.
• It lays an unnecessary burden on the usage of other renderers and editors. For example, MathJax doesn't recognize "\uarr", but only "\uparrow".

So I think we should settle for a purely validating parser.

Omitting the transformations will of course also change the output and thus the outputhash column.  So a migration plan would probably have to include expiring PNGs generated before the deployment.

Interestingly, sometimes MathJax seems to emulate texvc's behaviour: If you TeX "m_\mbox{t} = m \cdot \frac {1}{1 - \sum k}", the result is "m" with the rest of the line as subscript. texvc transforms that to: "m_{{\mbox{t}}}=m\cdot {\frac {1}{1-\sum k}}" which only puts "t" as subscript, and at least the MathJax demo at http://www.kenschutte.com/mathjax uses the same rendering.

I fear if we do not clearly define the grammar and its meaning, we will face a situation similar to MediaWiki's parser.