Dear all,

today I did some changes to simplify the math plug in a little bit and as a preparation to the LaTeXML integration: Please review and comment at: https://gerrit.wikimedia.org/r/#/c/28824/

Best regards Moritz PS: I had some problems with the commit messages. They were not accepted so the text is lost. What I basically did that I separated the code of the divers ways of rendering into separate classes that extend a common base class.

On Mon, Oct 15, 2012 at 12:13 AM, Moritz Schubotz physik@physikerwelt.de wrote:

HI DJ,

first of all thanks a lot for your helpful comments.

On Sun, Oct 14, 2012 at 10:31 PM, Derk-Jan Hartman d.j.hartman+wmf_ml@gmail.com wrote:

...

On 14 okt. 2012, at 17:45, Moritz Schubotz physik@physikerwelt.de wrote:

...

Summary: I'm looking for people to discuss about the parsing of math.

The last person who really did some work on the extension was Brion Vibber. Almost all of the (limited) work that has gone into the extension over the last 1,5 year was actually based on the efforts of User:Nageh on the English wikipedia and his MathJax userscript. Beyond these MathJax changes, the extension has not seen much developer involvement for quite some while.

...

I came up with a proposal for a new version of the rendering of the <math> tag. I proposed to use LaTeXML to convert the LaTeX expressions in the math tag to MathML. If the browser is not capable of displaying MathML I use MathJaX to display the MathML output in the browser. My implementation (the LaTeXML branch) has only a few very little differences in contrast to the master branch. I have the feeling that the php code of the math extensions could be improved. For example I'd suggest to put all the texvc related stuff to another class. Furthermore I was thinking about an asynchronous rendering of the formula, which would speed up page loading time especially for major edits. At http://wiki.physikerwelt.de/images/text_math_search.pdf you find the draft of a paper where I describe in detail what I changed and why it is an improvement. The paper will appear soon in the postconference proceedings of CICM2012. Now I want to figure out, who is working on the development of the math extension, and who wants to discuss with me about our ideas. I'm open to any kind of suggestions and questions.

Discussion is always welcomed and in this case probably best on the mailinglist.

You seem to propose switching our current texvc -> image converter (by some users extended by using MathJax) with a LaTeX -> MathML presentational renderer that would use MathJax as a backup. The renderer would also output Content MathML to make the content easier to index and search. In general I would support this, definitely morally.

I see MathJaX rather as a renering engine that supports browsers that do not such a good job in displaying MathML, rather than a fall back alternative.

...

There are however some problems here as well.

• If a browser does not support MathML and not support (have enabled) Javascript, you would see nothing reading the article.

MathML is also capable of producing images. However this feature is more designed for the rendering of whole documents rather than for small latex snippets. So this would require a little customization.

...

• Browser support for MathML is still spotty. It would be nice if we can gather some actual numbers on this. There are apparently also significant bugs in rendering implementations. I believe this is one of the reasons that the default of MathJax is HTML+CSS now, and not MathML.

• You are determining support serverside right now ? This would require fragmenting the cache when we move forward.

I'm nor sure if I understand, what you mean by that. The math table becomes much larger as it is if you use texvc, since the matml column contains all the mathml code which is not really optimized with regard to space.

...

• I think that currently a large group of users would fall into your MathJax group. Especially those with older browsers. However, we don't particularly like MathJax for that usergroup. It is incredibly slow on the client side and this greatly affects the userbase with less capable computers (of which we have quite a lot). This is one of the main reasons why we are not particularly pushing for this solution as a default right now (let alone as the primary backup method). I would seriously consider just skipping MathJax (at least by default) and just show images for the fallback option.

MathJaX is slow if you want to display LateX stuff. If you want to display MathML with the help of MathJaX- that's what I propose- MathJaX is reasonably fast.

...

• In order to provide proper MathML support, a proper font is required. Many Operating Systems do not yet have this font support. This is something that possibly could be solved with WebFonts, but will require a bit of work.

• MathML does not, I believe, provide everything that is possible with texvc. This is another issue that we are currently seeing with MathJax, though we (Nagheh actually) has added most of them to MathJax by now. For MathML the solution seems not so simple however. This probably means we either need to phase out part of what we currently support (always complicated) or fragment the implementation for these particular situations.

This would be really interesting to see which are those things, that can not be displayed with MathML.

...

• I'm not aware of the current quality of LaTexML. Is there some information about the 'completeness' of the capabilities and quality of the implementation? Comparisons with other implementations ? That would be useful information.

I'll provide some information soon.

...

• Outputting both Presentational and Content MathML into the HTML will significantly increase the page size.

This is true. I saw that in my experiments.

...

• LaTeXML will have to be reviewed by one of the developers of the security team.

Yes. I think the major advantage of LaTeXML in contrast to texvc is that it can be run on a separate machine, with very restricted rights.

...

These would be the issues that I think would require assessment and solving at some point before a full solution can be deployed on Wikipedia. But of course, we can just start and expose the functionality at a later time.

DJ

P.S. Seems that MathJax 2.1 will be released soon: http://www.mathjax.org/2012/10/01/news/mathjax-v2-1-beta-now-available-on-th... _______________________________________________ MediaWiki-l mailing list MediaWiki-l@lists.wikimedia.org https://lists.wikimedia.org/mailman/listinfo/mediawiki-l

Thanks again for the feedback.

Best regards Moritz

-- Mit freundlichen Grüßen Moritz Schubotz

Telefon (Büro): +49 30 314 22784 Telefon (Privat):+49 30 488 27330 E-Mail: schubotz@itp.physik.tu-berlin.de Web: http://www.physikerwelt.de Skype: Schubi87 ICQ: 200302764 Msn: Moritz@Schubotz.de