Re: Collection of pages using native MathML?

On 8/5/19 1:26 PM, Neil Soiffer wrote:
>>     there is also the original mathml test suite at W3C of course.
>>
>>
>>     Right, I can use it too although my preference would be for
>>     "concrete" content rather than tests
> 
> Real world examples will give you realistic speed comparisons and help 
> with robustness, but a test suite will be much better at testing 
> robustness and correctness of your implementation and any polyfills. The 
> collections, especially if they are generated from TeX, won't test whole 
> categories of functionality. E.g,. mpadded, mphantom, or stretchy 
> min/max may not be part of the output of the converters. Having said 
> that, the MathML test suite is a disaster in terms of being focused -- 
> lots of people contributed tests so there is a ton of overlap in what is 
> tested. Still, it does ASFAIK cover all element and all attributes and 
> includes some tests that were meant to challenge implementations in 
> terms of size of expression and speed of MathML rendering (see Torture 
> Tests section). I was pleased to see that Firefox does a pretty good job 
> for many of these Torture tests, especially speed-wise, when I just checked.

I'll second Neil's comments. In fact, LaTeXML intentionally avoids a 
number of constructs that aren't well supported
(mlabelledtr jumps immediately to mind).  OTOH, The same
reason the constructs are avoided is why they're candidates
for removal from core, so...

That said, real-world examples are a nice focus, since it's
what people are actually using. For testing purposes, you'll
want to download snapshot samples from DLMF or arXiv, since
they'll always be subject to updates, improvements and corrections.  And 
if you do that, you have the possibility of
hacking those samples to remove/disable the calls to load
MathJax.

[A good system for browser sniffing &/or cookies has long
been on my wishlist]

bruce

Received on Monday, 5 August 2019 17:50:08 UTC