Performance improvements for utils.isHidden

[ckundo]

I evaluated axe-core with Firefox performance tools, running it in http://www.html5accessibility.com/html5elements/ page. I saw that utils.isHidden costs about 20% of the total execution time of axe.a11yCheck call. On this particular page that equates to about 200ms. Thoughts on improving performance in that method? I'm happy to chip in or share ideas, but wanted to capture the issue first.

[WilcoFiers]

Interesting. You confirmed something I've been wondering about myself. I'll have a go at it. Maybe memoization could help out here. We are touching that function an awful lot after all.

[jnurthen]

Have you considered checking the CSS offsetWidth , offsetHeight etc. (I use offsetTop and offsetLeft too - but I'm not sure that is necessary). If something is display:none these should be 0 (or null in some browsers). This avoids walking up the DOM tree to find out if an ancestor is hidden.

[dylanb]

@jnurthen If you look at the code, you will see that it is a bit more complicated than just display:none https://github.com/dequelabs/axe-core/blob/master/lib/core/utils/is-hidden.js but we could possibly improve the performance somewhat by doing that test first and then the other tests. Anyone want to take a shot at doing that?

[jnurthen]

I question a little your testing of aria-hidden here. I think (only a 5 minute look so apologies if incorrect) that your isHidden routine gets called on nodes when excludeHidden is not set to false in a rule.

I don't think it is correct to exclude nodes hidden using aria-hidden=true from a colour contrast check, only nodes hidden using techniques which hide from nodes visually should be excluded from contrast checks. I really think you need either 2 different isHidden checks or remove the aria-hidden check from this basic check which prevents the node even getting checked by the real check and check aria-hidden in the actual rule when appropriate.

Testcase for why this fails at http://output.jsbin.com/xulehoniyu.
Both lines should fail colour contrast, but only 1 gets picked up by aXe currently.

[dylanb]

@jnurthen Can you open a separate issue for that specific rule?

[jnurthen]

Had a feeling you would say that
#170