Slow rules should turn themselves off on large pages

[WilcoFiers]

Testing a page with 70k DOM nodes, axe was unable to complete. Turning off just a few rules known to be particularly slow solves this problem. I created this configuration, which enables axe to complete in 26s.

{
  "rules": [{
     "id": "color-contrast",
     "enabled": false
   }, {
     "id": "duplicate-id-active",
     "enabled": false
   }, {
     "id": "duplicate-id-aria",
     "enabled": false
   }, {
     "id": "duplicate-id",
     "enabled": false
  }],
  "checks": []
}

One possible solution would be to provide a "slow" flag, which when running on very large pages turns those rules off, returning a single "incomplete" result. We can add an option to axe that will force to alway run slow rules. But having this feature will help axe-core complete even on very large pages. We could consider different thresholds for different rules based on the number of DOM nodes.

[AutoSponge]

If you're looking for a benchmark, I can suggest https://js.tensorflow.org/api/latest/ (only 23k nodes but painfully slow).

[mohanraj-r]

A way to label and exclude slow rules seems like a good idea 👍 @WilcoFiers

Btw in my tests I have found that oddly link-name rule also takes almost the same amount of time as color-contrast

[dylanb]

@WilcoFiers perhaps we could use the AOM to speed up the calculation of accessible name for situations that are simple (i.e. do not fall into the "lowest common denominator" category) - thoughts?