fix(aria-allowed-attr): no inconsistent aria-checked on HTML checkboxes #3895
Conversation
| import { getRole, allowedAttr, validateAttr } from '../../commons/aria'; | ||
| import { isFocusable } from '../../commons/dom'; | ||
| import cache from '../../core/base/cache'; |
There was a problem hiding this comment.
I didn't think this cache was useful, so I've left it out on my rewrite. If we ever want to cache getRole, it should be memoized. And unlike this current implementation I'm only looking up the table role once per row, not for every attribute.
| if (validateAttr(attrName) && preChecks[attrName]?.()) { | ||
| invalid.push(attrName + '="' + virtualNode.attr(attrName) + '"'); | ||
| } else if (validateAttr(attrName) && !allowed.includes(attrName)) { | ||
| invalid.push(attrName + '="' + virtualNode.attr(attrName) + '"'); |
There was a problem hiding this comment.
This was really weird and messy. Took some time refactoring and cleaning this up. Much easier to grok now. The gist of what this does is that if you're using one of these validTreeRowAttrs on a table or grid, it should fail because those attrs are only allowed on treegrid.
| export default function ariaConditionalAttrEvaluate( | ||
| node, | ||
| options, | ||
| virtualNode | ||
| ) { | ||
| const role = getRole(virtualNode); | ||
| if (!conditionalRoleMap[role]) { | ||
| return true; | ||
| } | ||
| return conditionalRoleMap[role].call(this, node, options, virtualNode); |
There was a problem hiding this comment.
I'm not sure if this was over-engineering or not. Having two new checks would be fine, but then if we wanted to add more that becomes kind of bloated. These things are mutually exclusive because they run for different roles.
There was a problem hiding this comment.
They certainly should be their own functions, but having them be "evaluate" methods is interesting, especially since they are two evaluate functions without an associated metadata file or test file.
I can see why having them function like evaluate methods is helpful so you can call data() in them, but you could do that using a destructured return and calling data in here. I think I would prefer them as a common function or something rather than their own evaluate files. That way they can't accidentally be used in a rule, and they can be tested in their own files rather than combining the tests into a single file for aria-conditional-attr.
There was a problem hiding this comment.
Guess I can put them into commons/aria then.
| @@ -7,7 +7,7 @@ | |||
| "description": "Ensures ARIA attributes are allowed for an element's role", | |||
| "help": "Elements must only use allowed ARIA attributes" | |||
| }, | |||
| "all": [], | |||
| "any": ["aria-allowed-attr"], | |||
| "all": ["aria-allowed-attr", "aria-conditional-attr"], | |||
There was a problem hiding this comment.
I really don't like "none" checks. Always messes with my head that false means pass. Can't undo that on the other checks but I don't want to make it worse.
| const isChecked = vNode => | ||
| vNode.actualNode ? !!vNode.actualNode.checked : vNode.hasAttr('checked'); |
There was a problem hiding this comment.
Using actualNode.checked because if you toggle a checkbox it doesn't update the checked attribute. It's not a prop on virtualNode
|
Back to draft, need to handle indeterminate checkboxes. |
| export default function ariaConditionalAttrEvaluate( | ||
| node, | ||
| options, | ||
| virtualNode | ||
| ) { | ||
| const role = getRole(virtualNode); | ||
| if (!conditionalRoleMap[role]) { | ||
| return true; | ||
| } | ||
| return conditionalRoleMap[role].call(this, node, options, virtualNode); |
There was a problem hiding this comment.
They certainly should be their own functions, but having them be "evaluate" methods is interesting, especially since they are two evaluate functions without an associated metadata file or test file.
I can see why having them function like evaluate methods is helpful so you can call data() in them, but you could do that using a destructured return and calling data in here. I think I would prefer them as a common function or something rather than their own evaluate files. That way they can't accidentally be used in a rule, and they can be tested in their own files rather than combining the tests into a single file for aria-conditional-attr.
Co-authored-by: Steven Lambert <2433219+straker@users.noreply.github.com>
WilcoFiers
changed the title
Because I want to add explicit messages without confusing the missing messages of the aria-allowed-attr check I decided to create a new
aria-conditional-attrcheck. I moved existing code that checks that row attributes that are allowed on treegrid but not on table or grid into that new check too.I put the checkbox code and row code into separate evaluate methods, that then get invoked by ariaConditionalAttrEvaluate. Alternatively they could have been different checks entirely. I'm not keen on adding a new check every time we have a role for which we want some extra logic, that felt excessive. Then again there's only two of them so maybe they could have just been two checks? I'm open to changing this.
Closes: #3307