Bug 259725

Whenever there is an HTML input with a role of "combobox", and the value in that input is a single character, Voiceover will read the input's label as a string of individual characters, rather than the whole word or series of words that makes up the label.
 
Simple minimal example – if you run the following HTML (from an HTML file running in the browser, or using a site like Codepen):
 
<label for="test">Label</label><input id="test" role="combobox" value="1" />
 
you will hear – and see – the label announced as “l – a – b – e – l”, rather than the word “label”.
 
For a more realistic example, you can try with the Aria Practices combobox example at https://www.w3.org/WAI/ARIA/apg/patterns/combobox/examples/combobox-autocomplete-list/ - if you either manually enter a single character in the combobox, or change the HTML so there are options consisting of a single character and then select one, the same behaviour is apparent.
 
Some further observations I’ve made in testing:
 - the bug happens with VoiceOver in Safari, Chrome and Firefox, but does not happen with NVDA on Windows in Chrome, Firefox or Edge. So it appears to be specific to VoiceOver (or MacOS) rather than anything in a browser. [I initially reported the bug to accessibility@apple.com but received a reply saying to raise it here, even though it seems to be not actually a Webkit bug...]
 - it happens both when the input value is set programmatically and when entered by the user
 - it still happens when the label is set with aria-label or aria-labelledby
 - it appears to be specific to the “combobox” role (I’ve tried other roles and not yet seen the bug with any other)
 - it does NOT happen with a native HTML <select> element though
 - it doesn’t require an HTML <input>, a <div> with a role of “combobox” and the value set by the text content behaves the same way with VoiceOver
 - it doesn’t appear to matter what the label is (eg length/casing)