I found this while working iwth my photos on Tweetphoto.com. The construct is like this: <a href="something" title="something"><img src="bla.png" alt=""/></a> or similar with nested spans, as can be seen with all of the photos. Result: neither the "click here to enlarge this photo" links for each photo appear in NVDA's virtual buffer, nor the "Edit geolocation" and similar links that follow each photo message. When calculating the name for a link, we fall back on the title if we detect that the inner text is empty. We should make this algorithm smarter to detect that the inner stuff is really nothing useful, and then use the title in any case.

Oops, was a bit too quick with filing the bug. We're actually doing it right, we're presenting the title as the name. However NVDA still ignores the link. I'll file a ticket against NVDA.

Created NVDA ticket 710: http://www.nvda-project.org/ticket/710