Webrat Webrat

...or at least surprising.

Since November, link_element has worked by finding all link elements whose text content, HTML content or title attribute matches the locator value, and then returning the element whose text content is the shortest.

Namely: there are now several criteria for deciding which elements are potential candidates, but only one of those criteria is used when deciding which element is the "best" match.

This change broke a whole bunch of stuff for me when I upgraded Webrat. In particular I have many pages that contain links with no text content (e.g. images) but whose HTML content (e.g. alt tags) often matches the locator value. These no-text-content links are never what I want to click on with click_link (and if they were I'd give the locator a value like alt="...") but they invariably get picked by link_element because they have the shortest (i.e. completely empty) text content.

The attached patch fixes this behaviour, or at least makes it more predictable and consistent: we now keep track of the matching content for each link, and use the actual length of that content when scoring the match candidates. This still might not be what everyone wants but it's much closer to being correct. I've left the old API in place in case anyone was depending on it, but otherwise matches_text?, matches_id? and matching_links can be removed.