An objection has been raised against longdesc (and the use cases which rely on information not presented in the same page as an image) that it is "hidden metadata" and its quality and relevance are likely to deteriorate over time. While it is apparent that content which is immediately visible can be more readily maintained in a simplistic content management workflow, this argument falsely assumes that is always the case, and further falsely assumes that the presence of some level of degradation is a fatal problem for the use of longdesc to improve accessibility.
In fact substantial amounts of Web content are maintained in processes which assume the presence of "hidden" metadata (which is actually readily discoverable) and require maintenance of that data as well as of the "primary" content (that is immediately visible by default). In addition, where an image is not changed, it is unlikely that a well crafted description needs to be changed, so there is no inherent degradation.
While longdesc does not require "hidden" metadata (it can be used simply to unambiguously identify inline content of the page as a description for an image), there are use cases which benefit from the ability to support it. Images maintained as resources in a content management system, or even just by copying and pasting the img tag with a link inside it such as longdesc provides, can easily re-use the description rather than requiring that it too be replicated. This matches common workflows for managed content, and there is no reason to make it difficult. It is normal in authoring tools that copying objects between pages may require rewriting links appropriately (their destinations are, after all "hidden"), and this does not seem to break the web.