One of the problems is that the anchors that Haddock currently generate aren't always legal in HTML, XHTML, or XML. I'd like to fix the anchor generation so that they are. If I do, then links between old and new generated Haddock pages will land on the right page, but won't always get to the right spot on the page.
If I recall correctly, Haddock's relative links hit a spot in the specs that was open to interpretation, as in: different browsers chose to interpret the specs differently. Hence the doubled 'NAME' tags: http://trac.haskell.org/haddock/ticket/45
Will this be a problem for anyone? On one's own machine, I imagine we can come up with a simple script that will just rebuild all the Haddock docs and that will take care of it.
What is the problem you have in mind, and what is your solution? Apart from differing browsers, there used to be tools out there that scrape information from Haddock pages to provide access to docs, such as haskellmode for Vim or the Hayoo! indexer: http://trac.haskell.org/haddock/ticket/113 To what extent any of these would be affected depends on the nature of the changes. Also, while haskellmode no longer constructs the relative urls itself, being able to find the relative url for any identifier with Haddocks in a scriptable/predictable way would be helpful (eg, to open Haddocks from GHCi). Claus