What will happen when you need to insert a paragraph? Or delete? Will al unique links shift or stay the same?

Adding a new paragraph won't change any of the existing IDs. The ID-generating code just adds a new one. Deleting a paragraph would remove its ID, but I'll probably avoid deleting paragraphs for that reason. FWIW, here's my current hack for adding IDs:

 (defun blog:idparas () 
  (interactive)
  (save-excursion
    (let ((paracount 0)
	  (paraid 0))
      (beginning-of-buffer)
      (re-search-forward "\/info\\|\/essayinfo" nil t nil)
      (while (re-search-forward "\\|]*\\)>"
				nil t nil)
	(if (match-string 3)
	    (let ((attributes (match-string 3)))
	      (if (string-match " id=[\"\']\\([^\"\']+\\)[\"\']" attributes)
		  (let ((id (match-string 1 attributes)))
		    (if (string-match "^p\\([0-9]+\\)$" id)
			(let ((idnum (string-to-number (match-string 1 id))))
			  (if (> idnum paraid)
			      (setq paraid idnum)))))))))
      (setq paraid (+ paraid 1))
      (beginning-of-buffer)
      (re-search-forward "\/info\\|\/essayinfo" nil t nil)
      (while (re-search-forward "\\|]*\\)>"
				nil t nil)
	(if (or (not (match-string 3))
		(not (string-match " id=[\"\']\\([^\"\']+\\)[\"\']" (match-string 3))))
	    (progn
	      (forward-char -1)
	      (insert (format " id='p27" paraid))
	      (setq paracount (+ paracount 1))
	      (setq paraid (+ paraid 1)))))
      (message (format "Inserted IDs" paracount)))))(defun blog:idparas () 
  (interactive)
  (save-excursion
    (let ((paracount 0)
	  (paraid 0))
      (beginning-of-buffer)
      (re-search-forward "\/info\\|\/essayinfo" nil t nil)
      (while (re-search-forward "\\|]*\\)>"
				nil t nil)
	(if (match-string 3)
	    (let ((attributes (match-string 3)))
	      (if (string-match " id=[\"\']\\([^\"\']+\\)[\"\']" attributes)
		  (let ((id (match-string 1 attributes)))
		    (if (string-match "^p\\([0-9]+\\)$" id)
			(let ((idnum (string-to-number (match-string 1 id))))
			  (if (> idnum paraid)
			      (setq paraid idnum)))))))))
      (setq paraid (+ paraid 1))
      (beginning-of-buffer)
      (re-search-forward "\/info\\|\/essayinfo" nil t nil)
      (while (re-search-forward "\\|]*\\)>"
				nil t nil)
	(if (or (not (match-string 3))
		(not (string-match " id=[\"\']\\([^\"\']+\\)[\"\']" (match-string 3))))
	    (progn
	      (forward-char -1)
	      (insert (format " id='p27" paraid))
	      (setq paracount (+ paracount 1))
	      (setq paraid (+ paraid 1)))))
      (message (format "Inserted IDs" paracount)))))

I think I'd rather do that with XSLT, but I haven't bothered to get that working just yet.

Except, of course, "id=" in that code should now be "xml:id=". For the time being, at least.

Hmm, I hope they are not going to change it to xmlid just because some specfication is flawed. The discussion it generates on www-tag is interesting though. Especially about other specifications introducing backwards incompatible changes in the new namespace. (As far as I understand it.)

I hope so too. Alas, I don't expect the TAG to get a chance to discuss it before the 21st and perhaps not before the plenary.

The comments feeds have now been upgraded to Atom Format 05 too.