Hi Nicolas,
I think this patch is a welcome simplification. Would it be possible to
merge the code that is used for reading babel header args (things like
“:results output :file foo.txt”) with the code from the exporter?
Unless they are parsed by the same code, the two syntaxes will differ in
subtle and headache-inducing ways (for users and developers).
Both methods as it stands have their pros and cons.
org-export-read-attribute seems* to give the wrong result for the case of
#+ATTR_FOO: :key "one :two three" -> (:key "\"one" :two "three\"")
Whereas org-babel-parse-header-arguments (returning an alist, not a
plist) gives correctly ((:key . "one :two three"))
OTOH ":key (one :two)" makes o-b-parse-header-arguments give an error
because it tries to read (one :two) as elisp, and “one” isn’t an elisp
function. For o-x-read-attribute we correctly(?) get
(:key "(one :two)").
o-b-parse-header-arguments can cope with ":key '(one :two three)" (note
the single quote to not barf on the lisp-like syntax) giving
((:key one :two three)) – which is a notational variant of
((:key . (one :two three))) – whereas o-x-read-attribute gives
(:key "'(one" :two "three)")
In this last case I think that either value is arguably correct – but it
should be consistently one and not the other.
* For testing, I extracted the following function from
org-export-read-attribute.
#+begin_src emacs-lisp
(defun o-test (value)
(let ((s value) result)
(while (string-match
"\\(?:^\\|[ \t]+\\)\\(:[-a-zA-Z0-9_]+\\)\\([ \t]+\\|$\\)"
s)
(let ((value (substring s 0 (match-beginning 0))))
(push (and (not (member value '("nil" ""))) value) result))
(push (intern (match-string 1 s)) result)
(setq s (substring s (match-end 0))))
;; Ignore any string before the first property with `cdr'.
(cdr (nreverse (cons (and (org-string-nw-p s)
(not (equal s "nil"))
s)
result)))) )
#+end_src
--
Aaron Ecay
