I wrote:
> This code solves many problems of `hs-hide-block' mentioned in the
> first mail, but it breaks `hs-hide-level'.  It seems that current
> MDATA-SELECTOR support is limited and needs some fixes if we want to
> do something like above.  I will look into this a little bit more.

There are some problems in this approach.  The biggest one is the way
of processing in `hs-hide-level-recursive'.  It basically repeats:

1. hs-hide-block-at-point
2. forward-comment
3. re-search-forward hs-block-start-regexp

Let me explain with the following code:

#+begin_src python
class C:
    def a():
        pass

    def b():
        pass
#+end_src

Let's assume that point is at the beginning of the line "def a():".
The first step hides the method "a" and moves the point to the end of the
block.  The second step moves the point to the position "d" of "def b():",
not the beginning of the line.  Therefore, if hs-block-start-regexp
contains rx line-start, it fails to match "def b():".

Although this can be solved by adding a flag or a mode specific
forward-comment function to hs-special-modes-alist, I'm not sure if it
is the right approach.

So I think the following setting is best at this moment.

#+begin_src elisp
  (add-to-list
   'hs-special-modes-alist
   `(python-mode
     ,(python-rx (seq (* (syntax ?-)) block-start))
     ;; Use the empty string as end regexp so it doesn't default to
     ;; "\\s)".  This way parens at end of defun are properly hidden.
     ""
     "#"
     python-hideshow-forward-sexp-function
     nil))
#+end_src

(* (syntax ?-)) corresponds to "\\s-*" of the current regexp.  It
allows the point to be located before the block start symbols.

Another idea to improve the behavior of hiding a block is to provide a
wrapper function of `hs-hide-block' as follows and remap the key
bindings to this function like `python-mark-defun'.  With this
approach, C-c @ C-d and C-c @ C-h work as expected with the example
shown in the first mail.

#+begin_src elisp
(defun python-hideshow-hide-block (&optional end)
  "Python specific `hs-hide-block' function for `hs-minor-mode'.
Argument END is passed to `hs-hide-block'."
  (interactive "P")
  (python-nav-beginning-of-block)
  (hs-hide-block end))
#+end_src

Attached is a PoC level patch of this approach.  What do you think of
this approach?

Regards,