Re: [PATCH] allow for multiline headers

[Nicolas Goaziou]

Hello,

Mario Frasca <mario@anche.no> writes:

> On 13/06/2020 17:18, Nicolas Goaziou wrote:

>> Unit tests are not worth a formal definition. However, "test-ox.el"
>> contains unit tests.
>
> I'm not sure what you mean by the first sentence

I mean that, even though unit tests are great, and certainly welcome, we
first and foremost need a clear definition of a table header. This
concept is suggested in the manual, and more accurately defined in
ox.el. It would be nice to define it properly in the syntax.

> but I found the header tests in the test-ox.el file, and added one
> where multiple lines header is accepted.  is it related enough as to
> be included in my proposal?

Sure.

> I can move the org-table-collapse-header function from org-table.el to
> org-plot.el, but to me it makes little sense, relegating a generic
> function to a sub-module: others will look for the functionality in
> org-table, not see it, and duplicate the function somewhere else.

This doesn't seem to be a generic function, but a very specific one.
More on this below.

> +  (setq glue (or glue " "))
> +  (setq max-header-lines (or max-header-lines 4))

Please use `let' instead of `setq' whenever possible, e.g., here.

> +  (while (equal 'hline (car table))

equal -> eq

> +    (setq table (cdr table)))

(pop table)

> +  (let* ((trailer table)
> +      (header-lines (cl-loop for line in table
> +                             until (equal line 'hline)
> +                             collect line
> +                             do (setq trailer (cdr trailer)))))

You couldn't resist. 

This could be extracted as an independent function, which would return
the header, or nil. We can also imagine a function returning a cons cell
(HEADER . BODY), both HEADER and BODY being list of rows (possibly
empty).

Note that the function may be more complicated than the above, because,
in the following tables

    | a |
    | b |

or

    |---|
    | a |
    | b |
    |---|

there's a body, but no header. It depends on the definition chosen for
headers.

> +    (if (and trailer (<= (length header-lines) max-header-lines))
> +     (cons (apply #'cl-mapcar
> +                  #'(lambda (&rest x)

The #', aka, `function', is just noise before `lambda'.

   --> (lambda (&rest x))

> +                      (org-trim
> +                       (mapconcat #'identity x glue)))
> +                  header-lines)
> +           trailer)
> +      table)))

Here you are going further than setting the definition for tables
headers. You imply that:

    | header line 1 |
    | header line 2 |
    |---------------|
    | body          |

is equivalent to

    | header line 1 header line 2 |
    |-----------------------------|
    | body                        |

I'm not so sure this is a good idea. It might be a good idea for Org
Plot, I don't know, but generally speaking, is it?

Regards,
-- 
Nicolas Goaziou