Re: [AUCTeX-devel] Extending `LaTeX-array-count-columns'

[Arash Esbati]

Hi Ikumi,

Ikumi Keita <address@hidden> writes:

>> 2 thing occurred to me when hitting `M-RET' in tabular environments:
>> Current code in `LaTeX-array-count-columns' cannot handle multi column
>> specs like `*{num}{spec}' and optional args to column specs S and s from
>> siunitx.el.  I have a patch to handle both issues.  Can people please
>> test this code against their tabular's and see if it does the right
>> thing?
>
> I found two cases that the code does not work as expected.
> (1) When the second argument of *-operator contains more than 1
> columns.
>
> \begin{tabular}[t]{*{3}{|cr}l}
>   1 & 2 & 3 & 4 & 5 & 6 & 7 \\
>     &&&
> \end{tabular}
>
> (2) When *-operator is nested.
> \begin{tabular}[t]{*{3}{c*{2}{r}}l}
>   1 & 2 & 3 & 4 & 5 & 6 & 7 & 8 & 9 & 0 \\
>     &&&
> \end{tabular}

Thanks for looking at this.  Indeed, I totally ignored these cases!  I
admit that I've never seen case (2), but it works and your code supports
it.

> Apparently recursive call of LaTeX-array-count-columns is needed do the
> task.  How about this one?
>
> (defun LaTeX-array-count-columns (start end)
>   "Count number of ampersands to be inserted.
> The columns are specified by the letters found in the string
> `LaTeX-array-column-letters' and the number of those letters within the
> text between START and END is basically considered to be the number of
> columns.  The arguments surrounded between braces such as p{30pt} do not
> interfere the count of columns.
>
> Return one less number than the columns, or nil on failing to count the
> right number."
>   (save-excursion
>     (let (p (cols 0))
>       (goto-char start)
>       (while (< (setq p (point)) end)
>
>       ;; The below block accounts for one unit of move for
>       ;; one column.
>         (setq cols (+ cols
>                     ;; treat *-operator specially.
>                       (if (eq (following-char) ?*)
>                         ;; *-operator is there.
>                           (progn
>                           ;; pick up repetition number and count
>                           ;; how many columns are repeated.
>                             (re-search-forward
>                              "\\*[ \t\r\n%]*{[ \t\r\n%]*\\([0-9]+\\)[ 
> \t\r\n%]*}" end)
>                           (let ((n (string-to-number
>                                       (match-string-no-properties 1)))
>                                 ;; get start and end of repeated spec.
>                                 (s (progn (down-list 1) (point)))
>                                 (e (progn (up-list 1) (1- (point)))))
>                             (* n (1+ (LaTeX-array-count-columns s e)))))
>                       ;; not *-operator.
>                         (skip-chars-forward
>                          LaTeX-array-column-letters end))))
>       (skip-chars-forward (concat
>                            "^" LaTeX-array-column-letters "*"
>                            TeX-grop LaTeX-optop) end)
>         (when (or (eq (following-char) ?\{)
>                   (eq (following-char) ?\[))
>           (forward-list 1))
>
>       ;; Not sure whether this is really necessary or not, but
>       ;; prepare for possible infinite loop anyway.
>       (when (eq p (point))
>         (setq cols nil)
>         (goto-char end)))
>       ;; The number of ampersands is one less than column.
>       (if cols (1- cols)))))
>
> The above version also changes the construct
>       (if (looking-at-p "\\*[ \t\r\n%]*{[ \t\r\n%0-9]+}")
>           (progn
>             (re-search-forward
>              "\\*[ \t\r\n%]*{[ \t\r\n%]*\\([0-9]+\\)[ \t\r\n%]*}" end t)
> to
>       (if (eq (following-char) ?*)
>           (progn
>             (re-search-forward
>              "\\*[ \t\r\n%]*{[ \t\r\n%]*\\([0-9]+\\)[ \t\r\n%]*}" end)
> based on the following two reasons.
> [a] Simple check of `the next char is * or not?' seems sufficient for
> the purpose.  (In addition, looking-at-p is not available in older
> emacsen.)
> [b] The noerror argument of re-search-forward is dropped.  If that
> re-search-forward fails, the string-to-number on the next line does not
> make sense.  Even if the error is raised here, it is captured by
> ignore-erros in LaTeX-insert-ampersands so the users are not bothered by
> the error.

Thanks!  I'm fine with your suggestions.  I have already pushed some
patches to git for other styles.  Once we have this function sorted out,
I would add some additional tests (see my other message to Mosè) and
update latex.el.  Then `M-RET' should work for all tabular env's.

Best, Arash