Re: bug#9782: 24.0.90; move-to-window-line not taking header line into account

[Eli Zaretskii]

> Date: Mon, 06 May 2013 07:38:43 +0400
> From: Dmitry Gutov <address@hidden>
> CC: address@hidden
> 
> >> But it's a seemingly reasonable piece of code, it worked, and your
> >> change made it break in some peculiar special case important to some
> >> users.
> >
> > That's not true.  It was not a peculiar special case. The function did
> > not behave according to its contract, and my change fixed that.
> 
> Comparing the descriptions of `move-to-window-line' and 
> `posn-actual-col-row', it doesn't seem obvious that one considers the 
> header line part of the window, and another doesn't. The words "row" and 
> "line" are often synonymous in English.

Perhaps we need to improve the documentation of the posn-* family of
functions, then.  It is important to keep in mind that this family
serves mainly the mouse events, so its coordinate system is
window-relative.  The use of columns and rows there is just to measure
in character units, rather than in pixels.  But it is still
window-relative and agnostic to the header line, like the mouse is.

> `company-pseudo-tooltip-show-at-point' passes the current row and column 
> to `company-pseudo-tooltip-show', which collects are required lines and 
> prepares the `before-string' for the overlay.
> 
> These two would be the easiest to convert to direct point manipulation, 
> except for the detail that they calculate, use and store the beginning 
> column in the overlay property, for use later.
> 
> `company--pseudo-tooltip-height' calculates the appropriate rectangle 
> height, based on the current row and the inner height of the window.
> 
> `company--electric-do' (somewhat unrelated) calls `recenter' using the 
> value returned by `posn-actual-col-row'.
> 
> `company-select-mouse' and `company--inside-tooltip-p' handle mouse 
> interaction with the rectangle, comparing the coordinates of the mouse 
> click with the coordinates and dimensions of the rectangle (some of them 
> saved previously; the current row is again retrieved using 
> `posn-actual-col-row').

I don't see company--inside-tooltip-p anywhere in today's ELPA company
package, but anyway, why do you need to use coordinates to determine
whether the mouse clicked on the pseudo-tooltip?  The pseudo-tooltip
is a display or overlay string, right?  So the OBJECT part of the
event's POSITION member will tell you whether you clicked on the
string or not: if it's nil, the click is on some buffer text, but if
it's the display string, the click was on that string.

> 1) Using `move-to-window-line', `vertical-motion' and a counter.

Actually, I think you want something like

  (count-screen-lines (window-start) (point))

And for the mouse, see the suggestion above.