Re: Coordinates and Windows

[martin rudalics <rudalics@gmx.at>]

 > I don't see why.  Having the top-left corner of the frame as the
 > reference point sounds simpler and more clear to me, and allows one to
 > use it in the same way in all the builds.  No?

If by top-left corner you intend the top-left position of the menu bar,
or if that's absent that of the top/left tool bar, or if that's absent
too that of the internal border, or if that is absent too that of the
root window, I agree.  Just that it's not all too trivial to calculate
it.

We know the size of the tool bar only after we have displayed it.  This
is already a problem on Lucid, Motif, Windows.  And Windows has certain
difficulties telling us the height of the menubar when it wraps.  This
too is already a problem, so nothing new here.  But there might be yet
unknown problems with Gtk and NS.

 >> Actually, I started to look into this when I detected that it's
 >> virtually impossible to position a tooltip frame via `x-show-tip' at a
 >> window position obtained via `pos-visible-in-window-p' consistently on
 >> all platforms.
 >>
 >> To illustrate how little care has been applied to this so far, consider
 >> the function `window-absolute-pixel-edges'.  With a maximized frame on
 >> Windows XP this function returns (-4 -4 1676 964) here.
 >
 > I get (-8 28 1912 984) on Windows 7 and (-4 32 1916 1022) on XP, so at
 > least the Y coordinate seems OK.

Just 32 pixels for title, menu, and tool bar?  Here `x-frame-geometry'
tells me that they alone take 74 pixels.  BTW, has Windows 7 an 8 pixel
wide border by default?

 >> Apparently, this was designed for Gtk users, never really tested on
 >> other platforms.  But this also implies that changing things for Gtk
 >> might not be a good idea.
 >
 > I don't follow the logic, can you elaborate why this implies what you
 > think it implies?  Are you saying that on GTK a maximized frame has
 > some of its GTK decorations off-screen?

No, off-screen decorations are a Windows feature.  What I meant is that
Gtk has some established, working features that albeit never worked on
Windows.  And I'm afraid to break these features.

 >> Because most programmers think (and applications work) in terms of rows
 >> and columns.  `window-edges' is still quite in use although it's bound
 >> to provide inaccurate results in many occasions now.
 >
 > I'm afraid I still don't see why this requires rounding, especially if
 > the results are already inaccurate.  E.g., can we consider the menu
 > bar a single line, and the tool bar another N lines (N is usually 1,
 > but doesn't have to be).  IOW, can we treat each "row" of display in
 > these windows as a single line, for the purposes of counting rows?  If
 > not, why not?

The windmove code, for example, determined adjacency of windows from the
fact that their respective edges matched.  It took me some time to
adjust the line/column calculation code to handle that in an artificial
way.  I don't know how much code exists that uses similar heuristics.

martin