I'm not good with a mouse, so I found it hard to edit animation lines.
The mouse position has to be quite precise, with only a small threshold.
Looking at the code, AutomationLine sets the threshold for the PolyLine
to 4.0 . That seems to be a distance, and better for me than what I
experience.
The actual code in PolyLine is however comparing it directly to the
squared distance, making it more sensitive than expected.
Fixed by computing the squared threshold - also including the line
width.
This fixes the following problem:
When automation lines with significant change are zoomed in time, the
slope gets smaller (towards horizontal) as the control points moves
further away from the visible area. That was rendered correctly, but the
corresponding mouse events happened as if the line had a steeper slope.
The problem was caused by the X value being clamped to the visible area,
without scaling the Y value correspondingly. It has apparently been like
that for a decode, since introduced in c4f0063a68.
The problem is fixed by introducing a clamp2 function that scales the Y
value if clamping the X value.
Note: An old comment says that math goes wrong unless clamping below
COORD_MAX. It is not clear to me what math it refers to, and especially
why we don't need similar clamping on the lower bounds.
And while rarely a real problem, I guess it would be more correct and a
slight optimization to skip all lines where both ends are outside the
same bound. In theory, as it is now, the mouse could catch an invisible
line close to the border.
This fixes automation lanes bleeding into the ruler area.
The bounding box of all items in the main canvas group starts
at -0.5, and the cursor-scroll-group at -1.5. This is calculated
to include line-width, and outlines outside the item(s).
A scroll-group however must not extend its render area to
render those.
If queue_draw is "frozen", we simply accumulate drawing
requests in a (union) rectangle, and when finally "thawed"
the canvas submits a single redraw request for the entire
accumulated rect.
Although in theory this is all that GTK/GDK does for
draw requests, callgrind reveals significant costs
associated with the actual calltree for GtkWidget::queue_draw_area().
One potential cost is that GDK also maintains a list of
invalidated rectangles in addition to the union, and
for MIDI regions with thousands of notes, this can represent
real overhead. This approach dispenses with the rect list,
since our Canvas drawing model only uses the union rectangle
anyway.
The tooltip should remain visible until the mouse leaves
the item's bounding box. Also do not start the tooltip
timeout if the item does not have a tooltip.
On Intel systems ArdourCanvas::COORD_MAX (1.7e+307) was rounded
to (gint) -2147483648. gtk+ treats negative window size-requests
as 1px.
However on ARM, COORD_MAX was truncated to +2147483648, gtk+ limits
this to 65535. Most WM/Xwin systems cannot handle windows this large.
It also exceeds the max size of cairo [image] surfaces.
This issue was introduced in a1c67b4ad7
when "natural_size" was removed. Before that change infinitely large
canvas had a natural_size of 2x2 px.
The AudioClipEditor features a scroll bar that is a part of the canvas. Because scroll
groups are at the top level of a canvas, the scroll bar is necessary within a scroll
group, which causes it to get confused about the difference between its own
position within the canvas and that of the scroll group. This commit introduces
a per-Item flag, _scroll_translation, which is true by default. If false, the
item will not have coordinates translated to reflect scroll group position.
We need to be able to construct Metrics after the ruler is constructed
in order to deal with Triggers which are potentially short-lived. We
stored the reference as a pointer anyway, so it's not as much of a change
as it appears.
If there's a grabbed item (GtkCanvas only at present) then unless it belongs to the scroll
group used for scroll offset translation, the event coordinates should not be translated,
even if the mouse pointer moves into the scroll group.
This was changed in fc9840f2a9 to hard code a "false" argument describing
whether or not the child bbox had changed. This causes the parent to not
recompute its own bounding box when a child is shown/hidden, which is
incorrect. We now force it to true, which may be overkill for some purposes,
but at worst will just force an unnecessary bbox re-computation and nothing
more.
