On low-DPI screens, the rounding order of operations made it impossible
to produce an odd-pixel spacing. Specifying 1, for instance, produced
zero space between panes.
This approach subtracts half the spacing from the first pane prior to
rounding and uses the whole spacing for the second pane size and
coordinate.
This patch documents the fact that a couple of the image
handle constructors know how to guess the image format based
on the data that they are provided.
I had to dig through `iced` sources until I discovered that those
routines ultimately boil down to stuff like [image::load_from_memory][1]
from the `image` crate, so I thought I would save others the trouble
of doing the same reverse-engineering
[1]: https://docs.rs/image/0.23.4/image/fn.load_from_memory.html
`winit` seems to produce `CursorEntered` and `CursorLeft` events after
wheel scrolling for no apparent reason (?). This causes annoying flickering when hovering some widgets on a scrollable.
I should investigate this further. We are rolling back the fix for the
time being.
For now, we just set the cursor position to some negative coordinates.
This is a temporary hack until we are able to encode cursor
availability. Layers and/or multi-window support will make this
apparent.
At this stage, it is important to allow the library to change rapidly.
Because of this, keeping a log of changes can be counter-productive. We
do not want pin down design decisions by writing detailed changelogs
(sunk cost fallacy).
Once the library and its different parts start becoming more stable and
mature, we will reintroduce changelogs accordingly.
For now, we will keep a main `CHANGELOG` file just for the `iced` crate.
