Opaque lines are turning out to be a total pain, plus they ruin absolute
positioning on the lines div. The slight speed boost isn't seeming worth
it anymore.
Because lines are opaque on the GPU for sub pixel antialiasing, the
lines obscure the selections which were formerly rendered behind the
lines. This commit renders selection fragments *on* each opaque line
layer so the selections look correct again. Still needs cleanup and
optimization.
Since lines and line numbers are now on the GPU, their text won't be
properly anti-aliased on low-resolution displays unless their layers
have a solid background.
This commit breaks the initial render of the editor component into two
stages.
The first stage just renders the shell of the editor so the height,
width, line height, and default character width can be measured. Nothing
that depends on these values is rendered on the first render pass.
Once the editor component is mounted, all these values are measured and
we force another update, which fills in the lines, line numbers,
selections, etc.
We also refrain from assigning an explicit height and width on the
model if these values aren't explicitly styled in the DOM, and just
assume the editor will stretch to accommodate its contents.
When the target of a mousewheel event is removed, it breaks velocity
scrolling.
Previously, we were preserving the entire screen range when scrolling
with the mouse wheel, which caused a lot of DOM nodes to accumulate. Now
we only preserve the individual line and line number associated with the
target of the mousewheel event, moving them just off screen below all
the on-screen lines and line numbers. This keeps the number of DOM nodes
limited while retaining velocity effects.
We accumulate pending changes and pass them to the lines and the gutter
to help them determine whether to update. The lines only update if the
visible row range changed or if there was a change in the visible row
range.
