Fixes#1416
At a high level, the issue is that when mouse capture is enabled (i.e. in
neovim), "shift" escapes the capture. So "cmd+shift" is equal to "cmd"
which doesn't get sent to the TUI program and so on. For link
highlighting which now requires "cmd" (super) is held, we were sending
"cmd+shift" to the renderer so we weren't checking for links.
So the core of this commit is respecting this scenario and stripping the
shift modifier.
This commit also found that when the mouse wasn't over a link, we were
always checking and highlighting links on line one of the visible
screen. This bug is fixed and should also result in a very slight
performance improvement on rendering in all cases.
Up to this point, every font I've experienced with ligatures has
replaced the codepoints that were replaced for combining with a space.
For example, if a font has a ligature for "!=" to turn it into a glyph,
it'd shape to `[not equal glyph, space]`, so it'd still take up two
cells, allowing us to style both.
Monaspace, however, does not do this. It turns "!=" into `[not equal
glyph]` so styles like backgrounds, underlines, etc. were not extending.
This commit detects multi-cell glyphs and inserts synthetic blank cells
so that styling returns. I decided to do this via synthetic blank cells
instead of introducing a `cell_width` to the shaper result because this
simplifies the renderers to assume each shaper cell is one cell. We can
change this later if we need to.
Annoyingly, this does make the shaper slightly slower for EVERYONE to
accomodate one known font that behaves this way. I haven't benchmarked
it but my belief is that the performance impact will be negligible
because to figure out cell width we're only accessing subsequent cells
so they're likely to be in the CPU cache and also 99% of cells are going
to be width 1.
Rather than immediately converting a color palette index into an RGB
value for a cell color, when a palette color is used track the palette
color directly in the cell state and convert to an RGB value in the
renderer.
This causes palette color changes to take effect immediately instead of
only for newly drawn cells.
Fixes#1037
Renderers must convert the internal Kitty graphics state to a GPU
texture for rendering. For performance reasons, renderers cache the GPU
state by image ID. Unfortunately, this meant that if an image was
replaced with the same ID and was already cached, it would never be
updated on the GPU.
This PR adds the transmission time to the cache. If the transmission
time differs, we assume the image changed and replace the image.
Fixes#882
We previously had a hardcoded limit of 16 codepoints. In #882, a user
pointed out that in Chinese, is reasonable for a preedit input to be
longer than this.
To avoid any future issues, this commit moves to a heap-allocated
variant. Preedits aren't that common, they aren't high throughput, and
they're generally pretty small, so using a heap allocation is fine. The
memory is owned by the person who set it.
Fixes#895
Every loaded font face calculates metrics for itself. One of the
important metrics is the baseline to "sit" the glyph on top of. Prior to
this commit, each rasterized glyph would sit on its own calculated
baseline. However, this leads to off-center rendering when the font
being rasterized isn't the font that defines the terminal grid.
This commit passes in the font metrics for the font defining the
terminal grid to all font rasterization requests. This can then be used
by non-primary fonts to sit the glyph according to the primary grid.
This uses WCAG2 algorithms with a minimum ratio hardcoded of 3:1. This
is not shippable in its current state because we want the ratio to be
configurable and I'm not happy with the way data is being sent to the
shader.
