Constructing the HashMap in DeclarativeEnvironment was by far the most
expensive thing when making JavaScript function calls.
As it turns out, we don't really need this to be a HashMap in the first
place, as lookups are cached (by EnvironmentCoordinate) after the first
access, so after that we were not even looking in the HashMap, going
directly to the bindings Vector instead.
This reduces function_declaration_instantiation() from 16% to 9% when
idling in "Biolab Disaster". It also reduces has_binding() from 3% to
1% on the same content.
With these changes, we now actually get to idle a little bit between
game frames on my machine. :^)
Note that only the first test actually functions currently.
Single-number ratios instead get parsed as a `<number>`, and will do
until the parser gets smarter. (The alternative, where all
single-numbers get parsed as `<ratio>`, does make the tests succeed,
but numbers are more common than ratios so I have given numbers
preference for now.)
Also simplified the styling and text a bit. Now, red = fail, green =
success. No more "unstyled = fail" stuff.
As noted, the Parser can't handle the `<number>` syntax for this - it
gets parsed instead by the `<number>` branch. We can't actually resolve
the ambiguity without making the Parser aware of what type each
media-feature is, but I will get to that soon. :^)
This replaces the fixed point subpixel precision logic.
GLQuake now effectively renders artifact-free. Previously white/gray
pixels would sometimes be visible at triangle edges, caused by slightly
misaligned triangle edges as a result of converting the vertex window
coordinates to `int`. These artifacts were reduced by the introduction
of subpixel precision, but not completely eliminated.
Some interesting changes in this commit:
* Applying the top-left rule for our counter-clockwise vertices is now
done with simpler conditions: every vertex that has a Y coordinate
lower than or equal to the previous vertex' Y coordinate is counted
as a top or left edge. A float epsilon is used to emulate a switch
between `> 0` and `>= 0` comparisons.
* Fog depth calculation into a `f32x4` is now done once per triangle
instead of once per fragment, and only if fog is enabled.
* The `one_over_area` value was previously calculated as `1.0f / area`,
where `area` was an `int`. This resulted in a lower quality
reciprocal value whereas we can now retain floating point precision.
The effect of this can be seen in Tux Racer, where the ice reflection
is noticeably smoother.
Incorrect is in quotes because the spec (both 1.7 and 2.0) specify this
multiplication as it was originally! However, flipping the order of
operations here makes the text in all of my test cases render in the
correct position.
The CTM is a transformation matrix between the text coordinate system
and the device coordinate system. However, being on the right-side of
the multiplication means that the CTM scale parameters don't have any
influence on the translation component of the left-side matrix. This
oddity is what originally led to me just trying this change to see if
it worked.
Previously, text spacing on a page would only look correct on very
zoomed-in pages. When the page was zoomed out, the spacing between
characters was very large. The cause for this was incorrect initial
values for the Tc (character spacing) and Tw (word spacing) text
parameters. The initial values were too large, but they were only
about 3-5 pixels, which is why the error was only observable for
smaller pages.
The text placement still isn't perfect, but it is _much_ better!
If the utilization of a HashTable (size vs capacity) goes below 20%,
we'll now shrink the table down to capacity = (size * 2).
This fixes an issue where tables would grow infinitely when inserting
and removing keys repeatedly. Basically, we would accumulate deleted
buckets with nothing reclaiming them, and eventually deciding that we
needed to grow the table (because we grow if used+deleted > limit!)
I found this because HashTable iteration was taking a suspicious amount
of time in Core::EventLoop::get_next_timer_expiration(). Turns out the
timer table kept growing in capacity over time. That made iteration
slower and slower since HashTable iterators visit every bucket.
Just walk the table from start to finish, deleting buckets as we go.
This removes the need for remove() to return an iterator, which is
preventing me from implementing hash table auto-shrinking.
This patch adds support for "crisp-edges", "high-quality" and "smooth"
for the CSS image-rendering property.
"crisp-edges" maps to nearest-neighbor scaling for <canvas> and <img>
elements, while "high-quality" and "smooth" both use bilinear blending.
Previously, the decoration was painted behind the text. This probably
wasn't noticed before, as we didn't compute `text-decoration-color`
values yet and the decoration had the same color anyway.
Previosly, we used only the text color as a line decoration color.
The FIXME comment has been directly copy-pasted from the border color
note a few lines below.
The spec says "fire an event named resize at the Window object
associated with doc."
However, we were accidentally firing it at `doc` instead of the Window.
Shell can now use LibLine's `on_paste` hook to more intelligently escape
pasted data, with the following heuristics:
- If the current command is invalid, just pile the pasted string on top
- If the cursor is *after* a command node, escape the pasted data,
whichever way yields a smaller encoding
- If the cursor is at the start of or in the middle of a command name,
paste the data as-is, assuming that the user wants to paste code
- If the cursor is otherwise in some argument, escape the pasted data
according to which kind of string the cursor is in the middle of
(double-quoted, single-quoted or a simple bareword)
