This reverts commit 1cca5142af.
This appears to be causing intermittent triple-faults and I don't know
why yet, so I'll just revert it to keep the tree in decent shape.
Background: DoubleBuffer is a handy buffer class in the kernel that
allows you to keep writing to it from the "outside" while the "inside"
reads from it. It's used for things like LocalSocket and PTY's.
Internally, it has a read buffer and a write buffer, but the two will
swap places when the read buffer is exhausted (by reading from it.)
Before this patch, it was internally implemented as two Vector<u8>
that we would swap between when the reader side had exhausted the data
in the read buffer. Now instead we preallocate a large KBuffer (64KB*2)
on DoubleBuffer construction and use that throughout its lifetime.
This removes all the kmalloc heap traffic caused by DoubleBuffers :^)
Some inline stylesheets use HTML comments like "<!--blah blah-->".
The HTML parser currently generates a comment child node of the <style>
element whenever this happens, and we don't want the comment itself to
be interpreted as part of the stylesheet.
This is a total hack to get around the auto-detection mechanism for
whether a block has inline or block children. We'll say that tables
never have inline children for now, and then anything that actually
turns out to be an inline child will just be ignored by layout.
Instead of computing whether a block's children are inline based on the
first child, make it an imperatively-set flag.
This gives us some flexibility to ignore things like text nodes inside
a <table>, for example. I'm still unsure what the "correct" way to deal
with those will be. We'll find out sooner or later. :^)
This class introduces LayoutTable, LayoutTableRow and LayoutTableCell.
These are produced by "display" values table, table-row and table-cell
respectively.
Note that there's no layout happening yet, I'm just adding the classes.
When playing an ABuffer, the count of samples were determined by the
size of the SharedBuffer. This caused small pauses of up to 512
samples during the playback, when the size of the shared buffer was
rounded up to a multiple of 4096. This problem was amplified by the
fact that the AResampleHelper was created every time a new chunk of
audio was to be processed, causing inconsistencies in the playback of
wav files.
We now wait until the pixels are actually needed before fully decoding
images in <img> elements.
This needs some more work and is currently a bit memory-wasteful since
we'll hang on to the raw image data forever.
An ImageLoader is a generic interface for loading encoded image data of
any supported format. It has an ImageLoaderPlugin internally that does
all the work.
This patch adds an initial PNGImageLoaderPlugin that knows how to
retrieve the size of a PNG, and the bitmap. The API is divided into
size() and bitmap() to facilitate geometry-only decoding.
This will be useful in places like LibHTML where we need dimensions for
layout purposes but can wait with the bitmap until later.
If a LayoutNode is split into line box fragments, we need to walk our
fragments and invalidate them. It was not enough to do this only for
LayoutBox nodes.
To streamline the layout tree and remove irrelevant data from classes
that don't need it, this patch adds two new LayoutNode subclasses.
LayoutNodeWithStyleAndBoxModelMetrics should be inherited by any layout
node that cares about box model metrics (margin, border, and padding.)
LayoutBox should be inherited by any layout node that can have a rect.
This makes LayoutText significantly smaller (from 140 to 40 bytes) and
clarifies a lot of things about the layout tree.
I'm also adding next_sibling() and previous_sibling() overloads to
LayoutBlock that return a LayoutBlock*. This is okay since blocks only
ever have block siblings.
Do also note that the semantics of is<T> slightly change in this patch:
is<T>(nullptr) now returns true, to facilitate allowing to<T>(nullptr).
This is an implementation of syslog with some OpenBSD extensions.
There is no syslogd support (so it only logs to dbgprintf/stderr),
but otherwise is functional.
Many weird defines are always present, because some syslog users in
the wild check for their existence.
Minor patch, but I was watching one of your videos on YouTube and
thought that the pop-up was unecessary/annoying in this case. Love
your enthusiasm for the project :^)
This patch makes it possible to call Node::invalidate_style() and have
that node and all of its ancestors recompute their style.
We then figure out if the new style is visually different from the old
style, and if so do a paint invalidation with set_needs_display().
Note that the "are they visually different" code is very incomplete!
Use this to make hover effects a lot more efficient. They no longer
cause a full relayout+repaint, but only a style invalidation.
Style invalidations are still quite heavy though, and there's a lot of
room for improvement there. :^)
