Currently, we use UnsafeFlow and UnsafeLayoutNode, both of which are aliases
for (usize, usize) and thus interconvertible. This change should make it
clearer that the WorkQueue is not limited to one particular type.
those instead of working on nodes one-by-one.
This reduces the overhead of the work-stealing traversal function
significantly. It's especially important on ARM, where memory barriers are
expensive.
A rebuild after touching components/profile/mem.rs now takes 48 seconds (and
only rebuilds `profile` and `servo`) which is much lower than it used to be.
In comparison, a rebuild after touching components/profile_traits/mem.rs takes
294 seconds and rebuilds many more crates.
This change also removes some unnecessary crate dependencies in `net` and
`net_traits`.
- Most of util::memory has been moved into profile::mem, though the
`SizeOf` trait and related things remain in util::memory. The
`SystemMemoryReporter` code is now in a submodule
profile::mem::system_reporter.
- util::time has been moved entirely into profile::time.
This makes these parameters self-documenting.
This patch does not attempt to push those enums into the data
structures that feed calls to this function.
Fixes#4158.
Now that DOM/Flow traversals have been refactored out, the `recalc_style_for_subtree`
function in `css/matching.rs` can be removed, in lieu of just running the standard
`recalc_style_for_node` and `construct_flows` traversals sequentially. Now we
no longer have the maintenance headache of duplicating selector matching logic
in two places! \o/
r? @pcwalton
a separate `ClipDisplayItem`.
We push down clipping areas during absolute position calculation. This
makes display items into a flat list, improving cache locality. It
dramatically simplifies the code all around.
Because we need to push down clip rects even for absolutely-positioned
children of non-absolutely-positioned flows, this patch alters the
parallel traversal to compute absolute positions for
absolutely-positioned children at the same time it computes absolute
positions for other children. This doesn't seem to break anything either
in theory (since the overall order remains correct) or in practice. It
simplifies the parallel traversal code quite a bit.
See the relevant Gecko bug:
https://bugzilla.mozilla.org/show_bug.cgi?id=615734
stretching.
This preserves the usage of the Bloom filter throughout style recalc,
but the implementation is rewritten. Provides a 15% improvement on
Guardians of the Galaxy.
DOM traversals and Flow traversals look very similar. This patch unifies them
with the preorder/postorder pattern. Hopefully, it also opens the door for writing
the traversal code only once, instead of the duplication we have today.
This patch puts in the initial framework for incremental reflow. Nodes' styles
are no longer recalculated unless the node has changed.
I've been hacking on the general problem of incremental reflow for the past
couple weeks, and I've yet to get a full implementation that actually passes all
the reftests + wikipedia + cnn. Therefore, I'm going to try to land the different
parts of it one by one.
This patch only does incremental style recalc, without incremental flow
construction, inline-size bubbling, reflow, or display lists. Those will be coming
in that order as I finish them.
At least with this strategy, I can land a working version of incremental reflow,
even if not yet complete.
r? @pcwalton
absolutely-positioned elements declared with `display: inline`.
Although the computed `display` property of elements with `position:
absolute` is `block`, `position: absolute; display: inline` can still
behave differently from `position: absolute; display: block`. This is
because the hypothetical box for `position: absolute` can be at the
position it would have been if it had `display: inline`. CSS 2.1 §
10.3.7 describes this case in a parenthetical:
"The static-position containing block is the containing block of a
hypothetical box that would have been the first box of the element if
its specified 'position' value had been 'static' and its specified
'float' had been 'none'. (Note that due to the rules in section 9.7 this
hypothetical calculation might require also assuming a different
computed value for 'display'.)"
To handle this, I had to change both style computation and layout. For
the former, I added an internal property
`-servo-display-for-hypothetical-box`, which stores the `display` value
supplied by the author, before the computed value is calculated. Flow
construction now uses this value.
As for layout, implementing the proper behavior is tricky because the
position of an inline fragment in the inline direction cannot be
determined until height assignment, which is a parallelism hazard
because in parallel layout widths are computed before heights. However,
in this particular case we can avoid the parallelism hazard because the
inline direction of a hypothetical box only affects the layout if an
absolutely-positioned element is unconstrained in the inline direction.
Therefore, we can just lay out such absolutely-positioned elements with
a bogus inline position and fix it up once the true inline position of
the hypothetical box is computed. The name for this fix-up process is
"late computation of inline position" (and the corresponding fix-up for
the block position is called "late computation of block position").
This improves the header on /r/rust.
