We're seeing occasional fuzzy-mismatch failures with a single pixel differing,
where an antialiased "a" serif overlaps an adjacent table border. Hopefully a
sans-serif font will reduce the likelihood of this fuzziness.
Differential Revision: https://phabricator.services.mozilla.com/D35266
--HG--
extra : moz-landing-system : lando
This patch implements the majority of the planned picture caching
improvements. It supports most of the functionality required to
(as a follow up) support OS compositor integration. It also improves
on the robustness and functionality of the previous picture caching
implementation.
There are some expected temporary performance regressions in
some cases (such as content that is constantly invalidating) and
during initial page render when many render targets must be drawn
to. These performance regressions will be resolved in follow up
commits by supporting multi-resolution tiles.
The scene is split into a number of slices, determined by the scroll
root of each primitive, which can be found by the primitive's
spatial node indices. If a scene contains too many slices, then
picture caching is disabled on the page, to avoid excessive texture
memory usage, and rendering falls back to rasterizing each frame.
The specific changes in this patch are:
* Support tile caches for multiple scroll roots, allowing the
entire page (including fixed divs and the main UI bar) to be
cached in most cases, in addition to the main content.
* Remove requirement to read tiles back from the framebuffer.
Instead, they are drawn into the picture cache target tiles,
and blitted to the screen. This is slightly slower than the
existing picture caching when content is constantly changing,
however this cost will disappear / become irrelevant when
the OS compositor integration work is complete.
* Switch picture cache render targets to be nearest sampled (they
are always rendered 1:1) and support depth buffer targets.
* Make use of the external scroll offset support to allow removal
of the primitive correlation hacks in the previous picture
caching implementation. Also allows storing of primitive
dependencies in picture space rather than world space, which
reduces floating point inaccuracies.
* Determine if each tile and picture cache can be considered
opaque. This is used to determine whether subpixel AA text
rendering is available on a slice, and for rendering optimizations
related to disabling blending and/or tile clears.
* Use the clip chain instance results from the recent visibility pass
work to determine clip chain dependencies. This results in fewer
clip item dependencies in tiles, which is faster to check validity
and reduces redundant invalidations.
* Remove extra overhead during batching related to batch lists,
and region iteration, as they are no longer required.
* Support PrimitiveVisibilityMask during batching. This allows a
single traversal of a picture (surface) root during batching to
efficiently construct multiple alpha batcher objects (typically
one per invalida tile).
* Picture caching is now handled implicitly by WR, depending on
the content of the scene. There is no requirement for client
code to manually select which stacking context should be cached.
* Simplify how clip chain / transform dependencies are tracked by
picture cache tiles.
* Support pushing / popping enclosing clip chain roots without
the need for a stacking context / picture in some cases. This
simplifies the logic to split the scene into multiple slices.
The main remaining work in this area is (a) extend the code to
optionally provide each slice as an input to the OS compositor
rather than drawing the tiles in WR, and (b) support multi-resolution
tiles so that we reduce the draw call, batching and render target
overhead in cases where much of the page content is changing.
Differential Revision: https://phabricator.services.mozilla.com/D34319
--HG--
extra : moz-landing-system : lando
Note that the geckoview annotation will override the previous Android
annotation, but I verified that on geckoview debug the test seems to
pass so overriding the Android&&debug annotation should be ok.
Depends on D34727
Differential Revision: https://phabricator.services.mozilla.com/D34728
--HG--
extra : moz-landing-system : lando
The fuzziness in the position-dynamic-changes reftests seems nondeterministic.
The fuzziness annotations in the previous patch were what I got after a few
iterations of do-a-try-push-and-update-annotations, but there are still more
failures showing up in subsequent try pushes. I visually checked all the
failures and they are all just fuzzy in different places, but intermittent.
This patch updates the fuzziness annotations on these tests to the maximum
that I encountered on any test, which is (2, 1382).
I'm keeping this as a separate patch because I think it might be valuable
in version control history to have the actual numbers seen on try which
are in the previous patch.
Depends on D34538
Differential Revision: https://phabricator.services.mozilla.com/D34539
--HG--
extra : moz-landing-system : lando
There are a number of failures, for which I've filed separate bugs.
And then a lot of fuzziness. I manually inspected the reftest analyzer
results on try pushes to distinguish failures vs fuzziness.
Depends on D34537
Differential Revision: https://phabricator.services.mozilla.com/D34538
--HG--
extra : moz-landing-system : lando
aDirtyRect in PaintSVG() is in outer <svg> coordinate. Image painting
API needs dirty rect in image coordinate. We need to transform it.
Differential Revision: https://phabricator.services.mozilla.com/D33813
--HG--
extra : source : d055d72f5e77f8ffd30331ed093800e87da8151a
aDirtyRect in PaintSVG() is in outer <svg> coordinate. Image painting
API needs dirty rect in image coordinate. We need to transform it.
Differential Revision: https://phabricator.services.mozilla.com/D33813
--HG--
extra : moz-landing-system : lando
(The pref is about to be removed, but even before its removal, it defaults to
'true' so these tests don't need to bother setting/checking it.)
Differential Revision: https://phabricator.services.mozilla.com/D33805
--HG--
extra : moz-landing-system : lando
The "-auto" and "-scroll" variants of this test used to be expected to mismatch
("!="), but our behavior changed with stylo such that they started matching, so
it was marked as "fails !=". But now we've decided that the post-stylo
matching behvavior is actually correct, so this patch is changing the manifest
line to "==" so that the expectation is that they match.
This is primarily to fix sizing of 'box-decoration-break:clone' inlines,
but also some 'slice' edge cases by recognizing more break opportunities,
and to improve sizing when BIDI-continuations are involved.
Differential Revision: https://phabricator.services.mozilla.com/D32915
--HG--
extra : rebase_source : 45a76e55e3eb5ac41fd60b11eb2acce24c6a1dd1
extra : amend_source : d8980854573522ddf761f0cf036aaa4ee0d96cac
The only different part is to resolve intrinsic image size. This patch
implements explicit requirements of the spec, but an edge case is tricky.
It's not clear per spec what the intrinsic image size is for an SVG
without explicit width/height, something like:
<svg>
<image href="data:image/svg+xml,<svg xmlns='http://www.w3.org/2000/svg'><rect width='40' height='90' fill='red' /></svg>"/>
</svg>
Chrome treats the intrinsic size of the href svg as the default size of
a replaced element (300x150), our image/VectorImage.cpp doesn't resolve
size in this case.
We can handle this particular case in some seperate bug if necessary, I think.
Differential Revision: https://phabricator.services.mozilla.com/D32415
--HG--
extra : moz-landing-system : lando
We previously (in bug 1491235) adjusted some utility code to make
layout-contained frames behave as if they have no baseline.
But that's not sufficient. To make frames fully report lack-of-a-baseline,
we now do the following for layout-contained frames, as of this patch:
(a) We now leave the ReflowOutput outparam's BlockStartAscent member at its
default value (which is what we do for frames without a baseline like
e.g. nsCheckboxRadioFrame and nsHTMLCanvasFrame). And if the parent cares
about the baseline, it'll then ask directly, using a baseline getter.
(b) We now return 'false' in more implementations of bool-returning
baseline-getter-methods (where 'false' indicates 'no baseline').
(c) We now return the margin-box-bottom edge, in the nscoord-returning
'GetLogicalBaseline()' getter method. (We typically do this by deferring
to the inherited method, which ultimately comes from nsFrame's
implementation). It's appropriate to use the margin-box-bottom edge when
there's no baseline, per the definition of 'vertical-align: baseline',
here: https://drafts.csswg.org/css2/visudet.html#propdef-vertical-align
Depends on D32182
Differential Revision: https://phabricator.services.mozilla.com/D32183
--HG--
extra : moz-landing-system : lando
In particular:
- In contain-layout-suppress-baseline-002.html, the test currently indirectly
relies on the 50px-tall-canvas being the tallest thing in each flex
container. This isn't robustly true (and in fact on windows, the textarea is
taller at 50.8px tall). So I'm adjusting this test so that it no longer has a
hardcoded flex container size and no longer depends on this.
- In contain-layout-baseline-005.html and its reference case, we need to
explicitly specify 'vertical-align:baseline' to test baseline-alignment,
because some of its tested form controls have other UA-stylesheet-provided
default values of 'vertical-align'.
(e.g. <select multiple> defaults to 'vertical-align:text-bottom")
- Also: in that same test, we need to reduce the width of the an <input>
textfield -- otherwise, it and the other elements on its line may not fit and
may linewrap, which prevents us from effectively testing baseline-alignment
on the linewrapped element.
- In contain-layout-button-001.html, the expectation was not correct. Before
this patch, the test expects that a layout-contained button will have the
same baseline as an empty button, and that's an invalid expectation. An empty
button uses a point inside of its content-box as its baseline, whereas a
layout-contained element *has no baseline*, which means that it does
'vertical-align:baseline' alignment by aligning its own margin-bottom edge
with the parent's baseline, per
https://drafts.csswg.org/css2/visudet.html#propdef-vertical-align
So, I'm amending the test to have this expectation and updating its meta tags
to reference the updated expectation and with a reference to that spec text.
Firefox fails the amended contain-layout-button-001.html test, so this patch
adds a .ini file to reflect that failure. The next patch in this series will
fix our implementation to make us pass the test, and will remove the .ini file.
Chrome also fails the amended contain-layout-button-001.html tests, and I filed
https://bugs.chromium.org/p/chromium/issues/detail?id=965740 on them with an
explanation.
Differential Revision: https://phabricator.services.mozilla.com/D32182
--HG--
extra : moz-landing-system : lando
We previously (in bug 1491235) adjusted some utility code to make
layout-contained frames behave as if they have no baseline.
But that's not sufficient. To make frames fully report lack-of-a-baseline,
we now do the following for layout-contained frames, as of this patch:
(a) We now leave the ReflowOutput outparam's BlockStartAscent member at its
default value (which is what we do for frames without a baseline like
e.g. nsCheckboxRadioFrame and nsHTMLCanvasFrame). And if the parent cares
about the baseline, it'll then ask directly, using a baseline getter.
(b) We now return 'false' in more implementations of bool-returning
baseline-getter-methods (where 'false' indicates 'no baseline').
(c) We now return the margin-box-bottom edge, in the nscoord-returning
'GetLogicalBaseline()' getter method. (We typically do this by deferring
to the inherited method, which ultimately comes from nsFrame's
implementation). It's appropriate to use the margin-box-bottom edge when
there's no baseline, per the definition of 'vertical-align: baseline',
here: https://drafts.csswg.org/css2/visudet.html#propdef-vertical-align
Depends on D32182
Differential Revision: https://phabricator.services.mozilla.com/D32183
--HG--
extra : moz-landing-system : lando
The relevant definition in the spec;
https://drafts.csswg.org/css-device-adapt/#min-scale-max-scale
Before this change, if both of initial-scale and maximum-scale are negative,
both values are clamped to 0.25. Whereas with this change, negative scale
values are treated as if it's not specified so that initial-scale value is
automatically calculated based on the layout viewport size.
negative-initial-and-maximum-scale.html is a test case for the case.
Also with this change, initial-scale values are going to be clamped to the
range [0.25, 10] during parsing it so that initial-scale-0.html and
initial-scale-100.html need to be modified, now the former is scaled by 0.25x,
the latter is scaled by 10x.
(Before this change, initial-scale=0 and initial-scale=100 were treated as
invalid scale values in nsViewportInfo::ConstrainViewportValues[1])
[1] https://searchfox.org/mozilla-central/rev/6c9f60f8cc064a1005cd8141ecd526578ae9da7a/dom/base/nsViewportInfo.cpp#15,19
Differential Revision: https://phabricator.services.mozilla.com/D32098
--HG--
extra : moz-landing-system : lando
This patch contains two isolated changes related to upcoming picture
caching improvements. Specifically:
* Determine the blit reason for stacking contexts with clips
earlier, during scene building. This simplifies the code and
allows better detection of redundant stacking contexts.
* Centralize the code for pushing batch instances into a small
number of places. This will simplify the switch to adding
a single primitive to multiple batch lists, in the case of
dirty regions with different batchers.
Differential Revision: https://phabricator.services.mozilla.com/D31898
--HG--
extra : moz-landing-system : lando
We don't store post-transform overflow areas for frames within preserve-3d, but we do store pre-transform overflow areas.
Rather than just recomputing the changed overflow for the root, we should recompute overflows for all ancestors up to the 3d root.
Differential Revision: https://phabricator.services.mozilla.com/D31213
--HG--
extra : moz-landing-system : lando
This is the main performance improvement, and means that we no longer have to iterate all the cells for each column.
It has a couple of behaviour changes:
The first is that we no longer apply stacking context effects (like opacity) to column and column group backgrounds.
I believe this is correct as per both CSS2.1 Appendix E, and css-tables-3 (quoted in nsTableColFrame::BuildDisplayList).
This matches the behaviour of blink and WebKit.
We also previously created items in column,row ordering, whereas now they will be in row,column. In cases where two cells
overlap (using rowspan and colspan to extend multiple neighbours in to the same place) this can render backgrounds in a
different order, but the new behaviour matches blink and WebKit.
Differential Revision: https://phabricator.services.mozilla.com/D29280
--HG--
extra : moz-landing-system : lando
This also changes behaviour a bit, previously we interleaved column and column group backgrounds. where we now put all the column group backgrounds behind all columns.
I believe this is the correct ordering as per CSS2.2 Appendix E.
Column backgrounds can overlap when using 'span', and we now render this in a different order, but this matches what other browsers do.
Differential Revision: https://phabricator.services.mozilla.com/D29278
--HG--
extra : moz-landing-system : lando