This removes nsStyleImageRequest by moving the load state to LoadData instead
(where other lazy state like the resolved URL and load id lives).
That way we can use cbindgen for more stuff (there's no blocker for using it for
all images now), and we can undo the image tracking shenanigans that I had to do
in bug 1605803 in nsImageFrame.
This removes the mDocGroup member because well, there's no real upside of that
now that quantum DOM is not a thing.
It also removes the static clones of the image requests, and the need for each
computed value instance to have its own request. These were needed because we
needed the image loader for the particular document to observe the image
changes. But we were also tracking the request -> loader for other purposes.
Instead, Now all the images get loaded with GlobalImageObserver as a listener,
which looks in the image map and forwards the notification to all the interested
loaders instead dynamically.
The style value is only responsible to load the image, and no longer tracks /
locks it. Instead, the loader does so, via the image tracker.
Differential Revision: https://phabricator.services.mozilla.com/D58519
--HG--
extra : moz-landing-system : lando
This removes nsStyleImageRequest by moving the load state to LoadData instead
(where other lazy state like the resolved URL and load id lives).
That way we can use cbindgen for more stuff (there's no blocker for using it for
all images now), and we can undo the image tracking shenanigans that I had to do
in bug 1605803 in nsImageFrame.
This removes the mDocGroup member because well, there's no real upside of that
now that quantum DOM is not a thing.
It also removes the static clones of the image requests, and the need for each
computed value instance to have its own request. These were needed because we
needed the image loader for the particular document to observe the image
changes. But we were also tracking the request -> loader for other purposes.
Instead, Now all the images get loaded with GlobalImageObserver as a listener,
which looks in the image map and forwards the notification to all the interested
loaders instead dynamically.
The style value is only responsible to load the image, and no longer tracks /
locks it. Instead, the loader does so, via the image tracker.
Differential Revision: https://phabricator.services.mozilla.com/D58519
--HG--
extra : moz-landing-system : lando
Otherwise it'd error, which is not great. We could add a custom attribute in
there if needed or something, or fix cbindgen to generate these for us as
described in the issue linked from the comments.
Do this in ServoStyleConstsInlines as that's where we have the guarantee that
the templates have been defined, and that the file is included every time
ServoStyleConsts is included.
Differential Revision: https://phabricator.services.mozilla.com/D61625
--HG--
extra : moz-landing-system : lando
This simplifies a bit the code, and guarantees that all calc()s have percentages
and lengths.
I also wanted to remove unclamped_length() / specified_percentage() (for the
same reason as the above patch), but they're needed for animations for now. When
I implement min() / max() for <length-percentage> they'll be fixed.
Differential Revision: https://phabricator.services.mozilla.com/D60194
--HG--
extra : moz-landing-system : lando
This is just not a thing you can do if you have min() / max() / etc, as the min
/ max value may depend on the percentage basis.
Differential Revision: https://phabricator.services.mozilla.com/D60168
--HG--
extra : moz-landing-system : lando
This simplifies a bit the code, and guarantees that all calc()s have percentages
and lengths.
I also wanted to remove unclamped_length() / specified_percentage() (for the
same reason as the above patch), but they're needed for animations for now. When
I implement min() / max() for <length-percentage> they'll be fixed.
Differential Revision: https://phabricator.services.mozilla.com/D60194
--HG--
extra : moz-landing-system : lando
This is just not a thing you can do if you have min() / max() / etc, as the min
/ max value may depend on the percentage basis.
Differential Revision: https://phabricator.services.mozilla.com/D60168
--HG--
extra : moz-landing-system : lando
We generate ByteBuf by rust bindgen, so we can drop StyleVecU8.
One potential follow-up is that we can merge this together with WrVecU8.
Differential Revision: https://phabricator.services.mozilla.com/D60328
--HG--
rename : ipc/glue/ByteBuf.h => ipc/glue/ByteBufUtils.h
extra : moz-landing-system : lando
Though this may make us use more space when serializing
StyleTransform, but we don't have to do extra conversion on the compostior
side, and this makes us easier to maintain the Rust type.
Differential Revision: https://phabricator.services.mozilla.com/D60045
--HG--
extra : moz-landing-system : lando
We generate ByteBuf by rust bindgen, so we can drop StyleVecU8.
One potential follow-up is that we can merge this together with WrVecU8.
Differential Revision: https://phabricator.services.mozilla.com/D60328
--HG--
rename : ipc/glue/ByteBuf.h => ipc/glue/ByteBufUtils.h
extra : moz-landing-system : lando
Though this may make us use more space when serializing
StyleTransform, but we don't have to do extra conversion on the compostior
side, and this makes us easier to maintain the Rust type.
Differential Revision: https://phabricator.services.mozilla.com/D60045
--HG--
extra : moz-landing-system : lando
So that it takes one pointer instead of two, and doesn't make nsStylePosition's
size blow up.
This is not as ugly as I was fearing, thankfully, though it requires a bit of
boilerplate. I think it's acceptable.
Differential Revision: https://phabricator.services.mozilla.com/D58702
--HG--
extra : moz-landing-system : lando
This is needed to support min() / max() / clamp(), etc, as those need to be a
tree of values and thus need heap storage.
This unfortunately grows LengthPercentage to be two pointers, which is bad as
it blows up the size of nsStylePosition enough to trigger the size assertions.
This patch comments out the assertion for now, the follow-up patches will
uncomment them.
Differential Revision: https://phabricator.services.mozilla.com/D58700
--HG--
extra : moz-landing-system : lando
We need to pass these two types into the compositor, so we need a better
way to serialize these rust types. We use serde and bincode to
serialize/deserialize them, and use ByteBuf to pass the &[u8] data
through IPC. We define StyleVecU8 for FFI usage only.
Differential Revision: https://phabricator.services.mozilla.com/D50688
--HG--
extra : moz-landing-system : lando
This also fixes some of the issues with -moz-image-region, where we just minted
an auto out of the blue.
Depends on D43472
Differential Revision: https://phabricator.services.mozilla.com/D43474
--HG--
extra : moz-landing-system : lando
This change is motivated by `atomic::load` being `nodiscard` on Windows
and `atomic_thread_fence` being closer to what `Arc::drop` and
`ThreadSafeAutoRefCnt` does. We keep the `MOZ_TSAN` refinement of
`ThreadSafeAutoRefCnt` as well.
Differential Revision: https://phabricator.services.mozilla.com/D41945
--HG--
extra : moz-landing-system : lando
CORS only works on http channels, so anything else that tries to do a
CORS-enabled request fails catastrophically.
resource:// images are useful for extension developers, so don't perform CORS
checks on them. We may want to also do file:// and fix bug 1565509, while at it,
if we consider it's causing developer pain.
Differential Revision: https://phabricator.services.mozilla.com/D40651
--HG--
extra : moz-landing-system : lando
I'm _really_ sorry for the size of the patch. I tried to do this in two steps
but it was a lot of work and pretty ugly.
This patch makes us use cbindgen for grid-template-{rows,columns}, in order to:
* Make us preserve repeat() at computed-value time. This is per spec since
interpolation needs to know about repeat(). Except for subgrid, which did the
repeat expansion at parse-time and was a bit more annoying (plus it doesn't
really animate yet so we don't need it to comply with the spec).
* Tweaks the WPT tests for interpolation to adopt the resolution at:
https://github.com/w3c/csswg-drafts/issues/3503.
Trade-off here, as this patch stands, is that this change makes us use less
long-living memory, since we expand repeat() during layout, but at the cost of a
bit of CPU time during layout (conditional on the property applying though,
which wasn't the case before). It should be very easy to store a cached version
of the template, should this be too hot (I expect it isn't), or to change the
representation in other ways to optimize grid layout code if it's worth it.
Another trade-off: I've used SmallPointerArray to handle line-name merging,
pointing to the individual arrays in the style data, rather than actually
heap-allocating the merged lists. This would also be pretty easy to change
should we measure and see that it's not worth it.
This patch also opens the gate to potentially improving memory usage in some
other ways, by reference-counting line-name lists for example, though I don't
have data that suggests it is worth it.
In general, this patch makes much easier to tweak the internal representation of
the grid style data structures. Overall, I think it's a win, the amount of magic
going on in that mako code was a bit huge; it took a bit to wrap my head around
it.
This patch comments out the style struct size assertions. They will be
uncommented in a follow-up patch which contains some improvements for this type,
which are worth getting reviewed separately.
Also, this patch doesn't remove as much code as I would've hoped for because of
I tried not to change most of the dom/grid code for inspector, but I think a
fair bit of the nsGridContainerFrame.cpp code that collects information for it
can be simplified / de-copy-pasted to some extent. But that was a pre-existing
problem and this patch is already quite massive.
Differential Revision: https://phabricator.services.mozilla.com/D36598
And move the useful bits of it somewhere else (ServoStyleConstInlines.h for the
inline function definitions, and nsFrame.cpp for the static assertions).
Differential Revision: https://phabricator.services.mozilla.com/D36120
And move the useful bits of it somewhere else (ServoStyleConstInlines.h for the
inline function definitions, and nsFrame.cpp for the static assertions).
Differential Revision: https://phabricator.services.mozilla.com/D36120
--HG--
extra : moz-landing-system : lando
We clamp earlier (parse time rather than computed value time), but that's the
only behavior change, which I think doesn't really matter.
Differential Revision: https://phabricator.services.mozilla.com/D35198
--HG--
extra : moz-landing-system : lando
We clamp earlier (parse time rather than computed value time), but that's the
only behavior change, which I think doesn't really matter.
Differential Revision: https://phabricator.services.mozilla.com/D35198
--HG--
extra : moz-landing-system : lando
This doesn't clean up as much as a whole, but it's a step in the right
direction. In particular, it allows us to start using simple bindings for:
* Filters
* Shapes and images, almost. Need to:
* Get rid of the complex -moz- gradient parsing (let
layout.css.simple-moz-gradient.enabled get to release).
* Counters, almost. Need to:
* Share the Attr representation with Gecko, by not using Option<>.
* Just another variant should be enough (ContentItem::{Attr,Prefixedattr},
maybe).
Which in turn allows us to remove a whole lot of bindings in followups to this.
The setup changes a bit. This also removes the double pointer I complained about
while reviewing the shared UA sheet patches. The old setup is:
```
SpecifiedUrl
* CssUrl
* Arc<CssUrlData>
* String
* UrlExtraData
* UrlValueSource
* Arc<CssUrlData>
* load id
* resolved uri
* CORS mode.
* ...
```
The new one removes the double reference to the url data via URLValue, and looks
like:
```
SpecifiedUrl
* CssUrl
* Arc<CssUrlData>
* String
* UrlExtraData
* CorsMode
* LoadData
* load id
* resolved URI
```
The LoadData is the only mutable bit that C++ can change, and is not used from
Rust. Ideally, in the future, we could just use rust-url to resolve the URL
after parsing or something, and make it all immutable. Maybe.
I've verified that this approach still works with the UA sheet patches (via the
LoadDataSource::Lazy).
The reordering of mWillChange is to avoid nsStyleDisplay from going over the
size limit. We want to split it up anyway in bug 1552587, but mBinding gains a
tag member, which means that we were having a bit of extra padding.
One thing I want to explore is to see if we can abuse rustc's non-zero
optimizations to predict the layout from C++, but that's something to explore at
some other point in time and with a lot of care and help from Michael (who sits
next to me and works on rustc ;)).
Differential Revision: https://phabricator.services.mozilla.com/D31742
This doesn't clean up as much as a whole, but it's a step in the right
direction. In particular, it allows us to start using simple bindings for:
* Filters
* Shapes and images, almost. Need to:
* Get rid of the complex -moz- gradient parsing (let
layout.css.simple-moz-gradient.enabled get to release).
* Counters, almost. Need to:
* Share the Attr representation with Gecko, by not using Option<>.
* Just another variant should be enough (ContentItem::{Attr,Prefixedattr},
maybe).
Which in turn allows us to remove a whole lot of bindings in followups to this.
The setup changes a bit. This also removes the double pointer I complained about
while reviewing the shared UA sheet patches. The old setup is:
```
SpecifiedUrl
* CssUrl
* Arc<CssUrlData>
* String
* UrlExtraData
* UrlValueSource
* Arc<CssUrlData>
* load id
* resolved uri
* CORS mode.
* ...
```
The new one removes the double reference to the url data via URLValue, and looks
like:
```
SpecifiedUrl
* CssUrl
* Arc<CssUrlData>
* String
* UrlExtraData
* CorsMode
* LoadData
* load id
* resolved URI
```
The LoadData is the only mutable bit that C++ can change, and is not used from
Rust. Ideally, in the future, we could just use rust-url to resolve the URL
after parsing or something, and make it all immutable. Maybe.
I've verified that this approach still works with the UA sheet patches (via the
LoadDataSource::Lazy).
The reordering of mWillChange is to avoid nsStyleDisplay from going over the
size limit. We want to split it up anyway in bug 1552587, but mBinding gains a
tag member, which means that we were having a bit of extra padding.
One thing I want to explore is to see if we can abuse rustc's non-zero
optimizations to predict the layout from C++, but that's something to explore at
some other point in time and with a lot of care and help from Michael (who sits
next to me and works on rustc ;)).
Differential Revision: https://phabricator.services.mozilla.com/D31742
--HG--
extra : moz-landing-system : lando
This avoids the expensive conversion, and cleans up a bunch.
Further cleanup is possible, just not done yet to avoid growing the patch even
more.
Differential Revision: https://phabricator.services.mozilla.com/D30748
--HG--
extra : moz-landing-system : lando
We could use ArcSlice if wanted I guess, your call. Though will change is not
supposed to be used very frequently.
Differential Revision: https://phabricator.services.mozilla.com/D30548
--HG--
extra : moz-landing-system : lando