This also removes some confusing comments around nsIDocument regarding some kind
of "special" stylesheets, which don't seem to exist anymore, and consolidates
StyleSheetList so that we only have one implementation.
I think that fixes a potential leak on the shadow root code (even though the API
is v0 only), given the pointer from the ShadowRootStyleSheetList to the
ShadowRoot wasn't being CCd.
Also, more stuff could be renamed, methods removed, etc, feel free to suggest
more cleanup, I've done mostly the minimal.
Next steps are moving the stylesets there and stop using the proto binding sheet
list / resources.
MozReview-Commit-ID: D9hnDgPQAS5
This also removes some confusing comments around nsIDocument regarding some kind
of "special" stylesheets, which don't seem to exist anymore, and consolidates
StyleSheetList so that we only have one implementation.
I think that fixes a potential leak on the shadow root code (even though the API
is v0 only), given the pointer from the ShadowRootStyleSheetList to the
ShadowRoot wasn't being CCd.
Also, more stuff could be renamed, methods removed, etc, feel free to suggest
more cleanup, I've done mostly the minimal.
Next steps are moving the stylesets there and stop using the proto binding sheet
list / resources.
MozReview-Commit-ID: D9hnDgPQAS5
See bug 1422633, there are assertions missing, and servo doesn't assert at all
anymore.
I don't think it's worth optimizing / lazily resolving it, each time the
document state changes.
We usually just restyle the world anyway (which requires recomputing it), and
the changes that it's optimizing (nsWindow::SetActive() and XUL root element
localedir attribute changes) aren't common enough to warrant the complexity I'd
say.
This doesn't handle invalidating the cache in the case the root element goes
away, I haven't bothered because it was already broken, and GetRootElement() is
already gone in RemoveSubtreeFromDocument.
MozReview-Commit-ID: 9RuQhmmy7Kr
This also makes the rule map not process all the stylesheets for the document,
which would be a mess with shadow DOM.
Far from the final, ideal state, but hey, progress.
MozReview-Commit-ID: 7TrifME9VZ
In the current implementation, we call SetStylistStyleSheetsDirty() every time
a style sheet is changed. However, the dirty bit setting may or may not always
update the style data. For example, the style data for undisplayed elements are
deliberately not updated in Stylo. However, the getComputedStyle API is supposed
to provide a way to get the up-to-date computed style data, even for undisplayed
elements.
In this patch, we increment RestyleGeneration for undisplayed elements when we
call SetStylistStyleSheetsDirty(). This could flush the cached data that
getComputedStyle API holds, and ensures the getComputedStyle API computes a new
one.
MozReview-Commit-ID: JDDhACOG3z4
This is a pre-patch for the real fix of Bug 1418433.
In the real fix, we'll add a IncrementUndisplayedRestyleGeneration() call into
SetStylistStyleSheetsDirty(). However, the IncrementUndisplayedRestyleGeneration()
call needs get through some deep structures in nsPresContext, RestyleManager,...
etc., and doing so means we need to move bunches of related include files, forward
declarations, from .cpp file to .h file, which doesn't make sense.
Instead, we move the implementation parts of SetStylistStyleSheetsDirty() to .cpp
file (since it is now a bit more complicated than it was), so we can use the existing
include files in the .cpp file to add IncrementUndisplayedRestyleGeneration() call
(this is in a following patch).
MozReview-Commit-ID: 3Vp9qyCf8NA
The textarea is inserted under a Shadow host, with no matching insertion point,
so its flattened tree parent node is null.
We're treating this case in the restyle root code as "the parent is the
document", but that's very wrong.
MozReview-Commit-ID: JlzUMRIYaYZ
There's nothing preventing the flat tree from changing while the document
doesn't have a shell. In that case, we really really don't want to lose track
of elements with stale style data, since then we'll mess up.
It's ok to _not_ clear the style data when the document goes into the BFCache
though, because the document is thrown away if other document runs script and
touches the cached DOM.
MozReview-Commit-ID: 4W3xDAnnLPL
They're useless now, provided we remove the hack to not traverse XBL-bound
elements on initial styling.
This also allows us to get rid of the fallback case.
MozReview-Commit-ID: AvBVdyF1wb6
We not only need to care about children getting inserted in the flat tree, but
also about children moving _out_ of the flat tree.
In particular, as of right now we may leave stale data on elements when they
disappear from the flattened tree.
We're lucky enough that in 99% of the situations we enter in[1] and that clears
all the stuff, including servo data. But my assertions for bug 1414999 caught
the template / observes case.
Thus, just clear the whole bound element subtree data, and restyle it in the
end, no need for StyleNewChildren. This matches what we do for shadow DOM
(though in the shadow DOM case we do it async in DestroyFramesForAndRestyle).
[1]: https://searchfox.org/mozilla-central/rev/9bab9dc5a9472e3c163ab279847d2249322c206e/dom/xbl/nsXBLBinding.cpp#368
MozReview-Commit-ID: 69A0aR0AFfU
(Path is actually r=froydnj.)
Bug 1400459 devirtualized nsIAtom so that it is no longer a subclass of
nsISupports. This means that nsAtom is now a better name for it than nsIAtom.
MozReview-Commit-ID: 91U22X2NydP
One remaining issue is that requesters of CSSRules might also want to know if those rules are complete, and to listen for completion if the rules are incomplete. Bug 1404538 will fix this up at least for devtools.
MozReview-Commit-ID: COnkS40Ooe2
This will allow us to verify the entire detection pipeline in real nightly
builds, which will give us confidence that real heap corruption will be
detected and reported properly.
MozReview-Commit-ID: 43Fp2HT8RYy
ServoStyleSetSizes now has two uses, one for the Stylist, and one for the UA
cache, and so the patch removes 'Stylist' from the field names.
Example output from about:memory:
> +----1,359,608 B (00.55%) -- layout
> | +----756,488 B (00.31%) -- style-sheet-cache [2]
> | +----393,968 B (00.16%) -- servo-ua-cache
> | | +--234,496 B (00.10%) -- element-and-pseudos-maps
> | | +---59,648 B (00.02%) -- revalidation-selectors
> | | +---58,320 B (00.02%) -- invalidation-map
> | | +---30,752 B (00.01%) -- other
> | | +---10,752 B (00.00%) -- precomputed-pseudos
MozReview-Commit-ID: 8oxuJO0ojp
MozReview-Commit-ID: 5eTW5dCcxzP
This patch reverts the code changes, but not the test changes from changeset bf9cf6f393eb. There is a crashtest here which is valuable to keep.
Example output from the Obama Wikipedia page:
> ├──2,315,600 B (01.16%) -- stylist
> │ ├──1,916,928 B (00.96%) ── invalidation-map
> │ ├────228,800 B (00.11%) ── rule-tree
> │ ├────142,336 B (00.07%) ── element-and-pseudos-maps
> │ ├─────14,336 B (00.01%) ── revalidation-selectors
> │ ├──────9,648 B (00.00%) ── other
> │ └──────3,552 B (00.00%) ── precomputed-pseudos
This change requires new code to measure HashMaps, which uses the new
'malloc_enclosing_size_of' functions that can measure a heap block from an
interior pointer.
The patch changes MallocSizeOfFn to a newtype, and introduces
MallocEnclosingSizeOfFn alongside.
It also adds new traits: MallocSizeOfBox, MallocSizeOfVec, MallocSizeOfHash.
These each contain a single method that does shallow measurement of the
relevant type, which is often useful. (This is a different style to the
existing MallocSizeOf trait, which does deep measurement, but I'm moving away
from the always-deep-measurement style because it's less flexible.)
MozReview-Commit-ID: FgJCCmdw0ZF
The motivation of this patch is that clearing mPresContext should be an
implementation detail of XBL style set, so I create a method for that, and
remove ClearPresContext().
MozReview-Commit-ID: Ftta0rcAqu6
