As expected, this is specific to the UA widget stuff.
What's going on here is that given we don't clear out the host when unattaching
the shadow tree, mutating that shadow tree still notifies all the way up to the
document, and that gets all the other code confused, thinking that the node is
connected.
Indeed, the first assertion that fails when loading that test-case in a debug
build is:
https://searchfox.org/mozilla-central/rev/17f55aee76b7c4610a974cffd3453454e0c8de7b/dom/base/nsNodeUtils.cpp#93
This seems the best fix to avoid confusion. Also clear the mutation observer,
to completely forget about the host.
Chrome code dealing with UA widgets needs to be careful, but I think this is
safe. All the code that assumes that GetHost() doesn't return null is in code
dealing with connected shadow trees only (style system / layout), or in
mutation observer notifications from the host.
Differential Revision: https://phabricator.services.mozilla.com/D11369
--HG--
extra : moz-landing-system : lando
Various places in dom/ use the pattern:
already_AddRefed<NodeInfo> ni = ...;
which is supposed to be disallowed by our static analysis code, but
isn't, for whatever reason. To fix our static analysis code, we need to
eliminate instances of the above pattern.
Unfortunately, eliminating this pattern requires restructuring how Nodes
are created. Most Node subclasses take `already_AddRefed<NodeInfo>&` in
their constructors, and a few accept `already_AddRefed<NodeInfo>&&`. We
need to enforce the latter pattern consistently, which requires changing
dozens of source files.
So that we don't waste a bunch of memory with stuff like <svg:use>. I
plan to shrink AuthorStyles further, but this should help regardless, and isn't
very complex.
Differential Revision: https://phabricator.services.mozilla.com/D4618
--HG--
extra : moz-landing-system : lando
This is on top of bug 1486623.
While at it removed the 'stale computed styles' report, since it was unused
since I removed the undisplayed maps.
Differential Revision: https://phabricator.services.mozilla.com/D4453
--HG--
extra : moz-landing-system : lando
This moves all the node-specific reporting to nsIDocument.
OrphanReporter delegates all the reporting to that and then returns the sum of
all sizes, which is not ideal but was pre-existing.
Also, I moved the main mStyleSheets size reporting to DocumentOrShadowRoot for
it to be shared between document and ShadowRoot.
I'll add memory reporting for the computed stylesheet maps and such in the
ShadowRoot in a followup.
I went through all the XBL bindings, though it seems I could just use
GetBindingWithContent(), since according to:
https://searchfox.org/mozilla-central/rev/55da592d85c2baf8d8818010c41d9738c97013d2/dom/xbl/nsXBLBinding.cpp#615
We don't allow multiple bindings to contribute anon content. Anyway it was the
same amount of code...
Differential Revision: https://phabricator.services.mozilla.com/D4433
--HG--
extra : moz-landing-system : lando
While trying to repro bug 1484293 I noticed that this assertion failed:
https://searchfox.org/mozilla-central/rev/ef8b3886cb173d5534b954b6fb7eb2d94a9473d0/dom/base/ShadowRoot.cpp#160
(during unlink, while unbinding the kids)
We rely on GetComposedDoc returning the right thing during unbind to cleanup
some stuff (see bug 1473637 for example), so it should probably be correct all
the time, regardless of whether something is unlinked or not.
Also this makes GetComposedDoc() much faster, which is nice too, since we call
it somewhat often.
I removed NodeHasRelevantHoverRules, since it's unused (was used by the old
style system).
I moved the SetIsConnected(false) call for the shadow root to before unbinding
the kids for consistency with what Element does with the uncomposed doc flag,
now that the children's connectedness doesn't depend on the shadow root's.
Differential Revision: https://phabricator.services.mozilla.com/D3715
--HG--
extra : moz-landing-system : lando
While trying to repro bug 1484293 I noticed that this assertion failed:
https://searchfox.org/mozilla-central/rev/ef8b3886cb173d5534b954b6fb7eb2d94a9473d0/dom/base/ShadowRoot.cpp#160
(during unlink, while unbinding the kids)
We rely on GetComposedDoc returning the right thing during unbind to cleanup
some stuff (see bug 1473637 for example), so it should probably be correct all
the time, regardless of whether something is unlinked or not.
Also this makes GetComposedDoc() much faster, which is nice too, since we call
it somewhat often.
I removed NodeHasRelevantHoverRules, since it's unused (was used by the old
style system).
I moved the SetIsConnected(false) call for the shadow root to before unbinding
the kids for consistency with what Element does with the uncomposed doc flag,
now that the children's connectedness doesn't depend on the shadow root's.
Differential Revision: https://phabricator.services.mozilla.com/D3715
--HG--
extra : moz-landing-system : lando
The DOM elements within the UA Widget Shadow DOM should have its reflectors in
the UA Widget Scope. This is done by calling nsINode::IsInUAWidget() which
would check its containing shadow and its UA Widget bit.
To prevent JS access of the DOM element before it is in the
UA Widget Shadom DOM tree, various DOM methods are set to inaccessible to
UA Widget script. It would need to use the two special methods in ShadowRoot
instead to insert the DOM directly into the shadow tree.
MozReview-Commit-ID: Jz9iCaVIoij
--HG--
extra : rebase_source : b7b17be68dcde00cfeb207cb39cf16b486f2ab02
Since sed on multiple lines ended up being such a pain and I didn't end up
writing a script for this because I didn't think it'd end up being so boring, I
may have made a couple cleanups here and there as well...
Differential Revision: https://phabricator.services.mozilla.com/D2887
--HG--
extra : moz-landing-system : lando
Adding the Places* files into unified sources pushed the
unified sources into a situation that exposed a strangely
large number of errors. This seems to be the minimum set of
changes I could make to resolve all of the issues.
MozReview-Commit-ID: C2H9ce8FmE4
--HG--
extra : rebase_source : 61afc5481dc8ec34caba1886bd74200cf3659fb4
Adding the Places* files into unified sources pushed the
unified sources into a situation that exposed a strangely
large number of errors. This seems to be the minimum set of
changes I could make to resolve all of the issues.
MozReview-Commit-ID: C2H9ce8FmE4
--HG--
extra : rebase_source : 4f8dd2996d820fdb5a07afe544be5e2d6ca6a5c7
Adding the Places* files into unified sources pushed the
unified sources into a situation that exposed a strangely
large number of errors. This seems to be the minimum set of
changes I could make to resolve all of the issues.
MozReview-Commit-ID: C2H9ce8FmE4
--HG--
extra : rebase_source : b01f47e439a61492ad999ae30677c48535e8cd4c
Adding the Places* files into unified sources pushed the
unified sources into a situation that exposed a strangely
large number of errors. This seems to be the minimum set of
changes I could make to resolve all of the issues.
MozReview-Commit-ID: C2H9ce8FmE4
--HG--
extra : rebase_source : 571fd3b1e6511daa5731da76fb5d6d97bce11db1
Adding the Places* files into unified sources pushed the
unified sources into a situation that exposed a strangely
large number of errors. This seems to be the minimum set of
changes I could make to resolve all of the issues.
MozReview-Commit-ID: C2H9ce8FmE4
--HG--
extra : rebase_source : 7a3b71596b4318f517ec4c3ac0180e2aa3b721c7
Just something I found while trying to construct a test-case for this.
MozReview-Commit-ID: A01qzQ14QG9
--HG--
extra : rebase_source : 5329a8791774b402b633a992fa9bca2cc5b320fb
Process and non-process managers have different script loader interfaces
(ProcessScriptLoader/GlobalProcessScriptLoader vs FrameScriptLoader). The WebIDL
conversion used the same interface for some process and
non-process managers, but because of the different script loader interfaces they really
should be using separate interfaces.
--HG--
rename : dom/base/ChromeMessageBroadcaster.cpp => dom/base/MessageBroadcaster.cpp
rename : dom/base/ChromeMessageBroadcaster.h => dom/base/MessageBroadcaster.h
rename : dom/base/ChromeMessageBroadcaster.cpp => dom/base/ParentProcessMessageManager.cpp
rename : dom/base/ChromeMessageBroadcaster.h => dom/base/ParentProcessMessageManager.h
rename : dom/base/ChromeMessageSender.cpp => dom/base/ProcessMessageManager.cpp
rename : dom/base/ChromeMessageSender.h => dom/base/ProcessMessageManager.h
extra : rebase_source : c9b0c543f9f367535919a6c6840e5ba038023112
extra : histedit_source : 7749f98e11e25423fcf414cc1f0415104343798a
Our behavior is correct, this uses the same setup that nsDocument and the
stylesets use, which I may look into fixing up / making more explicit in
bug 1465031.
MozReview-Commit-ID: 75AToXCw1pV
--HG--
extra : rebase_source : b7c11ca66b416c32b8fc0c5eedbc9383c63bad70
Now that BeginUpdate is useless for the UPDATE_STYLE case, we don't need the
update mechanism at all. Just ensure that ApplicableStylesChanged is called on
the pres shell via the relevant RuleChanged, etc. notifications.
There's a big hidden gotcha here. nsIDocument::BeginUpdate does put a script
blocker on the stack for these updates. However it's not needed, since no script
can run during these notifications (only the stylesheet events we post for
devtools, but those use AsyncEventDispatcher and PostDOMEvents, so they don't
try to run immediately).
nsIDocument::BeginUpdate also does XBL binding attached queue stuff, but we
can't change bindings during these notifications anyway, so it also doesn't
matter.
MozReview-Commit-ID: HJvK6zQfloh
This is sound because the unlink implementation of the stylesheet drops the
preserved wrapper, and there are no strong references back to any node from the
stylesheet or any of the non-unlinked members.
This almost is the same setup that works for document sheets. We need to account
for a double reference in case the sheet is applicable because Servo keeps
another reference to it in that case, instead of in the StyleSet / PresShell.
Added the testcase as a crashtest, in the hopes that if it regresses leak
reporting on automation will catch it.
MozReview-Commit-ID: Kcc5oaOvP9A
The dirty bit fiddling is nontrivial, but it's pretty much what we do for
invalidation and allows to keep this incrementally easily.
The only caller that we cared about for the GetFlattenedTreeParent check in
DestroyFramesForAndRestyle was the old ShadowRoot invalidation functions that
went away.
MozReview-Commit-ID: GmgPPJ6d2qX
Summary:
When the host is bound to the document, it'll also re-BindToTree the shadow
tree, which will update the stylesheets and all that.
Reviewers: xidorn
Bug #: 1446507
Differential Revision: https://phabricator.services.mozilla.com/D742
MozReview-Commit-ID: fBLq5owB7U
Summary:
Also, make stuff sound in presence of CSSOM and what not.
The dirty: false thing is reverting an accidental change that landed in the
de-XBL stuff, which was harmless, but now wouldn't let me assert stuff properly.
Reviewers: xidorn
Bug #: 1445682
Differential Revision: https://phabricator.services.mozilla.com/D748
MozReview-Commit-ID: K0W2Rv0qK8X
Summary:
Before that bug, we carefully told layout about all distribution changes before
they happened, so layout could cleanup frames and styles synchronously properly
(since otherwise there's no way afterwards to figure out what the tree shape
was).
That bug made it not do it correctly, causing this.
I obviously need to write a bunch of tests for this...
Reviewers: smaug
Bug #: 1438210
Differential Revision: https://phabricator.services.mozilla.com/D724
MozReview-Commit-ID: 8uupNhoFwme
More improvements to come. In particular, this still iterates through Shadow DOM
in each_xbl_cascade_data, but that should be changed later. That allows to
cleanup a bunch of stuff and finally fix Shadow DOM cascade order.
We still rely on the binding parent to be setup properly in the shadow tree, but
that requirement can go away later (we can walk the containing shadow chain
instead).
This mostly focuses on removing the XBL binding from the Shadow host.
It'd be nice to do EnumerateShadowRoots faster. I think that should also be a
followup, if needed.
MozReview-Commit-ID: Jf2iGvLC5de
Sylvestre Ledru
Olli Pettay
Emilio Cobos Álvarez
Nathan Froyd
shindli
Timothy Guan-tin Chien
Doug Thayer
Dorel Luca
Jeff Gilbert
Bogdan Tara
arthur.iakab
Peter Van der Beken
Xidorn Quan
Boris Zbarsky
Andreea Pavel