Not canceling the timer when run in a worker could lead to the timer firing
after the primary worker runnable finished running, leading to an assertion
failure in cycle collector code.
Because we're not updating language packs this release, we have to disable the
reporting portion of the reload flood guard to prevent errors. Functionality
of the guard is otherwise the same.
Fixes potential thread races between:
- nsSocketTransport::OnInputClosed and nsSocketTransport::OpenInputStream.
- nsSocketTransport::OnOutputClosed and nsSocketTransport::OpenOutputStream.
- nsSocketTransport::Close and nsSocketTransport::RecoverFromError called from
OnSocketDetached.
Critical note: this potentially reduces the strength of TLS 1.3 and should only
be enabled if absolutely necessary to access a site.
A browser restart is required for the pref change to take effect as it is set on
NSS initialization.
In almost all cases this is pointless to include, a ton of useless entries,
and there's no reason to have people post a full enumeration of installed
printers on their system when asked for TS information.
If there's a printer specific issue, people can always be asked to post relevant
info manually.
This rewrites the websocket channel network change detection function to not
skip part of its logic in a situation that has already been checked, preventing
a thread race. See analysis of the problem in the issue.
According to the original SVG spec, the SVG element in documents had to follow
special rules as to what was allowed for width/height attributes. This is no
longer the case, so we can decouple it from whatever internal units SVG/XML
language uses and allow all units like every other element.
The implementation is based on the work by Bernstein and Yang
(https://eprint.iacr.org/2019/266)
"Fast constant-time gcd computation and modular inversion".
It fixes the old mp_gcd and s_mp_invmod_odd_m functions. The patch also fixes
mpl_significant_bits s_mp_div_2d and s_mp_mul_2d by having less control flow to
reduce side-channel leaks.
Co-authored by : Billy Bob Brumley
With the changes to layout for WebComponents, it is now apparently possible to
pass in null for frame tree items to this function, which would cause a null
deref crash if not checked.
As of [da0c073a7] we no longer need this workaround because the issue is avoided
with proper sleep/wake logic restored.
This reverts commit 2fa993b563.
The root cause in this bug is that the connection info used by
'SpdyConnectTransaction' is the same instance as the connection info in
'nsHttpTransaction', so we should clone it and let 'SpdyConnectTransaction' use
the cloned one.
The original patch handled the grow case but not the shrink case. When the
current and new allocation sizes are in different size classes, jemalloc's
realloc will move the allocation when shrinking, not just truncate the existing
one.
Based on work by Jon Coppeard.
This involves refactoring the vibrancy and OpenGL/Pixel rendering changes contained
in the following Mozilla meta bugs: 1496823 and 1491445
Also add Big Sur to the features tests and update popup menu look and feel based
on Mozilla bug 1656301.
There don't seem to be any drawbacks to this; tested for the past month disabled
and there have been no issues with any sites visited. Adoption seems very low.
This is required for UniquePtr to accept <void>,
which is required for PseudoHandle = mozilla::UniquePtr<T, JS::FreePolicy>;
in turn for mozilla::SegmentedVector<PseudoHandle<void>> uniquePtrArena_;
Tag #1679
While we do fail a couple of tests, the other mainstream browsers also fail them and I think our implementation of tab-size is good enough to be unprefixed at this point. Having this patch also makes testing easier.
Currently -moz-tab-size only accepts <number> values, and both Chrome and Firefox currently support <length> values and have for some time now. So with this you would be able to support sizes in px or em, for instance. This was implemented in Firefox 53 and was trivial to backport.
This prevents a potential race and simplifies the code a bit by keeping the
bytes read separate instead of using mData, which is modified from another
thread from OnDataAvailable.
Relaxed atomics are fine for these, since they don't guard any memory.
This actually keeps both pseudo-elements for now, since the prefixed version is
still used internally, but we need the unprefixed version for web compat.
Note: while unprefixing a non-spec-compliant pseudo here, it's exactly in line
with what other browsers do. Nobody is following the spec here and at least
we'll be doing what everyone else is with our unprefixed version.
Mozilla's original implementation of this failed a couple of tests, but this seems to solve all the problems. Basically, the caret-color wasn't able to be set differently based on whether a link was visited, and the auto value implementation was incomplete. The only test we fail now is the one where you have grey text on a grey background and the caret is supposed to be visible, but I think that may have been removed from the spec. Even if it wasn't, no other browser supports it anyway.
This CSS property allows input carets (that blinking input cursor you see in text fields), to be given a custom color. This was implemented in Firefox 53, and it was such a minor feature that no one ever missed it, but I don't see any harm in implementing this.
https://bugzilla.mozilla.org/show_bug.cgi?id=1063162
This aligns with the current spec regarding overflow-wrap: break-word and
overflow-wrap: anywhere in if it affects intrinsic sized due to considering
soft-wrap opportunities or not.
See CSS Text Module Level 3, Editor’s Draft, 1 October 2020, Section 5.5
I got very anxious about making sure I included VARIANT_OPACITY in all the places VARIANT_NUMBER was included to make sure it couldn't possibly break unexpectedly, and that led to me accidentally breaking a mechanism that prevented percentages from serializing as numbers in other parts of the code. It was a total accident, and these additions were unnecessary. Basically, the situation is that there was one part of the code where it determines what's allowed for the flex statement (and possibly other statements) by checking whether it got stored as a "number", and basically only disallows percentages if it attempted to store/serialize them as percentages.
However, it only got to that part of the code because I accidentally allowed VARIANT_OPACITY as a valid way for certain tokens to parse where it wasn't necessary. If it tries to parse it that way under very specific circumstances... percentages will be marked valid and fed through the system as numbers rather than being rejected and not serialized at all, because the check to disallow percentages there relied on them being stored as percentages.
It's a really weird thing to have a problem with in a lot of ways, because if percentages aren't allowed in a field, you would think people wouldn't try to use them there, much less depend on the broken behavior that results from them not parsing as a related value.
A race condition seemed to exist between tab destruction and un-registering a ResizeObserver resulting in a null deref crash.
The original reporter in Forum Topic 25311 experienced this on msn.com so that was the functional test reference.
This make MediaQueryList inherit from EventTarget and adds MediaQueryListEvent
as an interface as well as the onchange() method.
This should not affect compatibility with other code; the event object is a
MediaQueryListEvent instance, which is recognized as a MediaListQuery instance.
This DiD measure ensures that our async HRTF database loading is completed
before we actually try to use it. If not done, database() simply returns null.
Even though percentages are already treated as floats internally by the style system for computation purposes, you have to go out of your way to stop them from being read back out as percentages. What I do here amounts to storing the percentage token in the "wrong" container, the one normally used for floats. This allows a value that was read in as a percentage to be read back out as something else, which is normally prevented by the design of the style system.
This preliminary step allows percentages to be computed and display correctly,
but unfortunately it fails a test after changing VARIANT_HN to VARIANT_HPN because that allows values to be serialized as percentages. However, not doing this means percentages are rejected as valid values for the user to input. The way the style system is setup makes it hard to change this for opacity without changing it for everything else, especially since some code-saving speed hacks in Bug 636029 and Bug 441367 that make a lot of assumptions about this stuff very rigid.
There are situations where nsCSSClipPathinstance->CreateClipPath(dt)
returns null. We need to check for this before trying to use its
functions. If there is no clip path, then always return "no hit".
Host name was purely being used for HPKP and since HPKP is killed,
this can also go. Currently it doesn't do anything other than
generating build warnings.
Since the local variable is always initialized to false, we don't actually need to declare it and can just pass "false" directly as a parameter to the PrepareSheet function's bool. I was worried about code readability at first, but some well-placed comments took care of that.
This clarifies the assumptions the code is making and the order in which the variables pass through the loading process. The new variable is set after the sheet is created and prepared, and is assumed to be false in the beginning.
This is just a clean port of 1322191 and follow-up 1325970. It really seems to add create a new way to access existing code relating to block formatting and floating elements rather than implementing new functionality, and it is mercifully straightforwards.
Unless a user is debugging media errors, this detail is unnecessary to report
and could include sensitive data which could be abused by third-party
requesters. This aligns it with the standard success/error paradigms in normal
browsing situations.
This commit does several things:
- Moves the pref check from ScriptLoader to ns[I]Document so it can be called on
the document.
- Changes the atrribute freezing function to a better name that takes the
document as a parameter.
- Sets the proper async/defer attributes on HTML script elements based on
keywords and whether they are module scripts or not.
This simplifies handling of combinations of async/defer by assigning one and
only one state to scripts.
If async then always async > if defer or module then defer > otherwise blocking.
This makes sure we don't block body-referred sub-resources by head-referenced
defer and async scripts. This is important for all script loads, not just
modules, but is added here because it was run into while implementing modules.
The added debug assertion does not work due to missing API.
They were added in BZ 1337491, 1395366, and others, but were primarily used for multi-threading.
This uses our existing non-multithreaded syntax instead, resolving a `no member named` build error.
According to the spec this isn't created until the module is instantiated, but
we create it when we compile the module.
We stored this previously in InitialEnvironmentSlot and copied it to
EnvironmentSlot when it was supposed to be created, but we can just store it in
the latter slot straight away and check the module's status and return null if
it shouldn't exist yet.
This reduces the number of slots needed on a moduleObject to 17.
Re: BZ 1420412 Part 1
We can't implement the second part to further reduce our number of slots,
because it relies on SetProxyReservedSlot which in turn relies on
rearchitecturing JS proxies to make reserved slots dynamic. That's a rabbit hole
we really don't want to fall into.
So, we'll end up being a bit slower because it can't be in-line allocated with
having more than 16 slots, but so be it. I sincerely doubt it will make any
practical difference.
This updates module implementation to match spec regarding handling of
instantiation errors, after it was changed yet again, this time to not remember
instantiation errors, but instead immediately rethrow applicable ones.
Ref: BZ 1420420
Our use of SQLite explicitly avoids using MSVC's _msize which the lib would
notrmally use. This was set up this way in BZ bug 719579 for legitimate reasons
(because _msize would end up using the system symbol, which would crash trying
to get the size information from jemalloc allocations). Since we build the
SQLite lib separately there is no easy way to avoid this warning as it's not
directly linked against mozglue.dll which has the jemalloc functions it wants
for _msize when not using the system allocator. It is implicitly assuming the
function is an extern returning an int (with the call specifically casting to
int) which is correct (well, correct _enough_, anyway) so the warning can
safely be ignored for our specific use as it all lines up at link time.
Newly introduced aspect-ratio property did not have CSS_PROP_LIST_EXCLUDE_INTERNAL defines, resulting in the following assertion:
\!nsCSSProps::PropHasFlags(p, (1<<28)) (properties defined outside of #ifndef CSS_PROP_LIST_EXCLUDE_INTERNAL sections must not have the CSS_PROPERTY_INTERNAL flag), at ...layout/style/nsCSSProps.cpp:289
This patch resolves the assertion by adding #ifndef around the aspect-ratio property.
Since this is supported as an alias by Firefox and Edge for the same
reasons and we have websites using this to (attempt to) override the
system-provided styling with their own, leaving out the only supported
keyword we'd otherwise have (with -moz- prefix) but still stating
-webkit-.
TODO: unprefix this completely and make the vendor prefixes aliases.
This is apparently used for fallback selection and if available it is "assumed"
Shadow DOM is also available, while this is a utility function.
Webcompat is a nightmare sometimes.
At first, HTMLEditor::GetActiveEditingHost might return null in this situation, we should check whether nullptr is returned. At second, SplitNodeDeep returns error since curent is design mode and selection node has no parent. So we should check error.
Ref: Bug 1350772
https://bugzilla.mozilla.org/show_bug.cgi?id=1547792
Aspect Ratio handling simplified by using floating point integers:
- Multiplication of value (or inverse value) to a known side for Scaling
- No unequal equal values such as "4/3" vs "8/6" vs "20/15"
- Truly "Empty" aspect ratios, even if one dimension is not 0
I meant to do this a long time ago, but basically it accounts for the new XP_SOLARIS build flag that never made it into the MailNews code. Additionally, it enables a compatibility flag for Solaris 11.4 that allows us to use the three-argument implementation of ctime_r still used by Solaris 11.3 and illumos (which also appears equivalent to the NSLDAPI_CTIME implementation used by libldap internally). Also, the ctime_r function has been added to the time.h header library for a while now, not sure why Mozilla thought we didn't have a ctime_r implementation.
When using AF_CONN addresses, don't put these in the COOKIE
chunk. For these addresses it is possible to reconstruct them
locally. Conceptually, addresses are something to be shared
with the peer, but in the case of AF_CONN this might not be
the case. Therefore, zero then out.
Thanks to Natalie Silvanovich of Google Project Zero for
finding and reporting the issue.
JIT optimizations involving the Javascript 'arguments' object could
potentially confuse later optimizations, so we simply disable these
optimizations as a DiD measure.
These should all be spec-compliant and were (for release-trickling of features)
arbitrarily disabled by Mozilla at our fork point. There's no real reason to
keep them disabled since they are used in the wild.
A version bump triggering add-on compatibility check will re-evaluate the "native" status (among other things) so no extra or special work is required to fix erroneous warnings.
tl;dr instafix!
The added debug assertion did not take into account microtask refactoring done in BZ 1405821.
Resulting in error: no member named 'IsInMicroTask' in 'nsContentUtils'.
This resolves the error.
This changes NativeObject::slotSpan() to get the class from the object group
rather than getting it from the base shape to avoid a race between
Shape::ensureOwnBaseShape and background sweeping.
The implementation is based on the work by Bernstein and Yang
(https://eprint.iacr.org/2019/266)
"Fast constant-time gcd computation and modular inversion".
It fixes the old mp_gcd and s_mp_invmod_odd_m functions. The patch also fixes
mpl_significant_bits s_mp_div_2d and s_mp_mul_2d by having less control flow to
reduce side-channel leaks.
Co-authored by : Billy Bob Brumley
Includes removing an error code for a function that never fails, and removing an
error return when the function successfully did what it said it would.
Ref: BZ 1644477
For checking if a module is in an error state and what the error is, it
shouldn't matter if we are currently GC-ing or not. So we don't need to check
for it, which removes the requirement to pass in the JS context (needed for
AssertHeapIsIdle's runtime check); this unblocks progress where otherwise we'd
have to figure out what the context is at the module level just to satisfy this
check.
This splits ScriptLoader up the same way Mozilla did with the exception of
ScriptRequest due to the fact that ScriptLoader and ScriptRequest are
interdependent and would create a circular dependency if split apart when not
using unified building.
In a lot of places, edit action handlers and their helper methods return nsresult and aHandled and aCanceled with out params. However, the out params cause the code complicated since:
* it's not unclear if the method will overwrite aHandled and aCanceled value.
* callers need to create temporary variable event if some of them are not necessary.
This patch rewrites the helper methods of HTMLEditRules::WillDeleteSelection() with it.
Tag #1563
When selection is collapsed and JoinBlocks() doesn't handle nor cancel the action, WillDeleteSelection() should move selection to the start/end of leftmost/rightmost editable leaf node and retry to handle the action again.
For avoiding infinite loop, it checks if selected node is changed actually before calling itself again.
Tag #1563
When HTMLEditRules::WillDeleteSelection() tries to remove something from the end/start of a block to its last/first text node but it's contained by block elements, it tries to join the container and the block. However, JoinBlocks() always fails to join them since it's impossible operation. In this case, HTMLEditRules::WillDeleteSelection() should retry to remove something in the leaf, however, it's impossible for now because JoinBlocks() and its helper methods don't return if it handles the action actually.
This patch renames |JoinBlocks()| to |TryToJoinBlocks()| for representing what it is. And this patch adds |bool* aHandled| to the helper methods. Then, *aHandled and *aCancel are now always returns the result of each method. Therefore, for merging the result of multiple helper methods, callers need to receive the result with temporary variables and merge them by themselves.
Note that when they modify DOM node actually or the action should do nothing (for example, selection is across tables), aHandled is set to true.
Tag #1563
Dynamic script loading/unloading (thank you modules) can yank the script
out from under us before the JS API for it is initialized, leading to
null deref crashes.
This adds a simple check if the passed-in object is sane and present.
Resolves#1602
This removes the (default disabled) node.rootNode readonly attribute
and replaces it with a node.getRootNode() function per WhatWG
spec discussion.
Based on work by John Dai <jdai@mozilla.com>
This is entangled with the IPC messaging configuration so
we need these IPDLs even if sppech synthesis is otherwise
not being built.
This resolves#1560
The JEMALLOC_USES_MAP_ALIGN code has turned out to be worse than useless, and in fact it breaks 64-bit SunOS. It's very old and turned out not to be needed anymore because the way the memory allocator works has changed since it was implemented. It prevented a fix I tried for 48-bit addressing from working properly. However, without either this code or the 48-bit addressing fix, the 64-bit version won't even start, which is why I thought the code was still needed.
https://bugzilla.mozilla.org/show_bug.cgi?id=457189https://hg.mozilla.org/mozilla-central/rev/a26c500b98ab
The 48-bit addressing fix is based on code found here:
https://github.com/OpenIndiana/oi-userland/blob/oi/hipster/components/web/firefox/patches/patch-js_src_gc_Memory.cpp.patch
I already applied these changes to js/src/gc/Memory.cpp, but as I was looking through jemalloc.c I saw that there were very similar ifdefs for Linux on SPARC as the ones I'd had to enaable in Memory.cpp, but for whatever reason the patches I found didn't touch them. So I tried doing for jemalloc.c what was already done for Memory.cpp, and it worked (but only after I removed the map align code).
The JEMALLOC_USES_MAP_ALIGN code has turned out to be worse than useless, and in fact it breaks 64-bit SunOS. It's very old and turned out not to be needed anymore because the way the memory allocator works has changed since it was implemented. It prevented a fix I tried for 48-bit addressing from working properly. However, without either this code or the 48-bit addressing fix, the 64-bit version won't even start, which is why I thought the code was still needed.
https://bugzilla.mozilla.org/show_bug.cgi?id=457189https://hg.mozilla.org/mozilla-central/rev/a26c500b98ab
The 48-bit addressing fix is based on code found here:
https://github.com/OpenIndiana/oi-userland/blob/oi/hipster/components/web/firefox/patches/patch-js_src_gc_Memory.cpp.patch
I already applied these changes to js/src/gc/Memory.cpp, but as I was looking through jemalloc.c I saw that there were very similar ifdefs for Linux on SPARC as the ones I'd had to enaable in Memory.cpp, but for whatever reason the patches I found didn't touch them. So I tried doing for jemalloc.c what was already done for Memory.cpp, and it worked (but only after I removed the map align code).
This removes speech recognition, pocketsphinx, training models
and the speech automated test interface.
This also re-establishes proper use of MOZ_WEBSPEECH to work
for the speech API (synthesis part only) that was a broken mess
before, with some synth parts being always built, some parts
being built only with it enabled and recognition parts being
dependent on it. I'm pretty sure it'd be totally busted if you'd
ever have tried building without MOZ_WEBPEECH before.
Tested that synthesis still works as-intended.
This resolves#1538
Excludes layers/ipc/CompositorBridgeParent and CrossProcessCompositorBridgeParent
because the latter purposely uses functions implemented in the former.
See also code comment in CrossProcessCompositorBridgeParent.cpp
Note: likely dead code for us anyway; likely not worth investing in to solve.
With de-unified building we have to re-apply the fix-up
for <cstdio>, and additionally the same for <cstring>
to prevent the compiler picking the wrong versions of
standard functions.
Before issue #146, border-radius on row groups, rows, column groups, or columns don't apply to the background of each cell, yet the border-radius on the cell itself does.
After issue #146, the behaviors changed. In this patch, I tried to revert the behaviors of border-radius on table row groups, rows, column groups, or columns back to what happened before issue #146.
Also: Don't override GetBorderRadii in nsBCTableCellFrame.
Part 1: Remove current table item, as it's never set.
Part 2: Get rid of generic table painting code, and handle each class separately.
Part 4: Hoist outline skipping into col(group) frame code.
Part 5: Skip box-shadow for table column and column groups.
Part 6: Store column and column group backgrounds separately, and then append them before the rest of the table contents.
Part 7: Pass rects in display list coordinates to AppendBackgroundItemsToTop.
Part 8: Create column and column group background display items as part of the cell's BuildDisplayList.
Part 9: Used cached values instead of calling nsDisplayListBuilder::ToReferenceFrame when possible, since it can be expensive when the requested frame isn't the builder's current frame.
Part 10: Make sure we build display items for table parts where only the normal position is visible, since we may need to create background items for ancestors at that position.
Part 11: Create an AutoBuildingDisplayList when we create background items for table columns and column groups, so that we initialize the invalidation state correctly.
In the printing preview, we create continuous table frame if table is too long to containing in a page. But the default value of NeedToCalcHasBCBorders is false which means we don't calculate HasBCBorders for continuous table frame. Thus, the border collapse is not shown when printing preview.
Bug 1441796 - Part 1: Optimize the (pseudo)-stacking context conditions
Bug 1441796 - Part 3: Reuse the results in nsIFrame::BuildDisplayListForStackingContext() for ChildrenHavePerspective(), IsTransformed(), and Combines3DTransformWithAncestors()
Bug 1512244 - Part 1: Cleanup DescendIntoChild
This reverts m-c Bug 668378 and completely removes Telemetry SQLite IO. As a bonus this fixes a potential crash in newer SQLite versions without the need for updating this useless telemetry shim.
This should ideally not be needed on Mac but compilation fails due to
sizeof `DxgiAdapterDesc` not known.
Didn't face this issue on Linux though, probably due to a different include
chain which already has this file included somewhere along the way.
This de-unifies up to and including dom/media/platforms with the exception of
dom/media/platforms/wmf, because that one throws weird errors in the Win
SDK which don't have a clear cause (yet)
This file not only somehow escaped my pass through here but was also mocking me by using AppConstants in a preprocessed file BUT ALSO right AFTER a preprocessor directive.
Reference:
15:03.06 ../../build/unix/gold/ld: error: /home/mattatobin/development/.obj/trunk-master/navigator-x64-gtk3/toolkit/library/../../dom/fetch/Request.o: requires dynamic R_X86_64_PC32 reloc against '_ZN7mozilla3dom9FetchBodyINS0_7RequestEED2Ev' which may overflow at runtime; recompile with -fPIC
15:03.06 ../../build/unix/gold/ld: error: /home/mattatobin/development/.obj/trunk-master/navigator-x64-gtk3/toolkit/library/../../dom/fetch/Response.o: requires dynamic R_X86_64_PC32 reloc against '_ZN7mozilla3dom9FetchBodyINS0_8ResponseEED2Ev' which may overflow at runtime; recompile with -fPIC
15:03.06 ../../build/unix/gold/ld: error: read-only segment has dynamic relocations
15:03.07 /home/mattatobin/development/.obj/trunk-master/navigator-x64-gtk3/toolkit/library/../../dom/fetch/Request.o:Request.cpp:function mozilla::dom::Request::~Request(): error: undefined reference to 'mozilla::dom::FetchBody<mozilla::dom::Request>::~FetchBody()'
15:03.07 /home/mattatobin/development/.obj/trunk-master/navigator-x64-gtk3/toolkit/library/../../dom/fetch/Response.o:Response.cpp:function mozilla::dom::Response::~Response(): error: undefined reference to 'mozilla::dom::FetchBody<mozilla::dom::Response>::~FetchBody()'
Running into the same problem as with dom/canvas: derived template
classes not being happy about being non-unified. Kept unified for now
to prevent linking errors.
KeyFrameEffect and KeyFrameEffectReadOnly are still built unified as
buddies because they have some complex interdependency that couldn't
easily be resolved without linking failures.
IsSelectionEditable should check whether the focus node and anchor node
aren't null before trying to use them.
This also changes the initialization of selections' aOutIndex to the
last range in the selection as a fallback in case we don't add a range
later (in AddItem) which could also end up with a null selection
otherwise if the additional selection nodes are removed.
To respond dynamically to the pref change, the mAppVersion string needs
to be rebuilt.
Includes some minor improvements for corner cases and removes leftover
b2g junk.
Since we are using vsync for timing anyway in the layout refresh driver,
there is no need to bump the system timer resolution down to 1 ms.
This resolves#1507.
Correct various pre-processor defines for sparc64 and in mozjemalloc use the JS arm64 allocator on Linux/sparc64.
This corrects build problems opn Linux sparc64 and is in line with bugzilla bug #1275204.
- Adds buildconfig option --enable-nss-sqlstore
- Prefixes NSS dbinit with either sql: or dbm: depending on config
- Pre-initializes mozStorage when NSS-SQL storage is used to prevent
an sqlite3_config race in NSS Init
whether they should be drawn.
This reduces the size of display lists for tables by only adding display
list items that are actually going to be visibly drawn, which will help
overall performance of table drawing.
There were actually two separate logical errors in this method:
The first part is that "origSizes.isSome()" is simply a bogus
requirement for applying min/max-sizes here. I'm still keeping
the optimization of not needlessly copying the mSizes array
(as originally intended) since it's a quite common case.
The second bug is that min/max-sizes were only applied under
the "if (fr != 0.0f)" block. This is bogus since the calculated
'fr' value depends on 'aAvailableSize' which might change by
applying min/max-sizes and thus 'fr' could become non-zero in
the second round.
To fix, this patch just moves "applyMinMax" block out one level.
* So far this seems to only affect Macintosh builds but other unix builds may be affected. Windows doesn't seem to be.
* This maintains consistency with other one-off file invocations and backwards compatibility with older comm setups
When setting up connectivity, e.g. with new network detection, the pref
wouldn't be checked (or honored), resulting in spurious startup calls
despite being preffed off. This fixes that.
Resolves#1474
Part 1: Remove current table item, as it's never set.
Part 2: Get rid of generic table painting code, and handle each class separately.
Part 4: Hoist outline skipping into col(group) frame code.
Part 5: Skip box-shadow for table column and column groups.
Part 6: Store column and column group backgrounds separately, and then append them before the rest of the table contents.
Part 7: Pass rects in display list coordinates to AppendBackgroundItemsToTop.
Part 8: Create column and column group background display items as part of the cell's BuildDisplayList.
Part 9: Used cached values instead of calling nsDisplayListBuilder::ToReferenceFrame when possible, since it can be expensive when the requested frame isn't the builder's current frame.
Part 10: Make sure we build display items for table parts where only the normal position is visible, since we may need to create background items for ancestors at that position.
Part 11: Create an AutoBuildingDisplayList when we create background items for table columns and column groups, so that we initialize the invalidation state correctly.
This should be the last code backout for this. merging this branch
should get us back to the way we were (+ additional code changes for
later changes) as fasr as the unused unboxed code is concerned.
This adds an addition to the environment set up for child processes
(plugin container) so that it may still be able to pass the omni
parameters there as-needed.
This stub is added because websites insist on considering this
very hardware-dependent and O.S.-variable low-level font-control
as a "critical feature" which it isn't as there is 0 guarantee
that font variation settings are supported or honored by any
operating system used by the client.
On top this is a WD status feature that sites shouldn't be using, and
the feature itself is strongly discouraged for use in favor of standard
CSS font manipulation keywords like `font-weight`.
This is a hack to deal UnifiedBindings trying to deal with an incomplete codegen implementation so it was worked around with a hack back in Firefox 30-something. If we have anymore of this non-sense crop up as stuff is added or removed from DOM then extend this conditional.
Least until something better comes along.
Do less work in PaintRowGroupBackgroundByColIdx for cells that are not in our desired set of columns.
crash fix:
Guard against empty column groups when building a display list for a table.
If EME isn't built, these functions would have no use and the callback
structures are unavailable. Stubbing these out will still allow the
callers to work unchanged, but in effect simply always return without
dispatching a promised task.
This fixes the dependency issue from part 2a.
HTML standardizes proper behavior of colSpan and rowSpan:
The main thing is that getting the .rowSpan and .colSpan IDL properties will now return the actual clamped value that we use.
There is flexibility in exactly the value the initialized length must
hold, i.e. if an array is completely empty, it is valid for the
initialized length to be any value between zero and the length of the
array, as long as the in-memory values below the initialized length have
been initialized with a hole value.
In the case of 0, the array is packed and the move operation would be a
nop, so simply convert the assert to a condition to save some cycles.
This was added as part of the WebExtension sec issue CVE-2017-5389.
Since we no longer have this interface, it needed to be removed to
prevent errors in the WebRequest channel callback.
This gets rid of platform-dependent hard-coded defaults, but keeps
build-time blocking if there is no GL provider (in which case layers
acceleration almost certainly won't work because it needs a GL
compositor and would likely crash without)
New prefs are
- layers.acceleration.enabled to enable HWA
- layers.acceleration.force to force it enabled (requires .enabled to be
set as well)
This is the platform part of this issue. The rest will be front-end work
(Preference UI integration and pref migration)
2020-04-14 12:14:11 +02:00
47 changed files with 415 additions and 197 deletions
Blocking a user prevents them from interacting with repositories, such as opening or commenting on pull requests or issues. Learn more about blocking a user.
Moonchild
Paul Adenot
Yaron Tausky
Steve Fink
Olli Pettay
Jan de Mooij
Matt A. Tobin
FranklinDM
J.C. Jones
Sohaib ul Hassan
trav90
Brian Smith
Werner Lemberg
Henri Sivonen
David Major
Alessio Vanni
adesh
athenian200
Gaming4JC
adeshkp
Kris Maglione
Jon Coppeard
Andy
Michael Tuexen
Andrea Marchesini
Byron Campen [:bwc]
wolfbeast
Lootyhoof
Fedor
Ascrod
Trushita
Cameron McCormack
Jiaxun Yang
win7-7
Bas Schouten
JustOff
athenian200
JMadgwick
Simon Giesecke
sotaro