We currently have 2 virtual methods to inform DOM::Element subclasses
when an attribute has changed, one of which is spec-compliant. This
patch removes the non-compliant variant.
Instead, smuggle it in as a `void*` private data and let Javascript
aware code cast out that pointer to a VM&.
In order to make this split, rename JS::Cell to JS::CellImpl. Once we
have a LibGC, this will become GC::Cell. CellImpl then has no specific
knowledge of the VM& and Realm&. That knowledge is instead put into
JS::Cell, which inherits from CellImpl. JS::Cell is responsible for
JavaScript's realm initialization, as well as converting of the void*
private data to what it knows should be the VM&.
NanBoxedValue is intended to be a GC-allocatable type which is not
specific to javascript, towards the effort of factoring out the GC
implementation from LibJS.
Attempt 2! Reverts 2a5dbedad4
This time, set up a different combinator when producing a relative
invalid selector rather than a standalone one. This fixes the crash.
Original description below for simplicity because it still applies.
---
Selectors like `:is(.valid, &!?!?!invalid)` need to keep the invalid
part around, even though it will never match, for a couple of reasons:
- Serialization needs to include them
- For nesting, we care if a `&` appeared anywhere in the selector, even
in an invalid part.
So this patch introduces an `Invalid` simple selector type, which simply
holds its original ComponentValues. We search through these looking for
`&`, and we dump them out directly when asked to serialize.
This gets rid of a couple FIXMEs and allows reusing the logic of
validating this field between different algorithms. While we're here,
expand its logic to match the constraints as outlined in RFC 7517.
LibWebView now knows how to launch RequestServer and ImageDecoderServer
without help from the UI, so let's move ownership of these services over
to LibWebView for de-duplication.
This was a silly mistake on my end and percentages values are not
covered by device-independent color space, so I had to add support for
srgb to run a WPT test that made me realize the mistake.
This makes the following test pass:
- css/css-color/predefined-002.html
It makes the following WPT tests pass:
- css/css-color/predefined-001.html
- css/css-color/xyz-003.html
- css/css-color/xyz-d50-003.html
- css/css-color/xyz-d50-004.html
- css/css-color/xyz-d65-003.html
Also we now render the reference of color-mix-currentcolor-nested-for-
color-property.html properly. Which means that it's now different from
the actual test, that is still rendered incorrectly. In other word, the
false positive for this test is now turned into a true negative.
Now that the heap has no knowledge about a JavaScript realm and is
purely for managing the memory of the heap, it does not make sense
to name this function to say that it is a non-realm variant.
The main motivation behind this is to remove JS specifics of the Realm
from the implementation of the Heap.
As a side effect of this change, this is a bit nicer to read than the
previous approach, and in my opinion, also makes it a little more clear
that this method is specific to a JavaScript Realm.
Selectors like `:is(.valid, &!?!?!invalid)` need to keep the invalid
part around, even though it will never match, for a couple of reasons:
- Serialization needs to include them
- For nesting, we care if a `&` appeared anywhere in the selector, even
in an invalid part.
So this patch introduces an `Invalid` simple selector type, which simply
holds its original ComponentValues. We search through these looking for
`&`, and we dump them out directly when asked to serialize.
We will want to re-inform WebContent of the system visibility state when
we create a new process after a crash. This changes the IPC to just send
the enum value directly, instead of a boolean, so that we can just store
that enum value directly on the ViewImplementation class.
These operations should still apply even if they are off screen, because
they affect painting of things outside of their bounding rectangles.
This commit makes us always apply these, regardless of if they are in
the visible region. However, if they are outside that region, we
replace them with simple clip-rect commands, which have the same
effect (not painting anything) but are cheaper than computing a full
mask bitmap.
The insertion steps for iframes were following an old version of the
spec, where it was checking if the iframe was "in a document tree",
which doesn't cross shadow root boundaries. The spec has since been
updated to check the shadow including root instead.
This is now needed for Cloudflare Turnstile iframe widgets to appear,
as they are now inserted into a shadow root.
Previously, the inclusive descendant, which is the node that
for_each_shadow_including_inclusive_descendant was called on, would not
have it's shadow root traversed if it had one.
This is because the shadow root traversal was in the `for` loop, which
begins with the node's first child. The fix here is to move the shadow
root traversal outside of the loop, and check if the current node is an
element instead.
While this does mean that we keep one copy of the stack info in the VM,
and another in the Heap; keeping a separate instance removes one more
instance of coupling between the heap and LibJS specific details.
There is definitely a possibility I am misunderstanding the reason
behind it - but this does not appear neccessary. The VM owns both the
string cache and Heap. On destruction, the VM should clear out both
the heap and its string cache.
While this is used in the implementation of Runtime objects itself, Heap
seems like a more appropriate home. This will also help in factoring out
the GC implementation into it's own library as the heap explicitly has
knowledge of WeakContainer.
When the cached value was not an accessor, it was simply ignored.
This is the value we really want, so we can just return it.
Shows up to 5x improvements on some benchmarks,
and 1.4x in general js-benchmarks.
Async functions whose promise is never resolved were leaking, since they
had a strong root JS::Handle on themselves.
This doesn't appear to actually be necessary, since the wrapper will be
kept alive as long as it's reachable (and if it's not reachable, nobody
is going to resolve/reject the promise either).
This fixes the vast majority of leaks on Speedometer, bringing memory
usage at the end of a full run from ~12 GiB to ~3 GiB.
This fixes an issue where a badly-timed garbage collection could swallow
a static field initializer.
Caught by running test262 in GC-on-every-allocation mode.
We were miscalculating the length of the buffer in pointer-sized chunks,
which is what the conservative root scan cares about.
This could cause some values to be prematurely garbage-collected.
This was preventing https://ubereats.com/ from fully loading, because
they are attempting to overwrite setItem. They seem to be trying to add
error logging to setItem if it throws, as all they do is add a
try/catch block that emits an error log to their monitoring service if
it throws.
However, because Storage is a legacy platform object with a named
property setter (setItem), it will call setItem with the stringified
version of the function. This is actually expected as per the spec,
Firefox (Gecko) and Epiphany (WebKit) does this too, but Chromium does
not as it actually overwrites the function with the new function and
does not store the stringified function.
The problem is that we had the LegacyOverrideBuiltIns flag accidentally
set, so it would return the stored string instead of the built-in
function (hence the name), then it would try and call it and throw a
"not a function" error. This prevented their JS from going any further.
This fix allows their UI to fully load and be fully interactive, though
it is quite slow at the moment!
This change removes the append_without_space, append_with_space,
prepend_without_space, and prepend_with_space functions from DOM::Node.
All those methods were added with the initial “Implement Accessible Name
and Description Calculation” commit in da5c918 and were only used in the
code related to accessible-name computation. But subsequent changes to
that code have removed all the calls to those functions — so now they’re
all completely unused.
This change ensures that when the aria-labelledby attribute is used, the
expected text from the element referenced in the aria-labelledby value
appears in the computed accessible name. Otherwise, without this change,
the expected text doesn’t appear in the computed accessible name.
This change fixes handling for substep ii of the “F. Name From Content”
step at https://w3c.github.io/accname/#step2F in the “Accessible Name
and Description Computation” spec — to correctly include any ::before
and ::after pseudo-element content in the computation of accessible
names. Otherwise, without this change, accessible names unexpectedly
don’t include that pseudo-element content.
This change implements the https://w3c.github.io/accname/#comp_append
step in the “Accessible Name and Description Computation” spec — so that
when an accessible name is computed from multiple sources in a document
subtree, the parts of the computed text are joined together with spaces.
Otherwise without this change, in accessible names computed from
multiple sources in a document subtree, the parts of the computed text
are unexpectedly run together, with no spaces between the parts.
This is really just a type alias for NonnullGCPtr<T>, but it provides
a way to have non-owning non-visited NonnullGCPtr<T> without getting
yelled at by the Clang plugin for catching GC errors.
We were previously dumping the address of the cell pointer instead of
the address of the cell itself. This was causing mysterious orphans
in GC dumps, and it took me way too long to figure this out.
compute_inset() was incorrectly retrieving the containing block size
because containing_block() is unaware of grid areas that form a
containing block for grid items but do not exist in the layout tree.
With this change, we explicitly pass the containing block into
compute_inset(), allowing it to correctly provide the containing block
sizes for grid items.
Explicitly pass containing block width in
resolve_vertical_box_model_metrics() instead of doing containing block
box lookup.
This is a part of refactoring towards removing containing_block() usage
that will allow us introduce partial layout.
The following syntax is valid:
```js
e?.example / 1.2
```
Previously, the `/` would be treated as a unterminated regex literal,
because it was calling the regular `consume` instead of
`consume_and_allow_division`.
This is what is done when parsing IdentifierNames in
parse_secondary_expression when a period is encountered.
Allows us to parse clients-main-[hash].js on https://ubereats.com/
We currently compile the Qt event loop files multiple times, for every
target which wants to use them. This patch moves these to LibWebView as
a central location to avoid this.
If available space is definite it should always match the size of the
containing block. Therefore, there is no need to do containing block
node lookup.
The inline capacity on ThreadEventQueue::Private::queued_events caused
us to reserve (and importantly, not initialize!) 2 KiB of stack memory
when entering ThreadEventQueue::process().
This was causing any leftover pointers to GC-allocated objects within
that memory range to keep those objects alive, even when all other
references were gone.
Problem:
- Many constructors are defined as `{}` rather than using the ` =
default` compiler-provided constructor.
- Some types provide an implicit conversion operator from `nullptr_t`
instead of requiring the caller to default construct. This violates
the C++ Core Guidelines suggestion to declare single-argument
constructors explicit
(https://isocpp.github.io/CppCoreGuidelines/CppCoreGuidelines#c46-by-default-declare-single-argument-constructors-explicit).
Solution:
- Change default constructors to use the compiler-provided default
constructor.
- Remove implicit conversion operators from `nullptr_t` and change
usage to enforce type consistency without conversion.
These changes are arbitrarily divided into multiple commits to make it
easier to find potentially introduced bugs with git bisect.
The modifications in this commit were automatically made using the
following command:
find . -name '*.h' -exec sed -i -E 's/dbg\(\) << ("[^"{]*");/dbgln\(\1\);/' {} \;
- Store history entries as (timestamp)::(entry)\n\n
- Merge the entries together when saving to avoid loss of history
entries
To ideally make having two concurrently open shells
(or `js` repls or whatever) not overwrite each others' history entries.
Problem:
- The implementation of `find` is coupled to the implementation of `Vector`.
- `Vector::find` takes the predicate by value which might be expensive.
Solution:
- Decouple the implementation of `find` from `Vector` by using a
generic `find` algorithm.
- Change the name of `find` with a predicate to `find_if` so that a
binding reference can be used and the predicate can be forwarded to
avoid copies.
- Change all the `find(pred)` call sites to use `find_if`.
Previously, it was "relying" on the cursor blink timer to update
the visual selection.
This caused some delay after the whole line was selected by
triple-clicking specifically the last word in a line.
Now, the widget is updated after triple-clicking.
This patch merges the profiling functionality in the kernel with the
performance events mechanism. A profiler sample is now just another
perf event, rather than a dedicated thing.
Since perf events were already per-process, this now makes profiling
per-process as well.
Processes with perf events would already write out a perfcore.PID file
to the current directory on death, but since we may want to profile
a process and then let it continue running, recorded perf events can
now be accessed at any time via /proc/PID/perf_events.
This patch also adds information about process memory regions to the
perfcore JSON format. This removes the need to supply a core dump to
the Profiler app for symbolication, and so the "profiler coredump"
mechanism is removed entirely.
There's still a hard limit of 4MB worth of perf events per process,
so this is by no means a perfect final design, but it's a nice step
forward for both simplicity and stability.
Fixes#4848Fixes#4849
Yay for more spec compliance! This is pretty easy as everything using
to_size_t() should just be using one of the other abstract operations we
already have implemented.
This allows us to get rid of get_length() in ArrayPrototype, which is
basically a slightly incorrect implementation of length_of_array_like(),
and then finally remove to_size_t()!
Also fixes a couple of "argument is undefined" vs "argument isn't given"
issues along the way.
The pseudo-code from the spec says "Assert: Type(obj) is Object.", so we
can just enforce this at compile time rather than taking it literally
and doing "ASSERT(value.is_object())".
Also fix an issue where the absence of a "length" property on the object
would cause a crash (to_number() on empty value).
This adds a String.prototype.split implementation modelled after
ECMA262 specification.
Additionally, `Value::to_u32` was added as an implementation of
the standard `ToUint32` abstract operation.
There is a tiny kludge for when the separator is an empty string.
Basic tests and visiting google.com prove that this is working.
This commit adds support for inserting in a "verbatim" mode where a
single uninterpreted key is appended to the buffer.
As this allows the user to input control characters, all control
characters except \n (^M) are rendered in their caret form, with
reverse video (SGR 7) applied to it.
To not break cursor movement, the concept of "masked" characters is
introduced to the StringMetrics interface, which can be mostly ignored
by the rest of the system.
It should be noted that unlike some other line editing libraries,
LibLine does _not_ render a hard tab as a tab, but rather as '^I',
which greatly simplifies cursor handling.
Let's adapt this class a bit better to how it's actually being used.
Instead of having valid/invalid states and storing an error in case
it's invalid, a MappedFile is now always valid, and the factory
function that creates it will return an OSError if mapping fails.
This avoids unintentionally adding a newline character at the end of
user passwords when they are set using passwd(1).
I also fixed these two issues:
- The return value of getline() was being saved in an `int` variable
instead of in a `ssize_t` variable; I replaced the `int` keyword with
`auto` to fix this issue.
- Prior to this patch, get_password() could potentially return
tcsetattr()'s errno instead of getline()'s errno in case of an error.
We now make sure it always returns the right errno in case of an error.
This was very obviously racy and would only succeed if we already own
the socket anyway. (And if we do, we can bind to it without unlinking!)
Work towards #4876.
This doesn't solve half of the problems with /tmp/rpc, but this way we
can at least make it sticky instead of having it fully world-writable
and owned by whoever was the first to bind an RPC socket.
This ioctl can fail if we're resizing the terminal right when the shell
inside it has exited. Instead of throwing up a crash reporter, whine a
little bit in the debug log and exit cleanly moments later.
This fixes an issue that shows up as a nice crash when "^R<enter>^C",
which is actually the event loop trying to call into a deleted object
(the search editor).
Move some more complex globals into a Singleton, which allows it being
used from global destructors. It solves problems where some global
variables, such as HashMaps may already be deleted, triggering crashes
trying to use them.
This allows adding and removing of asynchronous signal handlers while
executing signal handlers, even if it is for the same signal that is
being handled right now.
These changes are arbitrarily divided into multiple commits to make it
easier to find potentially introduced bugs with git bisect.Everything:
The modifications in this commit were automatically made using the
following command:
find . -name '*.cpp' -exec sed -i -E 's/dbg\(\) << ("[^"{]*");/dbgln\(\1\);/' {} \;
This patch moves the user account password hashes from /etc/passwd,
where they were world-readable, to /etc/shadow, where only root can
access them.
The Core::Account class is extended to support both authentication
against, and modification of /etc/shadow.
The default password for "anon" as of this commit is "foo" :^)
This was missing a "toInt32()" which returns 0 for NaN and Infinity.
From the spec:
6.1.6.1.2 Number::bitwiseNOT ( x )
The abstract operation Number::bitwiseNOT takes argument x (a Number).
It performs the following steps when called:
Let oldValue be ! ToInt32(x).
Return the result of applying bitwise complement to oldValue.
The mathematical value of the result is exactly representable as
a 32-bit two's complement bit string.
Fixes#4868.
Instead of doing a forced layout synchronously whenever an element's
style is changed, use a zero-timer to do the forced relayout on next
event loop iteration.
This effectively coalesces a lot of layouts and makes many pages such
as GitHub spend way less time doing redundant layout work.
Previously the client would only learn the mime type of what was being
dropped on it once the drop occurred. To enable more sophisticated
filtering of drag & drop, we now pass along the list of mime types being
dragged to the client with each MouseMove event.
(Note that MouseMove is translated to the various Drag* events in LibGUI
on the client side.)
If a widget accept()'s a "drag enter" event, that widget now becomes
the application-wide "pending drop" widget. That state is cleared if
the drag moves over another widget (or leaves the window entirely.)
We were using the "accept" flag on the event to break out of the
bubbling loop, but this had lasting consequences since all events that
bubbled too far came out looking as if someone had accepted them.
If an event is ignored by everyone, it should appear ignored.
These events allow widgets to react when a drag enters/leaves their
rectangle. The enter event carries position + mime type, while the
leave event has no information.
AT_SECURE is set in the auxiliary vector when we execute setuid/setgid
programs.
In those cases, we do not want to read environment variables that
influence the logic of the dynamic loader, as they can be controlled
by the user.
Previously, when trying to parse the location info of a member
variable, we asserted that the location info of its parent is of type
'Address'.
However, there are cases where we cannot compute the location info of
the parent (for example - because we do not yet support the type of
debug info generated for it).
In those cases, it is better to just leave the location info of the
member variable empty instead of crashing.
DebugSession now makes the loader stop after loading the libraries,
and parses the loaded libraries of the program before continuing its
execution.
DebugSession now also supports inserting a breakpoint at a given symbol
or source position.
Additionally, DebugInfo now takes the base address of its object into
consideration.
If set, the dynamic loader will perform a software breakpoint after
loading all libraries, and just before jumping to the main entry point.
This allows a debugger to inspect the loaded libraries before the
program starts executing.
This patch implements the "remove irrelevant boxes" and "generate
missing child wrappers" parts of table fixup.
"Generate missing parents" is left as a task for our future selves.
It seems like both BFC and IFC can have absolutely positioned children.
It's a bit strange, but consider the following HTML:
<html><body>foobar<img style="position: absolute"></body></html>
In such a document, the <img> element is an absolutely positioned child
of a block-level element (<body>) with no block-level children.
An IFC is established for <body>, and needs to handle layout for <img>.
Various whitespace-related issues have been fixed, and support for the
different CSS white-space values is improved enough that I think we can
stop doing the hack where we just prune whitespace nodes from the tree
and actually let them show up.
This is a nice step forward for correctness with the slight downside of
cluttering up layout tree dumps with tons of whitespace text nodes.
But hey, that's the web we know & love. :^)
Fixes#4427.
The StyleResolver can find the specified CSS values for the parent
element via the DOM. Forcing everyone to locate specified values for
their parent was completely unnecessary.
Layout nodes now only carry CSS computer values with them. The main
idea here is to give them only what they need to perform layout, and
leave the rest back in the DOM.
Put all the inherited members in one struct and all the non-inherited
ones in another.
This makes it clear which is which, and also makes it easy to copy all
the inherited values while ignoring the non-inherited ones.
Another step towards not having to carry the full specified style with
us everywhere. This isn't the ideal final layout, since we're mixing
computed and used values a bit randomly here, but one step at a time.
We shouldn't reject indexed palette PNGs just because they have fewer
palette entries than the bit depth allows. Instead, we need to check
for OOB palette accesses and fail the decode *then*.
When changing the mouse cursor (e.g when hovering over a link) we now
only change the InProcessWebView's override cursor instead of setting
the cursor at the window level.
This fixes an issue where the I-beam or hand cursors would somehow
"escape" from the web view and over to other widgets.
This patch adds a global (per-process) filter list to LibWeb that is
used to filter all outgoing resource load requests.
Basically we check the URL against a list of filter patterns and if
it's a match for any one of them, we immediately fail the load.
The filter list is a simple text file:
~/.config/BrowserContentFilters.txt
It's one filter per line and they are simple glob filters for now,
with implicit asterisks (*) at the start and end of the line.
