This drastically reduces the cost (in time AND space) of allocating
a Gfx::Bitmap.
Anything that needs to be shared is already using Core::AnonymousBuffer
anyway, so this shouldn't break anything important.
Do note that this makes it an error to create an empty (0x0) Bitmap,
which was previously allowed for some reason. Some small tweaks are
included to bail gracefully in such scenarios.
Previously, we would apply any adopted style sheet to the document if
its alternate flag was not set. This meant that all adopted style
sheets would be applied, since constructed style sheets never have this
flag set.
Changes compute_absolute_padding_rect_with_css_transform_applied() to
use cached absolute rect and CSS transform instead of doing expensive
containing block chain traversal.
Reduces refresh_clip_state() from 4% to 2% in Discord profiles.
This allows main UI processes created while there is a currently
running one to request a new tab or a new window with the initial urls
provided on the command line. This matches (almost) the behavior of
Chromium and Firefox.
Add a new IPC protocol between two UI processes. The main UI process
will create an IPC server socket, while secondary UI processes will
connect to that socket and send over the URLs and action it wants the
main process to take.
...and add a test case that shows why it's incorrect.
If one dimension is 2^n + 1 and the other side is just 1, then the
topmost node will have 2^n x 1 and 1 x 1 children. The first child will
have n levels of children. The 1 x 1 child could end immediately, or it
could require that it also has n levels of (all 1 x 1) children. The
spec isn't clear on which of the two alternatives should happen. We
currently have n levels of 1 x 1 blocks.
This test case shows that a VERIFY we had was incorrect, so remove it.
The alternative implementation is to keep the VERIFY and to add a
if (x_count == 1 && y_count == 1)
level = 0;
to the top of TagTreeNode::create(). Then we don't have multiple levels
of 1 x 1 nodes, and we need to read fewer bits.
The images in the spec suggest that all nodes should have the same
number of levels, so go with that interpretation for now. Once we can
actually decode images, we'll hopefully see which of the two
interpretations is correct.
(The removed VERIFY() is hit when decoding
Tests/LibGfx/test-inputs/jpeg2000/buggie-gray.jpf in a local branch that
has some image decoding implemented. That file contains a packet with
1x3 code-blocks, which hits this case.)
The main difference was that our implementation was writing
the final line of a series of repeated lines, whereas the
spec says "The second and succeeding copies of repeated adjacent
input lines shall not be written."
Additionally, there was a mistake in the -f flag implementation
causing the number of fields skipped to be one greater than
required.
Flags that rely on counting lines (-c and -d) were
producing results that were off by one. This is fixed
by initializing the `count` variable to 1, which is
consistent with behavior in the main loop, where it
is reset to 1 when lines don't match.
Calls to `read_line` are replaced with `read_line_with_resize`
and `swap`s of StringViews, which assume a consistent location
of the underlying ByteBuffers, are replaced. A test file has
been added for uniq, which includes a test case for long lines.
This was resulting in a whole lot of rebuilding whenever a new IDL
interface was added.
Instead, just directly include the prototype in every C++ file which
needs it. While we only really need a forward declaration in each cpp
file; including the full prototype header (which itself only includes
LibJS/Object.h, which is already transitively brought in by
PlatformObject) - it seems like a small price to pay compared to what
feels like a full rebuild of LibWeb whenever a new IDL file is added.
Given all of these includes are only needed for the ::initialize
method, there is probably a smart way of avoiding this problem
altogether. I've considered both using some macro trickery or generating
these functions somehow instead.
Before this change we were recording and executing sample/blit commands
for each painting phase, even if there are no painting commands
in-between sample and blit that produce result visible on a canvas.
This change adds an optimization pass that goes through recorded
painting commands list and marks sample and blit commands that could
be skipped.
Reduces sample and blit corners executing from 17% to 8% on Discord.
For this case to work correctly in the current bytecode world:
func(a, a++)
We have to put the function arguments in temporaries instead of allowing
the postfix increment to modify `a` in place.
This fixes a problem where jQuery.each() would skip over items.
We don't need `file_actions` to be a constant-reference. It's created
in-place by its one user (HackStudio). Because it is currently a const-
ref, if we try to create a ProcessSpawnOptions like so:
Core::ProcessSpawnOptions options { .name = "foo"sv };
We get the following error with clang 18:
lifetime extension of temporary created by aggregate initialization
using a default member initializer is not yet supported; lifetime of
temporary will end at the end of the full-expression
Introduces separate layout and paintable type for foreign element.
It is necessary to inherit SVGForeignObjectPaintable from SVGMaskable
in upcoming changes.
Preparation work before adding support for SVGForeignObjectElement
masking.
Having a mixin makes possible sharing mask calculation between
SVGForeignObjectElement's paintable and SVGGraphicsPaintable.
Both has to support masking:
- PaintableWithLines (from SVGForeignObjectElement) -> PaintableBox
- SVGGraphicsPaintable -> SVGPaintable -> PaintableBox
After this change it will be possible to introduce a new paintable type
for foreignObject that inherits from both PaintableWithLines and
SVGMaskable.
Fetched bodies can be on the order of gigabytes, so rather than crashing
when we hit OOM here, we can simply invoke the error callback with a DOM
exception. We use "UnknownError" here as the spec directly supports this
for OOM errors:
UnknownError: The operation failed for an unknown transient reason
(e.g. out of memory).
This is still an ad-hoc implementation. We should be using streams, and
we do have the AOs available to do so. But they need to be massaged to
be compatible with callers of Body::fully_read. And once we do use
streams, this function will become infallible - so making it infallible
here is at least a step in the right direction.
Previously, [Global] interfaces were not excluded from the
`internal_own_property_keys()` call. This caused a crash when iterating
over the properties of the Window object.
Fixes bug when CSS transform is applied twice to clip rect:
- While calculating absolute clip rectangles in `refresh_clip_state()`
- While executing `PushStackingContext` painting command.
Duplicated transform is already removed for PaintableBox and this change
adds this for InlinePaintable.
This change moves WebAssembly related data that was previously globally
accessible into the `WebAssemblyCache` object and creates one of these
per global object. This ensures that WebAssembly data cannot be
accessed across realms.
Previously we would look for a matching ID, and then for a matching
name. If there was an element in the collection which had a matching ID
as well as an element with a matching name, we would always return the
element with a matching ID irrespective of what order that element was
in.
Link elements that aren't "browsing-context connected" should not
trigger a resource fetch when their attributes change.
This fixes an issue where we'd waste time by loading every style sheet
twice! :^)
Otherwise, the thread will continue to run and access the media data
buffer, which will have been freed.
The test here is a bit strange, but the issue would only consistently
repro after several GC runs.
The spec doesn't tell us the exact value to use, but a minumum & maximum
range of supported values. Just to be consistent with another browser,
we follow the values that firefox appears to support from testing the
interface on my machine.
This function will be used in the AudioBuffer constructor, but is
defined in the spec as part of BaseAudioContext.
Fixes crashing after following steps:
1. Open https://github.com/SerenityOS/serenity
2. Click on "Pull requests" tab
The problem was `navigable` null pointer dereferencing in
`decode_favicon()`. But navigable is null because the document was
created by `parseFromString()` DOMParser API.
With this change we skip fetching initiated by HTMLLinkElement if
document does not have a browsing context:
- Favicon is not displayed for such documents so no need to fetch.
- Stylesheets fetching won't affect such document because style or
layout does not run for them.
This largely adapts the code from SingletonProcess.cpp to work a bit
closer with Core::Process. Ideally, we'll move the daemonizing feature
into Core::Process::disown() eventually.
We implement the move constructor already. A future commit will have
code of the form:
process = move(other_process);
which there is no reason to forbid.
The following command was used to clang-format these files:
clang-format-18 -i $(find . \
-not \( -path "./\.*" -prune \) \
-not \( -path "./Base/*" -prune \) \
-not \( -path "./Build/*" -prune \) \
-not \( -path "./Toolchain/*" -prune \) \
-not \( -path "./Ports/*" -prune \) \
-type f -name "*.cpp" -o -name "*.mm" -o -name "*.h")
There are a couple of weird cases where clang-format now thinks that a
pointer access in an initializer list, e.g. `m_member(ptr->foo)`, is a
lambda return statement, and it puts spaces around the `->`.
The only subclass was already GC-allocated, so let's hoist the JS::Cell
inheritance up one level. This ends up simplifying a bit of rather
dubious looking code where we were previously slicing ESOs.
This allows to skip iterating through all allocated blocks in
`find_min_and_max_block_addresses()`.
With this change `collect_garbage()` in profiles of Discord goes down
from 17% to 8%.
For `AK_OS_SERENITY`, the root path of the resources folder is "/res";
but otherwise it should be the `s_serenity_resource_root` variable set
in `platform_init()`.
However, a path provided on the command line, will override the default
path in both of those cases.
This change also makes sure that `RequestServer` can find the
certificates file `serenity/Build/lagom/share/Lagom/ladybird/cacert.pem`
Contrary to IEEE formats, x86's 80-bit extended floats need the topmost
mantissa bit to be 1 or else a special kind of NaN is produced. We fix
this by reinserting the special bit if necessary.
Whenever the floating-point values are in integer range, we can use the
various FCVT functions with static rounding mode to perform fast
rounding. I took this opportunity to clean up the architecture
differentiation for these functions, which allows us to use the software
rounding implementation for all extreme and unimplemented cases,
including AArch64.
Also adds more round & trunc tests.
Most of these now just await the image decoding, equivalent (ish) to
the old behavior. A more async-aware refactor should happen some time
in the future.
The HTMLLinkElement caller is a bit hairy, so we shove an await() in
there temporarily. This is sure to cause fun times for anyone debugging
task/microtask execution order.
ImageDecoder now queues up image decoding requests and returns the
images back to the caller later. ImageDecoderClient has a new
promise-based API that allows callers to attach their own resolve/reject
handlers to the responses from ImageDecoder.
Weak symbols can be left unresolved, and code is supposed to check if
they're null at runtime before using them. Handle them appropriately
instead of refusing to load.
There's no need to capture things as Handle when using HeapFunction.
In this case, it was even creating a strong reference cycle, which ended
up leaking.
It previously resided in LibWebView to hide the details of launching a
singleton process. That functionality now lives in LibCore. By moving
this to Ladybird, we will be able to register the process with the task
manager.
This just moves the code to launch a single process such as SQLServer to
LibCore. This will allow re-using this feature for other processes, and
will allow moving the launching of SQLServer to Ladybird.
This will be needed to collect statistics from processes that do not
have anything to do with LibWebView. The ProcessInfo structure must be
virtual to allow callers to add application-specific information.
This lets us remove the color/dimension-parsing lambdas, since these
always forwarded to the same methods. This will make it easier to later
convert those methods to take a TokenStream.
We have to unregister link element stylesheets from the old document's
StyleSheetList when moving them into a new document.
This makes it possible to load GitHub contributor graphs. :^)
The strategy of recreating an object for the second time after
DynamicLoader::map is interesting, of course, but it doesn't help
comprehensibility of the code, unfortunately.
Previously, the actual behavior of magic lookup and one described in its
commit description have not matched. Instead of being weak definitions
in a library that is always in the end of load order, the definitions
were normal ones and thus were able to override other weak definitions
in LibC. While this was consistent with how DynamicLoader resolves
ambiguity between normal and weak relocations, this is not the behavior
POSIX mandates -- we should always choose first available definition wrt
load order. To fix this problem, the patch makes sure we don't define
any of magic symbols in LibC.
In addition to this, it makes all provided magic symbols functions
(instead of objects), what renders MagicWeakSymbol class unnecessary.
When WebDriver accesses cookies, it specifically says to run:
the first step of the algorithm in RFC6265 to compute cookie-string
So we should skip subsequent steps. We already skip step 2, which sorts
the cookies, but neglected to skip step 3 to update their last access
time.
We can't rely on Element.setAttribute() in cloneNode() since that will
throw on weird attribute names. Instead, just follow the spec and copy
attributes into cloned elements verbatim.
This fixes a crash when loading the "issues" tab on GitHub repos.
They are actually sending us unintentionally broken markup, but we
should still support cloning it. :^)
This just reuses the layout to compute the bounding box rather than
implementing the full bounding box algorithm, which is likely the slower
option, but also much simpler to implement.
This is enough to fix the faces on https://zengm.com/facesjs/.
Capturing a struct that owns bunch of JS::Handle makes it very hard to
understand what keeps these handles alive in the GC-graph.
Instead let's capture only members of a struct used in the callback.
With FAT write support copying the file would show two errors because
it does not support chown and chmod and would return ENOTSUP. After
this commit these errors are ignored in that case.
Seems like a specification bug, but other browsers update url before
popstate event is fired and so should we.
Fixes back/forward navigation on GitHub.
We were inadvertently keeping all documents alive by installing a
console client for them. This patch fixes the issue by adding a
finalizer to Document, and having that be the way we detach console
clients. This breaks the cycle.
With this change, we can spam set .innerHTML in a loop and memory
usage remains stable.
Fixes#14612
currently crashes with an assertion failure in `String::repeated` if
malloc can't serve a `count * input_size` sized request, so add
`String::repeated_with_error` to propagate the error.
The C++ standard does not allow specifying the template parameters in
constructor declarations, see
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=97202#c8.
Converting constructors have a higher priority that user-defined
conversion functions; let's constrain `Gfx::Size<T>(Gfx::Size<U>)` to
only be considered when `U` is convertible to `T`. This lets us fall
back to conversion operators in the case of `UISize` -> `IntSize`, for
instance. Clang is still okay without this, but MSVC would error out
similarly: https://godbolt.org/z/PTbeYPM7s
Note that a not-yet-committed patch is required for full compilation:
https://gcc.gnu.org/bugzilla/show_bug.cgi?id=114784#c3
Changes the signature of queue_fetch_task() from AK:Function to
JS::HeapFunction to be more clear to the user of the function that this
is what it uses internally.
Changes the signature of queue_global_task() from AK:Function to
JS::HeapFunction to be more clear to the user of the function that this
is what it uses internally.
We currently only supported Variant II which is used by x86-64.
Variant I is used by both AArch64 (when using the traditional
non-TLSDESC model) and RISC-V, although with small differences.
The TLS layout for Variant I is essentially flipped. The static TLS
blocks are after the thread pointer for Variant I, while on Variant II
they are before it.
Some code using ELF TLS already worked on AArch64 and RISC-V even though
we only support Variant II. This is because only the local-exec model
directly uses TLS offsets, other models use relocations or
__tls_get_addr().
This removes the allocate_tls syscall and adds an archctl option to set
the fs_base for the current thread on x86-64, since you can't set that
register from userspace. enter_thread_context loads the fs_base for the
next thread on each context switch.
This also moves tpidr_el0 (the thread pointer register on AArch64) to
the register state, so it gets properly saved/restored on context
switches.
The userspace TLS allocation code is kept pretty similar to the original
kernel TLS code, aside from a couple of style changes.
We also have to add a new argument "tls_pointer" to
SC_create_thread_params, as we otherwise can't prevent race conditions
between setting the thread pointer register and signal handling code
that might be triggered before the thread pointer was set, which could
use TLS.
This refactor eliminates the need for a second "fd passing socket" on
Lagom, as it uses SCM_RIGHTS in the expected fashion, to send fds along
with the data of our Unix socket message.
These new methods combine send/receive with send_fd/receive_fd.
This is the 'correct' way to use SCM_RIGHTS, rather than trying to
emulate the Serenity behavior on other Unixes.
Most JPEG2000 files put the codestream in an ISOBMFF box structure
(which is useful for including metadata that's bigger than the
~65k marker segment data limit, such as large ICC profiles), but
some files just store the codestream directly, for example
https://sembiance.com/fileFormatSamples/image/jpeg2000/balloon.j2c
See https://www.iana.org/assignments/media-types/image/j2c for the
mime type.
The main motivation is to be able to use the test data in J.10 in
the spec as a test case.
Add factory functions to distinguish between when the owner of the File
wants to transfer ownership to the new IPC object (adopt) or to send a
copy of the same fd to the IPC peer (clone).
This behavior is more intuitive than the previous behavior. Previously,
an IPC::File would default to a shallow clone of the file descriptor,
only *actually* calling dup(2) for the fd when encoding or it into an
IPC MessageBuffer. Now the dup(2) for the fd is explicit in the clone_fd
factory function.
These changes are compatible with clang-format 16 and will be mandatory
when we eventually bump clang-format version. So, since there are no
real downsides, let's commit them now.
This fixes a regression on Acid3, since we are not expected to "best
effort" parse XML. The test specifically checks that you don't create an
incomplete, incorrect DOM.
This line changes padding top to align cell content to baseline:
`cell.padding_top += m_rows[cell.row_index].baseline - cell.baseline`
Which means available for distribution height `height_diff` could have
changed so it needs to be refreshed before assigning the rest of it to
padding bottom:
`cell_state.padding_bottom += height_diff;`
Fixes https://github.com/SerenityOS/serenity/issues/22032
When loading an XML resource into an iframe and the resource fails to
decode (e.g due to invalid UTF-8), we must still fire a load event.
This fixes the regression in subtest 69 of Acid3.
Navigation should not run for <object> element until it is inserted into
a document. Spec deoes not seem to explicitely say that, but that
matches other browsers behavior.
Fixes hanging after reloading in Acid3 test.
There were two things going wrong here:
- Transformed text (via CSS text-transform) was not invalidated after a
`@media` rule changed state.
- Removing the `style` attribute from an element didn't trigger a style
update.
This fixes the regression in subtest 46 of Acid3.
Fixes#21777
Some really cursed servers simply drop the TCP socket on the floor when
they're trying to close an HTTP connection going through a TLS socket.
This commit makes LibTLS tolerate these silly servers, and LibHTTP
accept their idea of "EOF == connection closed".
Fixes loading wpt.live/acid/acid3/test.html.
Note that this means TLSv12::on_ready_to_read can fire with an empty
buffer signifying EOF; one test refused this behaviour, and has been
changed in this commit.
I haven't found this being mentioned in the current spec, but that is
what chromium (atm the only other Navigation API implementer) does.
Fixes crashing when location.hash is changed subsequently more than once
using JS.
When launched with the new --enable-idl-tracing option, we now log
every call to web platform APIs declared via IDL, along with the
arguments passed.
This can be very helpful when trying to figure out what a site is
doing, especially if it's not doing what you'd expect.
When running with --log-all-js-exceptions, we will print the message
and backtrace for every single JS exception that is thrown, not just
the ones nobody caught.
This can sometimes be very helpful in debugging sites that swallow
important exceptions.
Fixes bug when "navigate forward" button in UI is disabled after
performing following steps:
1. Load page with an iframe (top step = 0, iframe step = 0)
2. Navigate iframe to different document (top step = 0, iframe step = 1)
3. Navigate back from browser UI (top step = 0, iframe step = 0)
No test because change is only observable from browser UI.
Previously, the parent CSS stylesheet, owner node and owner CSS rule
properties were not unset when removing a sheet from a StyleSheetList.
This change moves the methods for adding and removing sheets to and
from a StyleSheetList, directly into the StyleSheetList class and
ensures they are called as required by the CSSOM specification.
This reverts commit 9dbec601b0.
For KCOV to be performant (or at least not even slower) we need to
mmap the PC buffer from both user and kernel space at the same time.
You can't mmap a character device, so this change didn't make sense.
Plus even if we did invent a new method to exfiltrate the coverage
information out of the kernel, it would be incompatible with existing
kernel fuzzers. That would be kind of annoying. 🙃
A tag tree is a data structure used for deserializing JPEG2000
packet headers.
We don't use them for anything yet, except from tests.
The implementation feels a bit awkward to me, but we can always polish
it later.
The spec thankfully includes two concrete examples. The code is
correct enough to pass those -- I added them as test.
If a DOM node isn't connected, there's no need to invalidate, since it's
not going to be visible anyway. The node will be automatically inserted
if/when it becomes connected in the future.
This step assumes copyURL is modified in place, while
`URL::Parser::basic_parse()` returns it as a result.
> 6. Basic URL parse input, with copyURL as url and fragment state as
state override.
So copyURL has to be reassigned to navigate to new url on step 8:
> 8. Location-object navigate this to copyURL.