Problem:
- `BitmapView` permits changing the underlying `Bitmap`. This violates
the idea of a "view" since views are simply overlays which can
themselves change but do not change the underlying data.
Solution:
- Migrate all non-`const` member functions to Bitmap.
Problem:
- Static variables take memory and can be subject to less optimization.
- This static variable is only used in 1 place.
Solution:
- Move the variable into the function and make it non-static.
This was removed as part of the ByteBuffer changes but the allocation
optimization is still necessary at least for non-SerenityOS targets
where malloc_good_size() isn't supported or returns a small value and
causes a whole bunch of unnecessary reallocations.
As the parser now flattens out the instructions and inserts synthetic
nesting/structured instructions where needed, we can treat the whole
thing as a simple parsed bytecode stream.
This currently knows how to execute the following instructions:
- unreachable
- nop
- local.get
- local.set
- {i,f}{32,64}.const
- block
- loop
- if/else
- branch / branch_if
- i32_add
- i32_and/or/xor
- i32_ne
This also extends the 'wasm' utility to optionally execute the first
function in the module with optionally user-supplied arguments.
Problem:
- `BitmapView` permits changing the underlying `Bitmap`. This violates
the idea of a "view" since views are simply overlays which can
themselves change but do not change the underlying data.
Solution:
- Migrate all non-`const` member functions to Bitmap.
The current code is factored such that reads to the entirety of the last
byte should be dropped. This was relying on the fact that last would be
one past the end in that case. Instead of actually reading that byte
when it's completely out of bounds of the bitmask, just skip reads that
would be invalid. Add more tests to make sure that the behavior is
correct for byte aligned reads of byte aligned bitmaps.
As we removed the support of VBE modesetting that was done by GRUB early
on boot, we need to determine if we can modeset the resolution with our
drivers, and if not, we should enable text mode and ensure that
SystemServer knows about it too.
Also, SystemServer should first check if there's a framebuffer device
node, which is an indication that text mode was not even if it was
requested. Then, if it doesn't find it, it should check what boot_mode
argument the user specified (in case it's self-test). This way if we
try to use bochs-display device (which is not VGA compatible) and
request a text mode, it will not honor the request and will continue
with graphical mode.
Also try to print critical messages with mininum memory allocations
possible.
In LibVT, We make the implementation flexible for kernel-specific
methods that are implemented in ConsoleImpl class.
Previously ByteBuffer would internally hold a RefPtr to the byte
buffer and would behave like a reference type, i.e. copying a
ByteBuffer would not create a duplicate byte buffer, but rather
two objects which refer to the same internal buffer.
This also changes ByteBuffer so that it has some internal capacity
much like the Vector<T> type. Unlike Vector<T> however a byte
buffer's data may be uninitialized.
With this commit ByteBuffer makes use of the kmalloc_good_size()
API to pick an optimal allocation size for its internal buffer.
This commit replaces the former, hand-written parser with a new one that
can be generated automatically according to a state change diagram.
The new `EscapeSequenceParser` class provides a more ergonomic interface
to dealing with escape sequences. This interface has been inspired by
Alacritty's [vte library](https://github.com/alacritty/vte/).
I tried to avoid changing the application logic inside the `Terminal`
class. While this code has not been thoroughly tested, I can't find
regressions in the basic command line utilities or `vttest`.
`Terminal` now displays nicer debug messages when it encounters an
unknown escape sequence. Defensive programming and bounds checks have
been added where we access parameters, and as a result, we can now
endure 4-5 seconds of `cat /dev/urandom`. :D
We generate EscapeSequenceStateMachine.h when building the in-kernel
LibVT, and we assume that the file is already in place when the userland
library is being built. This will probably cause problems later on, but
I can't find a way to do it nicely.
By constraining two implementations, the compiler will select the best
fitting one. All this will require is duplicating the implementation and
simplifying for the `void` case.
This constraining also informs both the caller and compiler by passing
the callback parameter types as part of the constraint
(e.g.: `IterationFunction<int>`).
Some `for_each` functions in LibELF only take functions which return
`void`. This is a minimal correctness check, as it removes one way for a
function to incompletely do something.
There seems to be a possible idiom where inside a lambda, a `return;` is
the same as `continue;` in a for-loop.
This implements the macOS API malloc_good_size() which returns the
true allocation size for a given requested allocation size. This
allows us to make use of all the available memory in a malloc chunk.
For example, for a malloc request of 35 bytes our malloc would
internally use a chunk of size 64, however the remaining 29 bytes
would be unused.
Knowing the true allocation size allows us to request more usable
memory that would otherwise be wasted and make that available for
Vector, HashTable and potentially other callers in the future.
If we're constructing a FlyString from a StringView, and we already
have a matching StringImpl in the table, use HashTable::find() to
locate the existing string without creating a temporary String.
Creating a ByteBuffer involves two allocations:
-One for the ByteBufferImpl object
-Another one for the actual byte buffer
This changes the ByteBuffer and ByteBufferImpl classes
so only one allocation is necessary.
This adds an `AK::ByteReader` to help with that so we don't duplicate
the logic all over the place.
No more `*(const u16*)` and `*(const u32*)` for anyone.
This should help a little with #7060.
We call placement new for the newly added slots. However, we should
also specify an initializer so primitive data types like u64 are
initialized appropriately.
Unfortunately adopt_ref requires a reference, which obviously does not
work well with when attempting to harden against allocation failure.
The adopt_ref_if_nonnull() variant will allow you to avoid using bare
pointers, while still allowing you to handle allocation failure.
Unfortunately adopt_own requires a reference, which obviously does not
work well with when attempting to harden against allocation failure.
The adopt_own_if_nonnull() variant will allow you to avoid using bare
pointers, while still allowing you to handle allocation failure.
This patch adds two new methods to LexicalPath. LexicalPath::append
appends a new path component to a LexicalPath, and LexicalPath::join
constructs a new LexicalPath from one or more components.
Co-authored-by: Gunnar Beutner <gunnar@beutner.name>
The get_dir_entries syscall failed if the serialized form of all the
directory entries together was too large to fit in its temporary buffer.
Now the kernel uses a fixed size buffer, that is flushed to an output
buffer when it is full. If this flushing operation fails because there
is not enough space available, the syscall will return -EINVAL. That
error code is then used in userspace as a signal to allocate a larger
buffer and retry the syscall.
This allows the construction of `Variant<int, int, int>`.
While this might not seem useful, it is very useful for making variants
that contain a series of member function pointers, which I plan to use
in LibGL for glGenLists() and co.
typeid() and RTTI was a nice clutch to implement this, but let's move
away from the horrible slowness and implement variants using type
indices for faster variants.
This commit introduces the ability to parse the document catalog dict,
as well as the page tree and individual pages. Pages obviously aren't
fully parsed, as we won't care about most of the fields until we
start actually rendering PDFs.
One of the primary benefits of the PDF format is laziness. PDFs are
not meant to be parsed all at once, and the same is true for pages.
When a Document is constructed, it builds a map of page number to
object index, but it does not fetch and parse any of the pages. A page
is only parsed when a caller requests that particular page (and is
cached going forwards).
Additionally, this commit also adds an object_cast function which
logs bad casts if DEBUG_PDF is set. Additionally, utility functions
were added to ArrayObject and DictObject to get all types of objects
from the collections to avoid having to manually cast.
This can currently parse a really simple module.
Note that it cannot parse the DataCount section, and it's still missing
almost all of the instructions.
This commit also adds a 'wasm' test utility that tries to parse a given
webassembly binary file.
It currently does nothing but exit when the parse fails, but it's a
start :^)
This enables us to use keys of type NonnullRefPtr in HashMaps and
HashTables.
This commit also includes fixes in various places that used
HashMap<T, NonnullRefPtr<U>>::get() and expected to get an
Optional<NonnullRefPtr<U>> and now get an Optional<U*>.
Additionally, the const version of get() returns Optional<ConstPeekType>
for smart pointers.
For example, if the value in the HashMap is OwnPtr<u32>,
HashMap::get() const returns Optional<const u32*>.
Also, the PeekType of smart pointers is now T* instead of const T*.
Note: This commit doesn't compile, it breaks HashMap::get() for some
types. Fixed in the next commit.
This currently (obviously) doesn't support any actual 3D hardware,
hence all calls are done via software rendering.
Note that any modern constructs such as shaders are unsupported,
as this driver only implements Fixed Function Pipeline functionality.
The library is split into a base GLContext interface and a software
based renderer implementation of said interface. The global glXXX
functions serve as an OpenGL compatible c-style interface to the
currently bound context instance.
Co-authored-by: Stephan Unverwerth <s.unverwerth@gmx.de>
Also adds an AK::Empty struct, because 'empty' variants are useful, but
this implementation leaves that to the user (i.e. a variant cannot
actually be empty, but it can contain an instance of Empty - i.e. a
byte).
Note that this is more of a constrained Any type, but they basically do
the same things anyway :^)
Adding -fno-semantic-interposition to the GCC command
line caused this new warning.
I don't see how output.data() could be uninitialized here. Also,
commenting out the ensure_capacity() call for the Vector
also gets rid of this warning.
This allows everybody to create a String version of their number
in a arbitrary bijective base. Bijective base meaning that the mapping
doesn't have a 0. In the usual mapping to the alphabet the follower
after 'Z' is 'AA'.
The mapping using the (uppercase) alphabet is used as a standard but
can be overridden specifying 'base' and 'map'.
The code was directly yanked from the Spreadsheet.
We had some inconsistencies before:
- Sometimes "The", sometimes "the"
- Sometimes trailing ".", sometimes no trailing "."
I picked the most common one (lowecase "the", trailing ".") and applied
it to all copyright headers.
By using the exact same string everywhere we can ensure nothing gets
missed during a global search (and replace), and that these
inconsistencies are not spread any further (as copyright headers are
commonly copied to new files).
When the two chosen pivots happen to be the smallest and largest
elements of the array, three partitions will be created, two of
size 0 and one of size n-2. If this happens on each recursive call
to dual_pivot_quick_sort, the stack depth will reach approximately n/2.
To avoid the stack from deepening, iteration can be used for the
largest of the three partitions. This ensures the stack depth
will only increase for partitions of size n/2 or smaller, which
results in a maximum stack depth of log(n).
Picking the first and last elements as pivots makes it so that
a sorted array is the worst-case input for the algorithm.
This change instead picks pivots at approximately 1/3 and 2/3 in
the array. This results in desired performance for sorted arrays.
Of course this only changes which inputs result in worst-case
performance, but hopefully those inputs occur less frequently than
already sorted arrays.
As many macros as possible are moved to Macros.h, while the
macros to create a test case are moved to TestCase.h. TestCase is now
the only user-facing header for creating a test case. TestSuite and its
helpers have moved into a .cpp file. Instead of requiring a TEST_MAIN
macro to be instantiated into the test file, a TestMain.cpp file is
provided instead that will be linked against each test. This has the
side effect that, if we wanted to have test cases split across multiple
files, it's as simple as adding them all to the same executable.
The test main should be portable to kernel mode as well, so if
there's a set of tests that should be run in self-test mode in kernel
space, we can accomodate that.
A new serenity_test CMake function streamlines adding a new test with
arguments for the test source file, subdirectory under /usr/Tests to
install the test application and an optional list of libraries to link
against the test application. To accomodate future test where the
provided TestMain.cpp is not suitable (e.g. test-js), a CUSTOM_MAIN
parameter can be passed to the function to not link against the
boilerplate main function.
This is useful for CI where we don't want to spend a minute and a half
benchmarking Vector::append, and we don't have a good way to pass
test-specific arguments yet. :)
C++20 added std::source_location, which lets you capture the
callers __FILE__ / __LINE__ / __FUNCTION__ etc as a default
argument to functions.
See: https://en.cppreference.com/w/cpp/utility/source_location
During a bug investigation @ADKaster suggested we could use this
to make the LOCK_DEBUG feature of the kernel more user friendly
and allow it to automatically instrument all call sites.
We then implemented / tested it over discord. :^)
Co-Authored-by: Andrew Kaster <andrewdkaster@gmail.com>
Problem:
- Some classes have `virtual` destructors despite not having any
virtual functions. This causes the classes to have a v-table and
perform extra jumps at destruction time when there is no need.
Solution:
- Remove `virtual` keyword from destructors where there are no other
virtual functions.
- Remove the destructor completely when the default destructor can be
used.
Problem:
- Much of the `GenericLexer` can be `constexpr`, but is not.
Solution:
- Make it `constexpr` and de-duplicate code.
- Extend some of `StringView` with `constexpr` to support.
- Add tests to ensure `constexpr` behavior.
Note:
- Construction of `StringView` from pointer and length is not
`constexpr`-compatible at the moment because the VERIFY cannot be,
yet.
SPDX License Identifiers are a more compact / standardized
way of representing file license information.
See: https://spdx.dev/resources/use/#identifiers
This was done with the `ambr` search and replace tool.
ambr --no-parent-ignore --key-from-file --rep-from-file key.txt rep.txt *
Problem:
- Type and size information is known at compile-time, but computations
are being performed using run-time parameters.
Solution:
- Move function arguments to be template arguments.
- Convert to `consteval` where possible.
- Decorate functions with `constexpr` which are used in both run-time
and compile-time contexts.
Problem:
- Global variables (and variable templates) defined in header files
need to be decorated `inline` to avoid multiple definition issues.
Solution:
- Put back the `inline` keyword which was erroneously removed.
Problem:
- `constexpr` functions are additionally decorated with `inline`
keyword. This is redundant since `constexpr` implies `inline`.
Solution:
- Remove redundancies.
PR #6376 made IntrusiveList capable of holding RefPtr<T>, etc. however
there was a latent bug where take_first() / take_last() would fail to
compile because they weren't being converted to their container type.
Problem:
- Post-increment of loop index.
- `const` variables are not marked `const`.
- Incorrect type for loop index.
Solution:
- Pre-increment loop index.
- Mark all possible variables `const`.
- Corret type for loop index.
Problem:
- Hex digit decoding is not `constexpr`, but can be.
Solution:
- Move the body of the function to the header and decorate with
`constexpr`.
- Provide tests for run-time and compile-time evaluation.
Some of the expected values in test-math were wrong, which caused some
tests to fail.
The updated values were generated by Python's math library, and rounded
to 6 decimals places:
>>> import math
>>> round(math.exp(20.99), 6)
1305693298.670892
Examples of failure outputs:
FAIL: ../Userland/Tests/LibM/test-math.cpp:98:
EXPECT_APPROXIMATE(exp(v.x), v.exp) failed with
lhs=1305693298.670892, rhs=1304956710.432034, (lhs-rhs)=736588.238857
FAIL: ../Userland/Tests/LibM/test-math.cpp:99:
EXPECT_APPROXIMATE(sinh(v.x), v.sinh) failed with
lhs=652846649.335446, rhs=652478355.216017, (lhs-rhs)=368294.119428
FAIL: ../Userland/Tests/LibM/test-math.cpp💯
EXPECT_APPROXIMATE(cosh(v.x), v.cosh) failed with
lhs=652846649.335446, rhs=652478355.216017, (lhs-rhs)=368294.119429
We had an unusual optimization in AK::StringView where constructing
a StringView from a String would cause it to remember the internal
StringImpl pointer of the String.
This was used to make constructing a String from a StringView fast
and copy-free.
I tried removing this optimization and indeed we started seeing a
ton of allocation traffic. However, all of it was due to a silly
pattern where functions would take a StringView and then go on
to create a String from it.
I've gone through most of the code and updated those functions to
simply take a String directly instead, which now makes this
optimization unnecessary, and indeed a source of bloat instead.
So, let's get rid of it and make StringView a little smaller. :^)
Previously this would create new to_lowercase()'d strings from the
needle and the haystack. This generated a huge amount of malloc
traffic in some programs.
This should allow creating intrusive lists that have smart pointers,
while remaining free (compared to the impl before this commit) when
holding raw pointers :^)
As a sidenote, this also adds a `RawPtr<T>` type, which is just
equivalent to `T*`.
Note that this does not actually use such functionality, but is only
expected to pave the way for #6369, to replace NonnullRefPtrVector<T>
with intrusive lists.
As it is with zero-cost things, this makes the interface a bit less nice
by requiring the type name of what an `IntrusiveListNode` holds (and
optionally its container, if not RawPtr), and also requiring the type of
the container (normally `RawPtr`) on the `IntrusiveList` instance.
This warning informs of float-to-double conversions. The best solution
seems to be to do math *either* in 32-bit *or* in 64-bit, and only to
cross over when absolutely necessary.
Floating point numbers are casted to i64 and passed to the integer
formatting logic, and the floating point portion of the number is
handled separately. However, casting to i64 when the number is between
-1.0 and 0.0 produces 0, so the sign would be lost. This commit fixes
that by using put_u64 instead, which allows us to manually provide the
is_negative flag.
The find_last_of implementations were breaking out of the search loop
too early for single-character string views. This caused a crash in
CookieJar setting a cookie on google.com - CookieJar::default_path knew
there was at least one "/" in a string view, but find_last_of returned
nullopt, so dereferencing the optional caused a crash.
Fixes#6273
This container is similar to the RedBlackTree container, but instead of
transparently allocating tree nodes on insertion and freeing on removal
this container piggybacks on intrusive node fields in the stored class
This container is based on a balanced binary search tree, and as such
allows for O(logn) worst-case insertion, removal, and search, as well
as O(n) sorted iteration.
If the prefix path is just a slash the LexicalPath was removing too many
characters. Now only remove an extra character if the prefix is not just
the root path.
This commit makes the user-facing StdLibExtras templates and utilities
arguably more nice-looking by removing the need to reach into the
wrapper structs generated by them to get the value/type needed.
The C++ standard library had to invent `_v` and `_t` variants (likely
because of backwards compat), but we don't need to cater to any codebase
except our own, so might as well have good things for free. :^)
This is a pretty naive implementation that works well. The precision
parameter is interpreted as "maximum precision" instead of "minimum
precision", which in my opinion is the most useful interpretation.
The old approach was more complex and also had a very bad edge case
with lots of collisions. This approach eliminates that possiblility.
It also makes both reading and writing lookups a little bit faster.
Almost a year after first working on this, it's finally done: an
implementation of Promises for LibJS! :^)
The core functionality is working and closely following the spec [1].
I mostly took the pseudo code and transformed it into C++ - if you read
and understand it, you will know how the spec implements Promises; and
if you read the spec first, the code will look very familiar.
Implemented functions are:
- Promise() constructor
- Promise.prototype.then()
- Promise.prototype.catch()
- Promise.prototype.finally()
- Promise.resolve()
- Promise.reject()
For the tests I added a new function to test-js's global object,
runQueuedPromiseJobs(), which calls vm.run_queued_promise_jobs().
By design, queued jobs normally only run after the script was fully
executed, making it improssible to test handlers in individual test()
calls by default [2].
Subsequent commits include integrations into LibWeb and js(1) -
pretty-printing, running queued promise jobs when necessary.
This has an unusual amount of dbgln() statements, all hidden behind the
PROMISE_DEBUG flag - I'm leaving them in for now as they've been very
useful while debugging this, things can get quite complex with so many
asynchronously executed functions.
I've not extensively explored use of these APIs for promise-based
functionality in LibWeb (fetch(), Notification.requestPermission()
etc.), but we'll get there in due time.
[1]: https://tc39.es/ecma262/#sec-promise-objects
[2]: https://tc39.es/ecma262/#sec-jobs-and-job-queues
