The array which contains the actual parameters is always located
immediately after the base `TypeErasedFormatParams` object of
`VariadicFormatParams`. Hence, storing a pointer to it inside a `Span`
is redundant. Changing it to a zero-length array saves 8 bytes.
Secondly, we limit the number of parameters to 256, so `m_size` and
`m_next_index` can be stored in a smaller data type than `size_t`,
saving us another 8 bytes.
This decreases the size of a single-element `VariadicFormatParams` from
48 to 32 bytes, thus reducing the code size overhead of setting up
parameters for `dbgln()`.
Note that [arrays of length zero][1] are a GNU extension, but it's used
elsewhere in the codebase already and is explicitly supported by Clang
and GCC.
[1]: https://gcc.gnu.org/onlinedocs/gcc/Zero-Length.html
Xcode 15 betas 1-3 lack https://reviews.llvm.org/D135772, which fixes a
bug that causes trailing `requires` clauses to be evaluated twice,
failing the second time. Reported as FB12284201.
This caused compile errors when instantiating types derived from RefPtr:
> error: invalid reference to function 'NonnullRefPtr': constraints not
> satisfied
> note: because substituted constraint expression is ill-formed: value
> of type '<dependent type>' is not contextually convertible to 'bool'.
This commit works around the issue by moving the `requires` clauses
after the template parameter list.
In most cases, trailing `requires` clauses and those specified after the
template parameter list work identically, so this change should not
impact the code's behavior. The only difference is that trailing
requires clauses are evaluated *after* constrained placeholder types
(i.e. `Integral auto i` function parameter).
All elements of the vector were moved to the left, for each element to
remove. This patch makes the function move each element exactly once.
On the same test case as the previous commit, it makes the function
disappear from the profile. These two commits combined reduce the
decompression time by 12%.
As confusing as it may sound, reusing them is terrible performance wise.
When profiling the PNG decoder, the result (which is dominated by the
Zlib decompression) shows that the `cleanup_unused_chunks()` function
represented 14.26% of the profile before this patch and only 7.7%
afterward.
On a 6.5 MB PNG image, it reduces the decompression time by more than
5%.
This uses one of Sun OS's algorithms, for a comparison to other
algorithms please refer to
https://gist.github.com/Hendiadyoin1/f58346d66637deb9156ef360aa158bf9
This is used on aarch64 builds and for x86 floats and doubles
for performance gains check
https://quick-bench.com/q/_2jTykshP6cUqtgdepFaoQ53YC8
which shows approximately 2x gains
Co-Authored-By: Ben Wiederhake <BenWiederhake.GitHub@gmx.de>
Co-Authored-By: kleines Filmröllchen <filmroellchen@serenityos.org>
Co-Authored-By: Dan Klishch <danilklishch@gmail.com>
This now searches the memory in blocks, which should be slightly more
efficient. However, it doesn't make much difference (e.g. ~1% in LZMA
compression) in most real-world applications, as the non-hint function
is more expensive by orders of magnitude.
This factors out a lot of complicated math into somewhat understandable
functions.
While at it, rename `next_read_span_with_seekback` to
`next_seekback_span` to keep the naming consistent and to avoid making
function names any longer.
The "operation modes" of this function have very different focuses, and
trying to combine both in a way where we share the most amount of code
probably results in the worst performance.
Instead, split up the function into "existing distances" and "no
existing distances" so that we can optimize either case separately.
We will be adding extra logic to the CircularBuffer to optimize
searching, but this would negatively impact the performance of
CircularBuffer users that don't need that functionality.
By golly, this is a lot more spec comments than I originally thought
I would need to do! This has exposed some bugs in the implementation,
as well as a whole lot of things which we are yet to implement.
No functional changes intended in this commit (already pretty large
as is!).
ECMA-262 implies that `MIN_VALUE` should be a denormalized value if
denormal arithmetic is supported. This is the case on x86-64 and AArch64
using standard GCC/Clang compilation settings.
test262 checks whether `Number.MIN_VALUE / 2.0` is equal to 0, which
only holds if `MIN_VALUE` is the smallest denormalized value.
This commit renames the existing `NumericLimits<FloatingPoint>::min()`
to `min_normal()` and adds a `min_denormal()` method to force users to
explicitly think about which one is appropriate for their use case. We
shouldn't follow the STL's confusingly designed interface in this
regard.
Instead of checking the address of a temporary, grab the address of the
current frame pointer to determine how much memory is left on the stack.
This better communicates to the compiler what we're trying to do, and
resolves some crashes with ASAN in test-js while the option
detect_stack_use_after_return is turned on.
This was missed in 02b74e5a70
We need to disable consteval in AK::String as well as AK::StringView,
and we need to disable it when building both the tools build and the
fuzzer build.
These recursive templates have a measurable impact on the compile speed
of Variant-heavy code like LibWeb. Using these builtins leads to a 2.5%
speedup for the measured compilation units.
oss-fuzz ships a pre-release commit of clang-15 for all of their build
bots. Until they update to a release of clang-15 that includes the fix
for this bug, or a later release, we need to keep the workaround in
place.
The Windows CRT definition of assert() is not noreturn, and causes
compile errors when using it as the backing for VERIFY() in debug
configurations of applications like the Jakt compiler.
Apple Clang 14.0.3 (Xcode 14.3) miscompiles this builtin on AArch64,
causing the borrow flag to be set incorrectly. I have added a detailed
writeup on Qemu's issue tracker, where the same issue led to a hang when
emulating x86:
https://gitlab.com/qemu-project/qemu/-/issues/1659#note_1408275831
I don't know of any specific issue caused by this on Lagom, but better
safe than sorry.
GCC 14 (https://gcc.gnu.org/g:2b4e0415ad664cdb3ce87d1f7eee5ca26911a05b)
has added support for the previously Clang-specific add/subtract with
borrow builtins. Let's use `__has_builtin` to detect them instead of
assuming that only Clang has them. We should prefer these to the
x86-specific ones as architecture-independent middle-end optimizations
might deal with them better.
As an added bonus, this significantly improves codegen on AArch64
compared to the fallback implementation that uses
`__builtin_{add,sub}_overflow`.
For now, the code path with the x86-specific intrinsics stays so as to
avoid regressing the performance of builds with GCC 12 and 13.
The previous version had a sequence of calls that are likely not
optimized out, while this version is strictly a sequence of static type
conversion which are always fully optimized out.
The previous alignment would always resolve to 8-bytes, which is below
the required alignments of types that could exist in userspace (long
double, 128-bit integers, SSE, etc).
The FileSlash state was erroneously copying the base URL host, instead
of the base URL path excluding the last path component. This resulted in
invalid file URLs.
Calling `from_utf8` with a DeprecatedString will hide the fact that we
have a DeprecatedString, while using `from_deprecated_string` with a
StringView will silently and needlessly allocate a DeprecatedString,
so let's forbid that.
This does a few things:
- The decoder uses a 32- or 64-bit integer as a reservoir of the data
being decoded, rather than one single byte as it was previously.
- `read_bool()` only refills the reservoir (value) when the size drops
below one byte. Previously, it would read out a bit-sized range from
the data to completely refill the 8-bit value, doing much more work
than necessary for each individual read.
- VP9-specific code for reading the marker bit was moved to its own
function in Context.h.
- A debug flag `VPX_DEBUG` was added to optionally enable checking of
the final bits in a VPX ranged arithmetic decode and ensure that it
contains all zeroes. These zeroes are a bitstream requirement for
VP9, and are also present for all our lossy WebP test inputs
currently. This can be useful to test whether all the data present in
the range has been consumed.
A lot of the size of this diff comes from the removal of error handling
from all the range decoder reads in LibVideo/VP9 and LibGfx/WebP (VP8),
since it is now checked only at the end of the range.
In a benchmark decoding `Tests/LibGfx/test-inputs/4.webp`, decode times
are improved by about 22.8%, reducing average runtime from 35.5ms±1.1ms
down to 27.4±1.1ms.
This should cause no behavioral changes.
Change the name and return type of
`IPv6Address::to_deprecated_string()` to `IPv6Address::to_string()`
with return type `ErrorOr<String>`.
It will now propagate errors that occur when writing to the
StringBuilder.
There are two users of `to_deprecated_string()` that now use
`to_string()`:
1. `Formatted<IPv6Address>`: it now propagates errors.
2. `inet_ntop`: it now sets errno to ENOMEM and returns.
This has KString, KBuffer, DoubleBuffer, KBufferBuilder, IOWindow,
UserOrKernelBuffer and ScopedCritical classes being moved to the
Kernel/Library subdirectory.
Also, move the panic and assertions handling code to that directory.
This partially implements CSS-Animations-1 (though there are references
to CSS-Animations-2).
Current limitations:
- Multi-selector keyframes are not supported.
- Most animation properties are ignored.
- Timing functions are not applied.
- Non-absolute values are not interpolated unless the target is also of
the same non-absolute type (e.g. 10% -> 25%, but not 10% -> 20px).
- The JavaScript interface is left as an exercise for the next poor soul
looking at this code.
With those said, this commit implements:
- Interpolation for most common types
- Proper keyframe resolution (including the synthetic from-keyframe
containing the initial state)
- Properly driven animations, and proper style invalidation
Co-Authored-By: Andreas Kling <kling@serenityos.org>
This class takes on the duties of CLOCK_MONOTONIC, a time without a
defined reference point that always increases. This informs some
important design decisions about the class API: MonotonicTime cannot be
constructed from external time data, except as a computation based on
other monotonic time, or the current monotonic time. Importantly, there
is no default constructor, since the reference point of monotonic time
is unspecified and therefore without meaning as a default.
The current use of monotonic time (via Duration) includes some potential
problems that may be caught when we move most to all code to
MonotonicTime in the next commit.
The API restrictions have one important relaxation:
Kernel::TimeManagement is allowed to exchange raw time data within
MonotonicTime freely. This is required for the clock-agnostic time
accessors for timeouts and syscalls, as well as creating monotonic time
data from hardware in the first place.
"Wherever applicable" = most places, actually :^), especially for
networking and filesystem timestamps.
This includes changes to unzip, which uses DOSPackedTime, since that is
changed for the FAT file systems.
This is a generic wrapper for a time instant relative to the unix epoch,
and does not account for leap seconds. It should be used in place of
Duration in most current cases.
This is a trivial change, and since this batch of commits will make a
large-scale rebuild necessary anyways, it seems sensible. The feature is
useful for e.g. building compound constant durations at compile time in
a readable way.
That's what this class really is; in fact that's what the first line of
the comment says it is.
This commit does not rename the main files, since those will contain
other time-related classes in a little bit.
This attribute is used for functions in the kernel that are entirely
written in assembly, yet defined in C++ source files.
Without `__attribute__((naked))`, Clang might decide to inline these
functions, making any `ret` instructions within them actually exit the
caller, or discard argument values as they appear "dead". This issue
caused a kernel panic when using the `execve` syscall in AArch64
SerenityOS built by Clang.
While the empty definition so far appears to work fine with GCC, simpler
test cases do similarly suffer from unintended inlining, so define
`NAKED` as a synonym of `NEVER_INLINE` to avert future issues.
Perhaps we should move users of `NAKED` to plain assembly files?
This makes aarch64Clang builds boot :^)
This underlines that we still copy-construct and copy-assign HashMaps.
Primarily, this makes it easier to develop towards OOM-safe(r) internal
data structures, by providing a reminder (the FIXME) and an easy error-
checking switch (just change it to "delete" to see some of the errors).
See identical code in LittleEndianBitStream; even in the bytewise
reading BigEndianBitStream an offset of 8 is not inconsistent state and
handled just fine by read_bits.
Rather than tracking our position in the bit buffer, we can simply shift
away the bits that we read. This is mostly for simplicity, but also does
help performance a bit.
Using the "enwik8" file as a test (100MB uncompressed, commonly used in
benchmarks: https://www.mattmahoney.net/dc/enwik8.zip), compression time
decreases from:
3.96s to 3.79s on Serenity (cold)
1.08s to 1.04s on Serenity (warm)
0.83s to 0.82s on Linux
We have outln() and out(), warnln() and warn(),
now we have dbgln() and dbg().
This is useful for printing arrays element-by-element while still
only printing one line per array.
This class, in a similar fashion to what has been done with
`InputBufferedStream`, postpones write to the stream until an internal
buffer is full.
This patch also adds the `OutputBufferedFile` alias.
In a similar fashion to what have been done with `fill_from_stream`,
this new method allows to write CircularBuffer's data to a Stream
without additional copies.
Due to 582c55a, both `must_set()` and `set()` should be providing the
same behavior. Not only is that a reason to remove `must_set()`, but
it also performs erroneous behavior since it inserts an element at
a specified index instead of modifying an element at that index.
"The official project language is American English […]."
5d2e915623/CONTRIBUTING.md (L30)
Here's a short statistic of the occurrences of the word "behavio(u)r":
$ git grep -IPioh 'behaviou?r' | sort | uniq -c | sort -n
2 BEHAVIOR
24 Behaviour
32 behaviour
407 Behavior
992 behavior
Therefore, it is clear that "behaviour" (56 occurrences) should be
regarded a typo, and "behavior" (1401 occurrences) should be preferred.
Note that The occurrences in LibJS are intentionally NOT changed,
because there are taken verbatim from the specification. Hence:
$ git grep -IPioh 'behaviou?r' | sort | uniq -c | sort -n
2 BEHAVIOR
10 behaviour
24 Behaviour
407 Behavior
1014 behavior
Our current `peek_bits` function allows retrieving more bits than we can
actually provide, so whenever someone discards the requested bit count
afterwards we were underflowing the value instead.
Resolves#18618.
8134dcc changed `JsonArray::set()` to insert elements at an index
instead of changing existing elements in-place. Since no behavior
such as `Vector::try_at()` exists yet, it returns nothing.
GCC 13 was released on 2023-04-26. This commit fixes Lagom build errors
when using an updated host toolchain:
- Adds a workaround for a bug in constraint handling, which made LibJS
fail to compile: https://gcc.gnu.org/bugzilla/show_bug.cgi?id=109683
- Silences the new `-Wdangling-reference` diagnostic globally. It
produces multiple false positives with no clear way to silence them
without `#pragmas`.
- Silences `-Wself-move` in `RefPtr` tests as GCC 13 adds this
previously Clang-exclusive warning.
This further optimizes floating point parsing (specifically with a large
amount of digits). The commit shaves additional 20% of the run time for
750-digit numbers. No performance degradation is noticeable for small
numbers.
Although it might seem like we've switched to more generic functions,
which must run slower, it is not the case. The time required to parse
"1", for example, decreased by 1%. For numbers with more digits, the
effect is more noticeable: 8-digit numbers are parsed ~5% faster; for
gigantic 750-digit numbers, parsing is 2 times faster.
The later result is achieved by using UFixedBigInt<64>::wide_multiply
instead of u128::operator*(u128).
Additionally, split it into two versions (for IsIntegral<T> -- asking
to place value into register and for !IsIntegral<T> -- asking to place
value into memory with memory clobber), so that Clang is no more
completely confused about `taint_for_optimizer(AK::StringView&)`.
There's no real reason to make this a debug-only formatter, on top of
that, jakt has a optional formatter that prints None/foo instead of
OptionalNone/Optional(foo), which is more concise anyway, so switch to
that.
I found this handy for debugging, and so might others.
This now also adds a formatter for TimeZone::TimeZone. This is needed
for FormatIfSupported<Optional<TimeZone::TimeZone>> to compile. As
FormatIfSupported sees a formatter for Optional exists, but not that
there's not one for TimeZone::TimeZone.
While swap() is available in the global namespace in normal conditions,
!USING_AK_GLOBALLY will make this name unavailable in the global
namespace, making these calls fail to compile.
That pattern seems to show up a lot in code written by people that
aren't intimately familiar with the lifetime model of Error and Strings.
This commit makes the compiler detect it and present a more helpful
diagnostic than "garbage string at runtime".
This brings the function name in line with how we usually name those
functions, which is with a `read_` or `write_` prefix depending on what
they do.
While at it, make the internal `_impl` function private and not-virtual,
since there is no good reason to ever override that function.
This class can be used to run a task in another thread, and allows the
caller to wait for the task to complete to retrieve any error that may
have occurred.
Currently, it doesn't support functions returning a value on success,
but with some template magic that should be possible. :^)
Based on `out()` and `vout()` from AK/Format.h. As opposed to those,
this propagates errors.
As `Core::File` extends `AK::Stream`, this allows formatted
printing directly on `Core::File`s.
We now null out smart pointers *before* calling unref on the pointee.
This ensures that the same smart pointer can't be used to acquire a new
reference to the pointee after its destruction has begun.
I ran into this when destroying a non-empty IntrusiveList of RefPtrs,
but the problem was more general so this fixes it for all of RefPtr,
NonnullRefPtr, OwnPtr and NonnullOwnPtr.
This allows accessing and looping over the path segments in a URL
without necessarily allocating a new vector if you want them percent
decoded too (which path_segment_at_index() has an option for).
This now defaults to serializing the path with percent decoded segments
(which is what all callers expect), but has an option not to. This fixes
`file://` URLs with spaces in their paths.
The name has been changed to serialize_path() path to make it more clear
that this method will generate a new string each call (except for the
cannot_be_a_base_url() case). A few callers have then been updated to
avoid repeatedly calling this function.
The defaults selected for this are based on the behaviour of URL
when it applied percent decoding during parsing. This does mean now
in some cases the getters will allocate, but percent_decode() checks
if there's anything to decode first, so in many cases still won't.
ARCH() uses the AK_IS_ARCH_ macros internally since 349e54d5375a4a,
and all user code uses the ARCH() macro instead of AK_ARCH_.
(Why it's called ARCH() and not AK_ARCH(), I don't know.)
If any ports not in the main repo use AK_ARCH_, they should switch
to using ARCH() instead.
957f89ce4a added some tweaks for serenity-on-aarch64.
It broke anythingelse-on-aarch64 hosts though, so only do these tweaks
when targeting serenity.
(I wonder if AK/Math.h should fall back to the system math routines
when not targeting serenity in general. Would probably help ladybird
performance. On the other hand, the serenity routines would see less
use and hence exposure and love.)
Previously, if we copied the last byte for a length of 100, we'd
recalculate the read span 100 times and memmove one byte 100 times,
which resulted in a lot of overhead.
Now, if we know that we have two consecutive copies of the data, we just
extend the distance to cover both copies, which halves the number of
times that we recalculate the span and actually call memmove.
This takes the running time of the attached benchmark case from 150ms
down to 15ms.
The {sin,cos,tan} functions in AK are used as the implementation of the
same function in libm. We cannot use the __builtin_foo functions as
these would just call the libc functions. This was causing an infinite
loop. Fix this by adding a very naive implementation of
AK::{sin, cos,tan}, that is only valid for small inputs. For the other
functions in this file, I added a TODO() such that we'll crash, instead
of infinite looping.
This is a remnant from when we didn't have a `read_value` and
`write_value` implementation for `AK::Endian` and instead used the
generic functions for reading a span of bytes. Now that we have a more
ergonomic alternative, remove the helper that is no longer needed.
As noted in serval comments doing this goes against the WC3 spec,
and breaks parsing then re-serializing URLs that contain percent
encoded data, that was not encoded using the same character set as
the serializer.
For example, previously if you had a URL like:
https:://foo.com/what%2F%2F (the path is what + '//' percent encoded)
Creating URL("https:://foo.com/what%2F%2F").serialize() would return:
https://foo.com/what//
Which is incorrect and not the same as the URL we passed. This is
because the re-serializing uses the PercentEncodeSet::Path which
does not include '/'.
Only doing the percent encoding in the setters fixes this, which
is required to navigate to Google Street View (which includes a
percent encoded URL in its URL).
Seems to fix#13477 too
`vformat()` can now accept format specifiers of the form
{:'[numeric-type]}. This will output a number with a comma separator
every 3 digits.
For example:
`dbgln("{:'d}", 9999999);` will output 9,999,999.
Binary, octal and hexadecimal numbers can also use this feature, for
example:
`dbgln("{:'x}", 0xffffffff);` will output ff,fff,fff.
When BufferedFile.can_read_line() was invoked on files with no newlines,
t incorrectly returned a false result for this single line that, even
though doesn't finish with a newline character, is still a line. Since
this method is usually used in tandem with read_line(), users would miss
reading this line (and hence all the file contents).
This commit fixes this corner case by adding another check after a
negative result from finding a newline character. This new check does
the same as the check that is done *before* looking for newlines, which
takes care of this problem, but only works for files that have at least
one newline (hence the buffer has already been filled).
A new unit test has been added that shows the use case. Without the
changes in this commit the test fails, which is a testament that this
commit really fixes the underlying issue.
For whatever reason, when CLion does its code indexing thing, it doesn't
define __clang__ despite using Clang. This causes it to run into various
problems that we've solved by checking for Clang.
Since CLion does define __CLION_IDE__ (or sometimes __CLION_IDE_, no
idea why but I have seen this issue locally), let's make that part of
the AK_COMPILER_CLANG check.
This makes CLion stop highlighting various things as errors.
I was originally thinking in the wrong direction when adding this limit,
we can at most read from the buffer until we reach the current write
head. Since that write head is the reference point for the distance,
we need to limit ourselves to that instead of the seekback limit (which
is the maximum of how far back the distance can be).
Rather than the very C-like API we currently have, accepting a void* and
a length, let's take a Bytes object instead. In almost all existing
cases, the compiler figures out the length.
This helper constructor exists on the unspecialized Span<T> class also,
and is convenient for e.g. creating Bytes from:
u8 buffer[64];
Bytes bytes { buffer };
From the getentropy() man page, "The maximum permitted value for the
length argument is 256". Several of our tests were passing lengths of
several thousand bytes, causing getentropy() to fail with EIO, which we
were completely ignoring. This caused these tests to only test long
sequences of 0x00.
We now loop over the provided buffer to fill it 256 bytes at a time. If
getentropy() fails for any reason, we fall back to the default method of
filling it with one random byte at a time.
This is very similar to the LittleEndianInputBitStream bit buffer change
from 8e834d4bb2.
We currently buffer one byte of data for the underlying stream. And when
we put bits onto that buffer, we do so 1 bit at a time.
This replaces the u8 buffer with a u64. And instead of looping at all,
we perform bitwise operations to write the desired number of bits.
Using the "enwik8" file as a test (100MB uncompressed, commonly used in
benchmarks: https://www.mattmahoney.net/dc/enwik8.zip), compression time
decreases from:
13.62s to 10.9s on Serenity (cold)
13.62s to 9.22s on Serenity (warm)
2.93s to 2.32s on Linux
One caveat is that this requires explicitly flushing any leftover bits
when the caller is done with the stream. The byte buffer implementation
implicitly flushed its data every time the buffer was byte-aligned, as
doing so would always fill the byte. This is no longer the case. But for
now, this should be fine as the one user of this class, DEFLATE, already
has a "flush everything now that we're done" finalizer.
Instead of reading bytes from the output stream into a buffer, just to
immediately write them back out, we can skip the middle-man and copy the
bytes directly into the output buffer.
We currently only fill a buffer when it is empty. So if it has 1 byte
and 16 KB was requested, only that 1 byte would be returned. Instead,
attempt to refill the buffer when it's size is less than the requested
size.
When reading, we currently only fill a BufferedStream's buffer when it
is empty, and only with 1 KB of data. This means that while the buffer
defaults to a size of 16 KB, at least 15 KB is always unused.
The current implementation of `Array<T, 0>` has a zero-length C array as
its storage type. While this is accepted as a GNU extension, when
compiling with Clang 16, an UBSan error is raised every time an object
is accessed whose only field is a zero-length array.
This is likely a bug in Clang 16's implementation of UBSan, which has
been reported here: https://github.com/llvm/llvm-project/issues/61775
Else:
AK/BitStream.h:218:24:
error: inline function '...::lsb_mask<unsigned char>' is not
defined [-Werror,-Wundefined-inline]
static constexpr T lsb_mask(T bits)
^
We current buffer one byte of data from the underlying stream. And when
we pull bits off that buffer, we do so 1 or 8 bits at a time (depending
on whether the buffer is byte aligned). The 1-bit-at-a-time loop is by
far the most common during e.g. GZIP decompression.
This replaces the u8 buffer with a u64. And instead of looping at all,
we perform bitwise operations to extract the desired number of bits.
Using the "enwik8" file as a test (100MB uncompressed, commonly used in
benchmarks: https://www.mattmahoney.net/dc/enwik8.zip), decompression
time decreases from:
242s to 35s on Serenity
11.125s to 3.527s on Linux
Note that BigEndianInputBitStream can also use the same techniques,
and some of the methods here may make sense to live in an endianness-
agnostic base class. The focus is GZIP right now though, which only
uses the little endian stream.
This mirrors String::from_utf8(StringView).
Jakt will use this to construct strings instead of just assuming the
allocation will succeed, lowering the API difference between
Jakt::String and AK::String by one API :^)
This patch parses enough of GPOS tables to be able to support the
kerning information embedded in Inter.
Since that specific font only applies positioning offsets to the first
glyph in each pair, I was able to get away with not changing our API.
Once we start adding support for more sophisticated positioning, we'll
need to be able to communicate more than a simple "kerning offset" to
the clients of this code.
With Clang, the previous/next pointers in buckets of an
`OrderedHashTable` are not cleared when a bucket is being shifted up as
a result of a removed bucket. As a result, an unfortunate pointer mixup
could lead to an infinite loop in the `HashTable` iterator, which was
exposed in `HashMap::keys()`.
Co-authored-by: Luke Wilde <lukew@serenityos.org>
Similar to POSIX read, the basic read and write functions of AK::Stream
do not have a lower limit of how much data they read or write (apart
from "none at all").
Rename the functions to "read some [data]" and "write some [data]" (with
"data" being omitted, since everything here is reading and writing data)
to make them sufficiently distinct from the functions that ensure to
use the entire buffer (which should be the go-to function for most
usages).
No functional changes, just a lot of new FIXMEs.
We don't need to decode the entire code point to know its length. This
reduces the runtime of decoding a string containing 5 million instances
of U+10FFFF from over 4 seconds to 0.9 seconds.
Let's add FlyString::from_deprecated_fly_string() so we can use it
instead of FlyString::from_utf8(). This will make it easier to detect
potential unncessary allocations as we transfer to FlyString.
We currently fully casefold the left- and right-hand sides to compare
two strings with case-insensitivity. Now, we casefold one code point at
a time, storing the result in a view for comparison, until we exhaust
both strings.