Before this change, the global FlyString table looked like this:
HashMap<StringView, Detail::StringBase>
After this change, we have:
HashTable<Detail::StringData const*, FlyStringTableHashTraits>
The custom hash traits are used to extract the stored hash from
StringData which avoids having to rehash the StringView repeatedly like
we did before.
This necessitated a handful of smaller changes to make it work.
There's no need to copy the result. We can also avoid increasing the
size of the output buffer by 1 for each written byte.
This reduces the runtime of `./bin/base64 -d enwik8.base64 >/dev/null`
from 0.917s to 0.632s.
(enwik8 is a 100MB test file from http://mattmahoney.net/dc/enwik8.zip)
We don't really need the features provided by StringBuilder here, since
we know the exact size of the output. Avoiding StringBuilder avoids the
recurring capacity/size checks both within StringBuilder itself and its
internal ByteBuffer.
This reduces the runtime of `./bin/base64 enwik8 >/dev/null` from
0.976s to 0.428s.
(enwik8 is a 100MB test file from http://mattmahoney.net/dc/enwik8.zip)
We know we are only appending ASCII characters to the StringBuilder, so
do not bother validating the result.
This reduces the runtime of `./bin/base64 enwik8 >/dev/null` from
1.192s to 0.976s.
(enwik8 is a 100MB test file from http://mattmahoney.net/dc/enwik8.zip)
This encoding scheme comes from section 5 of RFC 4648, as an
alternative to the standard base64 encode/decode methods.
The only difference is that the last two characters are replaced
with '-' and '_', as '+' and '/' are not safe in URLs or filenames.
This URL library ends up being a relatively fundamental base library of
the system, as LibCore depends on LibURL.
This change has two main benefits:
* Moving AK back more towards being an agnostic library that can
be used between the kernel and userspace. URL has never really fit
that description - and is not used in the kernel.
* URL _should_ depend on LibUnicode, as it needs punnycode support.
However, it's not really possible to do this inside of AK as it can't
depend on any external library. This change brings us a little closer
to being able to do that, but unfortunately we aren't there quite
yet, as the code generators depend on LibCore.
This was copying the vector behind our backs, let's remove it and make
the copying explicit by putting it behind COWVector::mutable_at().
This is a further 64% performance improvement on Wasm validation.
Otherwise, we percent-encode negative signed chars incorrectly. For
example, https://www.strava.com/login contains the following hidden
<input> field:
<input name="utf8" type="hidden" value="✓" />
On submitting the form, we would percent-encode that field as:
utf8=%-1E%-64%-6D
Which would cause us to receive an HTTP 500 response. We now properly
percent-encode that field as:
utf8=%E2%9C%93
And can login to Strava :^)
We already have a helper to split a StringView by line while considering
"\n", "\r", and "\r\n". Add an analagous method to just count the number
of lines in the same manner.
If the BufferedStream is able to fill its entire circular buffer in
populate_read_buffer() and is later asked to read a line or read until
a delimiter, it could erroneously return EMSGSIZE if the caller's buffer
was smaller than the internal buffer. In this case, all we really care
about is whether the caller's buffer is big enough for however much data
we're going to copy into it. Which needs to take into account the
candidate.
The main difference between them is that IntrusiveBinaryHeap can
optionally maintain an index inside every stored node that allows
arbitrary nodes to be deleted.
This is a simple extension of GenericLexer, and is used in more than
just LibXML, so let's move it into AK.
The move also resolves a FIXME, which is removed in this commit.
This makes it possible to use MakeIndexSequqnce in functions like:
template<typename T, size_t N>
constexpr auto foo(T (&a)[N])
This means AK/StdLibExtraDetails.h must now include AK/Types.h
for size_t, which means AK/Types.h can no longer include
AK/StdLibExtras.h (which arguably it shouldn't do anyways),
which requires rejiggering some things.
(IMHO Types.h shouldn't use AK::Details metaprogramming at all.
FlatPtr doesn't necessarily have to use Conditional<> and ssize_t could
maybe be in its own header or something. But since it's tangential to
this PR, going with the tried and true "lift things that cause the
cycle up to the top" approach.)
Instead of polluting global namespace with definitions from
libkern/OSByteOrder.h and machine/endian.h on MacOS, just use AK
functions for conversions.
On argument swapping to put positional ones toward the end,
m_arg_index was pointing at "last arg index" + "skipped args" +
"consumed args" and thus was pointing ahead of the skipped ones.
m_arg_index now points after the current parsed option arguments.
Now it actually only exposes methods to allocate uninitialized storage
and to create substring with a shared superstring. All the details of
the memory layout are fully encapsulated.
The idea is to eventually get rid of protected state in StringBase. To
do this, we first need to remove all references to m_data and
m_short_string from String.
This starts separating memory management of string data and string
utilities like `String::formatted`. This would also allow to reuse the
same storage in `DeprecatedString` in the future.
This method (unlike can_read_line) ensures that the delimiter is present
in the buffer, and doesn't return true after eof when the delimiter is
absent.
`JsonValue::to_byte_string` has peculiar type-erasure semantics which is
not usually intended. Unfortunately, it also has a very stereotypical
name which does not warn about unexpected behavior. So let's prefix it
with `deprecated_` to make new code use `as_string` if it just wants to
get string value or `serialized<StringBuilder>` if it needs to do proper
serialization.
A bunch of users used consume_specific with a constant ByteString
literal, which can be replaced by an allocation-free StringView literal.
The generic consume_while overload gains a requires clause so that
consume_specific("abc") causes a more understandable and actionable
error.
The current algorithm is currently O(N^2) because we forward-search an
ever-increasing substring of the haystack. This implementation reduces
the search time of a 500,000-length string (where the desired needle is
at index 0) from 72 seconds to 2-3 milliseconds.
Caught by clang-format-17. Note that clang-format-16 is fine with this
as well (it leaves the const placement alone), it just doesn't perform
the formatting to east-const itself.
Using a vector to represent a list of painting commands results in many
reallocations, especially on pages with a lot of content.
This change addresses it by introducing a SegmentedVector, which allows
fast appending by representing a list as a sequence of fixed-size
vectors. Currently, this new data structure supports only the
operations used in RecordingPainter, which are appending and iterating.
Much of the UTF-8 data that we'll iterate over will be ASCII only,
and we can get a significant speed-up by simply having a fast path
when the iterator points at a byte that is obviously an ASCII character
(<= 0x7F).
Since we're already building up a percent-encoded ASCII-only string
in the internal parser buffer, there's no need to do a second UTF-8
validation pass before assigning each part of the parsed URL.
This makes URL parsing signficantly faster.
Instead of do a wrappy MUST(try_append_code_point()), we now inline
the UTF-8 encoding logic. This allows us to grow the buffer by the
right increment up front, and also removes a bunch of ErrorOr ceremony
that we don't care about.
Once we know that a code point must be a valid ASCII character,
we now cast it to `char` and avoid the expensive generic
StringView::contains(u32 code_point) checks.
This dramatically speeds up URL parsing.
UTF8Decoder was already converting invalid data into replacement
characters while converting, so we know for sure we have valid UTF-8
by the time conversion is finished.
This patch adds a new StringBuilder::to_string_without_validation()
and uses it to make UTF8Decoder avoid half the work it was doing.
Instead of using a StringBuilder, add a String::repeated(String, N)
overload that takes advantage of knowing it's already all UTF-8.
This makes the following microbenchmark go 4x faster:
"foo".repeat(100_000_000)
And for single character strings, we can even go 10x faster:
"x".repeat(100_000_000)
In a bunch of cases, this actually ends up simplifying the code as
to_number will handle something such as:
```
Optional<I> opt;
if constexpr (IsSigned<I>)
opt = view.to_int<I>();
else
opt = view.to_uint<I>();
```
For us.
The main goal here however is to have a single generic number conversion
API between all of the String classes.
Half the functions used are not readily available on windows, instead of
creating more ifdef soup, this commit simply disables the rich debug
stuff on windows.
If we don't have __builtin_add_overflow_p(), we can also try using
__builtin_add_overflow(). This makes debug builds with Clang
significantly faster since they no longer need to use the generic
implementation. Same for multiplication.
Previously, `replace` used `find_all` to find all of the positions to
replace. But `find_all` finds all the *overlapping* instances of the
needle, while `replace` assumed that the next position was always at
least `needle.length()` away from the last one. This led to crashes like
https://github.com/SerenityOS/jakt/issues/1159.
