This commit addresses the following shortcomings of our current, simple
and elegant memset function:
- REP STOSB/STOSQ has considerable startup overhead, it's impractical to
use for smaller sizes.
- Up until very recently, AMD CPUs didn't have support for "Enhanced REP
MOVSB/STOSB", so it performed pretty poorly on them.
With this commit applied, I could measure a ~5% decrease in `test-js`'s
runtime when I used qemu's TCG backend. The implementation is based on
the following article from Microsoft:
https://msrc-blog.microsoft.com/2021/01/11/building-faster-amd64-memset-routines
Two versions of the routine are implemented: one that uses the ERMS
extension mentioned above, and one that performs plain SSE stores. The
version appropriate for the CPU is selected at load time using an IFUNC.
.text sections of objects that contain textrels have to be writable
during the relocation procedure. Because of this, we would segfault if
we tried to execute IFUNC resolvers defined in them. Let's print a
meaningful error message instead.
Additionally, a warning is now printed when we load objects with
textrels, as in the future, additional security mitigations might
interfere with them being loaded.
IFUNC is a GNU extension to the ELF standard that allows a function to
have multiple implementations. A resolver function has to be called at
load time to choose the right one to use. The PLT will contain the entry
to the resolved function, so branching and more indirect jumps can be
avoided at run-time.
This mechanism is usually used when a routine can be made faster using
CPU features that are available in only some models, and a fallback
implementation has to exist for others.
We will use this feature to have two separate memset implementations for
CPUs with and without ERMS (Enhanced REP MOVSB/STOSB) support.
IFUNC resolvers depend on the resolved function's address having been
relocated by the time they are called. This means that relative
relocations have to be done first.
The linker is kind enough to put R_*_RELATIVE before R_*_IRELATIVE in
.rel.dyn, but .relr.dyn contains relative relocations too.
This check is here to make sure we only try to load serenity binaries.
However, with -fprofile-instr-generate -fcoverage-mapping, clang
sets the EI_OSABI field to 3, for GNU. The instrumentation uses a lot of
retained COMDAT sections for coverage instrumentation that get the
SHF_GNU_RETAINED section header flag set on them, forcing llvm to set
the ABI to GNU.
A function object's realm is not necessarily non-null (like when called
via the Reflect API), so we can't blindly dereference it. Instead use
the object's own GlobalObject.
Because `TimeHour TimeZoneNumericUTCOffsetNotAmbiguous[opt]
TimeZoneBracketedAnnotation[opt]` can be a subset of
`TimeHourNotValidMonth TimeZone` we would not exhaust the whole input
in some cases, which would result in an incorrectly thrown exception.
There are two different things in ThemeEditor that want to know when a
palette changes:
1. The PreviewWidget subclass, so it can update its preview.
2. The ThemeEditor itself, so we know that the palette is modified.
Using a protected virtual function for 1 means that we can do 2 without
them fighting over the same on_palette_change callback.
The `TextEditor::on_change` callback now only fires if the user types in
the box, or `set_text()` is called with `AllowCallback::Yes`.
Previously that callback was what set `m_color`, so I've rearranged
things a little so that the color still updates regardless of what
source the color came from.
...and change how the two parsing steps fit together.
The two steps were previously quite muddled. Both worked with the
TokenStream directly, and both were responsible for rewinding that
stream if there was an error. This is both confusing and also made it
impossible to replace the rewinding with StateTransactions.
This commit more clearly divides the work between the two functions: One
parses ComponentValues and produces a string, and the other parses that
string to produce the UnicodeRange. It also replaces manual rewinding
in the former with StateTransactions.
This should be a bit easier to follow.
parse_media_query() no longer rewinds if the media query is invalid,
because it then interprets all the tokens as a "not all" query.
This is modeled after the one in ISO8601Parser. It rolls back the
TokenStream state automatically at the end of scope unless told to
commit the changes. This should be less error-prone than remembering to
manually call `rewind_to_position()` at the correct time.
For convenience, a StateTransaction can have "child" transactions. When
a transaction is committed, it automatically commits its parents too.
This is useful in situations where you have several nested and don't
want to have to remember to manually `commit()` them all.
`a` and `b` had to be declared at the top of the function before since
they were used by the `make_return_value()` lambda. But now that
doesn't exist, we can move them to where they are used - or eliminate
them entirely.
parse_a_n_plus_b_pattern()'s job is to parse as much of the TokenStream
as it can as a An+B, and then stop. The caller can then deal with any
trailing tokens as it wishes.
...using a ParseErrorOr type alias.
This lets us replace a bunch of manual error-checking with TRY. :^)
I also replaced the ParsingResult::Done value with returning an
Optional. I wasn't happy with treating "Done" as an error when I first
wrote this, and this makes a clear distinction between the two.
Using ~~text~~ syntax will strike out the text between the two tildes.
Only missing portion is the terminal rendering of strike through text.
The ansi escape codes for strike through text are \e[9m and \e[29m but
it appears the terminal does not support these. Please correct me if I
am wrong.
I tested that the render_to_terminal function was being called by giving
it bold ANSI escape codes, and that did work so the function is being
called correctly.
