Similar to commit 0652cc4, we now generate 2-stage lookup tables for
case conversion information. Only about 1500 code points are actually
cased. This means that case information is rather highly compressible,
as the blocks we break the code points into will generally all have no
casing information at all.
In total, this change:
* Does not change the size of libunicode.so (which is nice because,
generally, the 2-stage lookup tables are expected to trade a bit
of size for performance).
* Reduces the runtime of the new benchmark test case added here from
1.383s to 1.127s (about an 18.5% improvement).
We currently produce a single table for all categories of code point
properties (GeneralCategory, Script, etc.). Each row contains a field
indicating the range of code points to which that property applies. At
runtime, we then do a binary search through that table to decide if a
code point has a property.
This changes our approach to generate a 2-stage lookup table for each of
those categories. There is an in-depth explanation of these tables above
the new `create_code_point_tables` method. The end effect is that code
point property lookup is reduced from a binary search to constant-time
array lookups.
In total, this change:
* Increases the size of libunicode.so from 2.7 MB to 2.9 MB.
* Reduces the runtime of the new benchmark test case added here from
3.576s to 1.020s (a 3.5x speedup).
* In a profile of resizing a TextEditor window with a 3MB file open,
the runtime of checking if a code point has a word break property
reduces from ~81% to ~56%.
We started generating this data in commit 0505e03, but it was unused.
It's still not used, so let's remove it, rather than bloating the size
of libunicode.so with unused data. If we need it in the future, it's
trivial to add back.
Note we *have* always used the block name data from that commit, and
that is still present here.
This was preventing some unqualified emoji sequences from rendering
properly, such as the custom SerenityOS flag. We rendered the flag
correctly when given the fully qualified sequence:
U+1F3F3 U+FEOF U+200D U+1F41E
But were not detecting the unqualified sequence as an emoji when also
filtering for emoji-presentation sequences:
U+1F3F3 U+200D U+1F41E
For example the words "can't" and "32.3" should not have boundaries
detected on the "'" and "." code points, respectively.
The String test cases fixed here are because "b'ar" is now considered
one word.
Case folding rules have a similar mapping style as special casing rules,
where one code point may map to zero or more case folding rules. These
will be used for case-insensitive string comparisons. To see how case
folding can differ from other casing rules, consider "ß" (U+00DF):
>>> "ß".lower()
'ß'
>>> "ß".upper()
'SS'
>>> "ß".title()
'Ss'
>>> "ß".casefold()
'ss'
Unicode declares that to titlecase a string, the first cased code point
after each word boundary should be transformed to its titlecase mapping.
All other codepoints are transformed to their lowercase mapping.
This fixes `combine_hangul_code_points` which would try to combine
a LVT syllable with a trailing consonant, resulting in a wrong
character.
Also added a test for this specific case.
Currently, LibUnicodeData contains the generated UCD and CLDR data. Move
the UCD data to the main LibUnicode library, and rename LibUnicodeData
to LibLocaleData. This is another prepatory change to migrate to
LibLocale.
Each of these strings would previously rely on StringView's char const*
constructor overload, which would call __builtin_strlen on the string.
Since we now have operator ""sv, we can replace these with much simpler
versions. This opens the door to being able to remove
StringView(char const*).
No functional changes.
Our generator is currently preferring the DST variant of the time zone
display names over the non-DST variant. LibTimeZone currently does not
have DST support, and operates in a mode that basically assumes DST does
not exist. Swap the display names for now just to be consistent until we
have DST support.
Note we will need to generate both of these variants and select the
appropriate one at runtime once we have DST support.
The following table in TR-35 includes a web of fall back rules when the
requested time zone style is unavailable:
https://unicode.org/reports/tr35/tr35-dates.html#dfst-zone
Conveniently, the subset of styles supported by ECMA-402 (and therefore
LibUnicode) all either fall back to GMT offset or to a style that is
unsupported but itself falls back to GMT offset.
This adds an API to use LibTimeZone to convert a time zone such as
"America/New_York" to a GMT offset string like "GMT-5" (short form) or
"GMT-05:00" (long form).
The generator parses metaZones.json to form a mapping of meta zones to
time zones (AKA "golden zone" in TR-35). This parser errantly assumed
this was a 1-to-1 mapping.
These were missed in 565a880ce5.
This wasn't an issue because these tests don't pledge/unveil anything,
so they could happily dlopen() the library at runtime. But this is now
needed in order to migrate LibUnicode towards weak symbols instead.
For example, consider the following adjacent entries in UnicodeData.txt:
3400;<CJK Ideograph Extension A, First>;Lo;0;L;;;;;N;;;;;
4DBF;<CJK Ideograph Extension A, Last>;Lo;0;L;;;;;N;;;;;
Our current implementation would assign the display name "CJK Ideograph
Extension A" to code points U+3400 & U+4DBF, but not to the code points
in between. Not only should those code points be assigned a name, but
the Unicode spec also has formatting rules on what the names should be
(the names for these ranged code points are not as they appear in
UnicodeData.txt).
The spec also defines names for code point ranges that actually are
listed individually in UnicodeData.txt. For example:
2F800;CJK COMPATIBILITY IDEOGRAPH-2F800;Lo;0;L;4E3D;;;;N;;;;;
2F801;CJK COMPATIBILITY IDEOGRAPH-2F801;Lo;0;L;4E38;;;;N;;;;;
2F802;CJK COMPATIBILITY IDEOGRAPH-2F802;Lo;0;L;4E41;;;;N;;;;;
Code points are only coalesced into a range if all fields after the name
are equivalent. Our parser will insert the range and its name formatting
pattern when it comes across the first code point in that range, then
ignore other code points in that range. This reduces the number of names
we generated by nearly 2,000.
As noted by ECMA-402, if a supported locale contains all of a language,
script, and region subtag, then the implementation must also support the
locale without the script subtag. The most complicated example of this
is the zh-TW locale.
The list of locales in the CLDR database does not include zh-TW or its
maximized zh-Hant-TW variant. Instead, it inlcudes the zh-Hant locale.
However, zh-Hant-TW is listed in the default-content locale list in the
cldr-core package. This defines an alias from zh-Hant-TW to zh-Hant. We
must then also support the zh-Hant-TW alias without the script subtag:
zh-TW. This transitively maps zh-TW to zh-Hant, which is a case quite
heavily tested by test262.
This file contains the list of locales which default to their parent
locale's values. In the core CLDR dataset, these locales have their own
files, but they are empty (except for identity data). For example:
https://github.com/unicode-org/cldr/blob/main/common/main/en_US.xml
In the JSON export, these files are excluded, so we currently are not
recognizing these locales just by iterating the locale files.
This is a prerequisite for upgrading to CLDR version 40. One of these
default-content locales is the popular "en-US" locale, which defaults to
"en" values. We were previously inferring the existence of this locale
from the "en-US-POSIX" locale (many implementations, including ours,
strip variants such as POSIX). However, v40 removes the "en-US-POSIX"
locale entirely, meaning that without this change, we wouldn't know that
"en-US" exists (we would default to "en").
For more detail on this and other v40 changes, see:
https://cldr.unicode.org/index/downloads/cldr-40#h.nssoo2lq3cba
Previously, LibUnicode would store the values of a keyword as a Vector.
For example, the locale "en-u-ca-abc-def" would have its keyword "ca"
stored as {"abc, "def"}. Then, canonicalization would occur on each of
the elements in that Vector.
This is incorrect because, for example, the keyword value "true" should
only be dropped if that is the entire value. That is, the canonical form
of "en-u-kb-true" is "en-u-kb", but "en-u-kb-abc-true" does not change
for canonicalization. However, we would canonicalize that locale as
"en-u-kb-abc".
Note that the algorithm in the Unicode spec is for checking that a code
point precedes U+0307, but the special casing condition NotBeforeDot is
interested in the inverse of this rule.
Using a file(GLOB) to find all the test files in a directory is an easy
hack to get things started, but has some drawbacks. Namely, if you add
a test, it won't be found again without re-running CMake. `ninja` seems
to do this automatically, but it would be nice to one day stop seeing it
rechecking our globbed directories.
Calendar subtags are a bit of an odd-man-out in that we must match the
variants "ethiopic-amete-alem" in that order, without any other variant
in the locale. So a separate method is needed for this, and we now defer
sorting the variant list until after other canonicalization is done.
Unicode TR35 defines how locale subtag aliases should be emplaced when
converting a locale to canonical form. For most subtags, it is a simple
substitution. Language subtags depend on context; for example, the
language "sh" should become "sr-Latn", but if the original locale has a
script subtag already ("sh-Cyrl"), then only the language subtag of the
alias should be taken ("sr-Latn").
To facilitate this, we now make two passes when canonicalizing a locale.
In the first pass, we convert the LocaleID structure to canonical syntax
(where the conversions all happen in-place). In the second pass, we form
the canonical string based on the canonical syntax.
