zg

zg provides Unicode text processing for Zig projects.

Unicode Version

The Unicode version supported by zg is 16.0.0.

Zig Version

The minimum Zig version required is 0.15.2.

Integrating zg into your Zig Project

You first need to add zg as a dependency in your build.zig.zon file. In your Zig project's root directory, run:

zig fetch --save https://codeberg.org/atman/zg/archive/v0.15.3.tar.gz

Then instantiate the dependency in your build.zig:

constzg=b.dependency("zg",.{});

A Modular Approach

zg is a modular library. This approach minimizes binary file size and memory requirements by only including the Unicode data required for the specified module. The following sections describe the various modules and their specific use case.

Init and Setup

The code examples will show the use of Module.init(allocator) to create the various modules. All of the allocating modules have a setup variant, which takes a pointer and allocates in-place.

Example use:

test"Setup form"{vargraphemes=tryallocator.create(Graphemes);deferallocator.destroy(graphemes);trygraphemes.setup(allocator);defergraphemes.deinit(allocator);}

Code Points

In the code_point module, you'll find a data structure representing a single code point, CodePoint, and an Iterator to iterate over the code points in a string.

In your build.zig:

exe.root_module.addImport("code_point",zg.module("code_point"));

In your code:

constcode_point=@import("code_point");test"Code point iterator"{conststr="Hi 😊";variter:code_point.Iterator=.init(str);vari:usize=0;while(iter.next())|cp|:(i+=1){// The `code` field is the actual code point scalar as a `u21`.if(i==0)tryexpect(cp.code=='H');if(i==1)tryexpect(cp.code=='i');if(i==2)tryexpect(cp.code==' ');if(i==3){tryexpect(cp.code=='😊');// The `offset` field is the byte offset in the// source string.tryexpect(cp.offset==3);tryexpectEqual(cp,code_point.decodeAtIndex(str,cp.offset).?);// The `len` field is the length in bytes of the// code point in the source string.tryexpect(cp.len==4);// There is also a 'cursor' decode, like so:{varcursor=cp.offset;tryexpectEqual(cp,code_point.decodeAtCursor(str,&cursor).?);// Which advances the cursor variable to the next possible// offset, in this case, `str.len`. Don't forget to account// for this possibility!tryexpectEqual(cp.offset+cp.len,cursor);}// There's also this, for when you aren't sure if you have the// correct start for a code point:tryexpectEqual(cp,code_point.codepointAtIndex(str,cp.offset+1).?);}// Reverse iteration is also an option:varr_iter:code_point.ReverseIterator=.init(str);// Both iterators can be peeked:tryexpectEqual('😊',r_iter.peek().?.code);tryexpectEqual('😊',r_iter.prev().?.code);// Both kinds of iterators can be reversed:varfwd_iter=r_iter.forwardIterator();// or iter.reverseIterator();// This will always return the last codepoint from// the prior iterator, _if_ it yielded one:tryexpectEqual('😊',fwd_iter.next().?.code);}}

Note that it's safe to call CodePoint functions on invalid UTF-8. Iterators and decode functions will return the Unicode Replacement Character U+FFFD, according to the Substitution of Maximal Subparts algorithm, for any invalid code unit sequences encountered.

Grapheme Clusters

Many characters are composed from more than one code point. These are known as Grapheme Clusters, and the Graphemes module has a data structure to represent them, Grapheme, and an Iterator and ReverseIterator to iterate over them in a string.

There is also graphemeAtIndex, which returns whatever grapheme belongs to the index; this does not have to be on a valid grapheme or codepoint boundary, but it is illegal to call on an empty string. Last, iterateAfterGrapheme or iterateBeforeGrapheme will provide forward or backward grapheme iterators of the string, from the grapheme provided. Thus, given an index, you can begin forward or backward iteration at that index without needing to slice the string.

In your build.zig:

exe.root_module.addImport("Graphemes",zg.module("Graphemes"));

In your code:

constGraphemes=@import("Graphemes");test"Grapheme cluster iterator"{constgraph=tryGraphemes.init(allocator);defergraph.deinit(allocator);conststr="He\u{301}";// Hévariter=graph.iterator(str);vari:usize=0;while(iter.next())|gc|:(i+=1){// The `len` field is the length in bytes of the// grapheme cluster in the source string.if(i==0)tryexpect(gc.len==1);if(i==1){tryexpect(gc.len==3);// The `offset` in bytes of the grapheme cluster// in the source string.tryexpect(gc.offset==1);// The `bytes` method returns the slice of bytes// that comprise this grapheme cluster in the// source string `str`.tryexpectEqualStrings("e\u{301}",gc.bytes(str));}}}

Words

Unicode has a standard word segmentation algorithm, which gives good results for most languages. Some languages, such as Thai, require a dictionary to find the boundary between words; these cases are not handled by the standard algorithm.

zg implements that algorithm in the Words module. As a note, the iterators and functions provided here will yield segments which are not a "word" in the conventional sense, but word boundaries. Specifically, the iterators in this module will return every segment of a string, ensuring that words are kept whole when encountered. If the word breaks are of primary interest, you'll want to use the .offset field of each iterated value, and handle string.len as the final case when the iteration returns null.

The API is congruent with Graphemes: forward and backward iterators, wordAtIndex, and iterateAfter and before.

In your build.zig:

exe.root_module.addImport("Words",zg.module("Words"));

In your code:

constWords=@import("Words");test"Words"{constwb=tryWords.init(testing.allocator);deferwb.deinit(testing.allocator);constword_str="Metonym Μετωνύμιο メトニム";varw_iter=wb.iterator(word_str);trytesting.expectEqualStrings("Metonym",w_iter.next().?.bytes(word_str));// Spaces are "words" too!trytesting.expectEqualStrings(" ",w_iter.next().?.bytes(word_str));constin_greek=w_iter.next().?;// wordAtIndex doesn't care if the index is valid for a codepoint:for(in_greek.offset..in_greek.offset+in_greek.len)|i|{constat_index=wb.wordAtIndex(word_str,i).bytes(word_str);trytesting.expectEqualStrings("Μετωνύμιο",at_index);}_=w_iter.next();trytesting.expectEqualStrings("メトニム",w_iter.next().?.bytes(word_str));}

Unicode General Categories

To detect the general category for a code point, use the GeneralCategories module.

In your build.zig:

exe.root_module.addImport("GeneralCategories",zg.module("GeneralCategories"));

In your code:

constGeneralCategories=@import("GeneralCategories");test"General Categories"{constgen_cat=tryGeneralCategories.init(allocator);defergen_cat.deinit(allocator);// The `gc` method returns the abbreviated General Category.// These abbreviations and descriptive comments can be found// in the source file `src/GenCatData.zig` as en enum.tryexpect(gen_cat.gc('A')==.Lu);// Lu: uppercase lettertryexpect(gen_cat.gc('3')==.Nd);// Nd: decimal number// The following are convenience methods for groups of General// Categories. For example, all letter categories start with `L`:// Lu, Ll, Lt, Lo.tryexpect(gen_cat.isControl(0));tryexpect(gen_cat.isLetter('z'));tryexpect(gen_cat.isMark('\u{301}'));tryexpect(gen_cat.isNumber('3'));tryexpect(gen_cat.isPunctuation('['));tryexpect(gen_cat.isSeparator(' '));tryexpect(gen_cat.isSymbol('©'));}

Unicode Properties

You can detect common properties of a code point with the Properties module.

In your build.zig:

exe.root_module.addImport("Properties",zg.module("Properties"));

In your code:

constProperties=@import("Properties");test"Properties"{constprops=tryProperties.init(allocator);deferprops.deinit(allocator);// Mathematical symbols and letters.tryexpect(props.isMath('+'));// Alphabetic only code points.tryexpect(props.isAlphabetic('Z'));// Space, tab, and other separators.tryexpect(props.isWhitespace(' '));// Hexadecimal digits and variations thereof.tryexpect(props.isHexDigit('f'));tryexpect(!props.isHexDigit('z'));// Accents, dieresis, and other combining marks.tryexpect(props.isDiacritic('\u{301}'));// Unicode has a specification for valid identifiers like// the ones used in programming and regular expressions.tryexpect(props.isIdStart('Z'));// Identifier start charactertryexpect(!props.isIdStart('1'));tryexpect(props.isIdContinue('1'));// The `X` versions add some code points that can appear after// normalizing a string.tryexpect(props.isXidStart('\u{b33}'));// Extended identifier start charactertryexpect(props.isXidContinue('\u{e33}'));tryexpect(!props.isXidStart('1'));// Note surprising Unicode numeric type properties!tryexpect(props.isNumeric('\u{277f}'));tryexpect(!props.isNumeric('3'));// 3 is not numeric!tryexpect(props.isDigit('\u{2070}'));tryexpect(!props.isDigit('3'));// 3 is not a digit!tryexpect(props.isDecimal('3'));// 3 is a decimal digit}

Letter Case Detection and Conversion

To detect and convert to and from different letter cases, use the LetterCasing module.

In your build.zig:

exe.root_module.addImport("LetterCasing",zg.module("LetterCasing"));

In your code:

constLetterCasing=@import("LetterCasing");test"LetterCasing"{constcase=tryLetterCasing.init(allocator);defercase.deinit(allocator);// Upper and lower case.tryexpect(case.isUpper('A'));tryexpect('A'==case.toUpper('a'));tryexpect(case.isLower('a'));tryexpect('a'==case.toLower('A'));// Code points that have case.tryexpect(case.isCased('É'));tryexpect(!case.isCased('3'));// Case detection and conversion for strings.tryexpect(case.isUpperStr("HELLO 123!"));constucased=trycase.toUpperStr(allocator,"hello 123");deferallocator.free(ucased);tryexpectEqualStrings("HELLO 123",ucased);tryexpect(case.isLowerStr("hello 123!"));constlcased=trycase.toLowerStr(allocator,"HELLO 123");deferallocator.free(lcased);tryexpectEqualStrings("hello 123",lcased);}

Normalization

Unicode normalization is the process of converting a string into a uniform representation that can guarantee a known structure by following a strict set of rules. There are four normalization forms:

Canonical Composition (NFC)

The most compact representation obtained by first decomposing to Canonical Decomposition and then composing to NFC.

Compatibility Composition (NFKC)

The most comprehensive composition obtained by first decomposing to Compatibility Decomposition and then composing to NFKC.

Canonical Decomposition (NFD)

Only code points with canonical decompositions are decomposed. This is a more compact and faster decomposition but will not provide the most comprehensive normalization possible.

Compatibility Decomposition (NFKD)

The most comprehensive decomposition method where both canonical and compatibility decompositions are performed recursively.

zg has methods to produce all four normalization forms in the Normalize module.

In your build.zig:

exe.root_module.addImport("Normalize",zg.module("Normalize"));

In your code:

constNormalize=@import("Normalize");test"Normalize"{constnormalize=tryNormalize.init(allocator);defernormalize.deinit(allocator);// NFC: Canonical compositionconstnfc_result=trynormalize.nfc(allocator,"Complex char: \u{3D2}\u{301}");defernfc_result.deinit(allocator);tryexpectEqualStrings("Complex char: \u{3D3}",nfc_result.slice);// NFKC: Compatibility compositionconstnfkc_result=trynormalize.nfkc(allocator,"Complex char: \u{03A5}\u{0301}");defernfkc_result.deinit(allocator);tryexpectEqualStrings("Complex char: \u{038E}",nfkc_result.slice);// NFD: Canonical decompositionconstnfd_result=trynormalize.nfd(allocator,"Héllo World! \u{3d3}");defernfd_result.deinit(allocator);tryexpectEqualStrings("He\u{301}llo World! \u{3d2}\u{301}",nfd_result.slice);// NFKD: Compatibility decompositionconstnfkd_result=trynormalize.nfkd(allocator,"Héllo World! \u{3d3}");defernfkd_result.deinit(allocator);tryexpectEqualStrings("He\u{301}llo World! \u{3a5}\u{301}",nfkd_result.slice);// Test for equality of two strings after normalizing to NFC.tryexpect(trynormalize.eql(allocator,"foé","foe\u{0301}"));tryexpect(trynormalize.eql(allocator,"foΎ","fo\u{03D2}\u{0301}"));}

The Result returned by normalization functions may or may not be copied from the inputs given. For example, an all-ASCII input does not need to be a copy, and will be a view of the original slice. Calling result.deinit(allocator) will only free an allocated Result, not one which is a view. Thus it is safe to do unconditionally.

This does mean that the validity of a Result can depend on the original string staying in memory. To ensure that your Result is always a copy, you may call try result.toOwned(allocator), which will only make a copy if one was not already made.

Caseless Matching via Case Folding

Unicode provides a more efficient way of comparing strings while ignoring letter case differences: case folding. When you case fold a string, it's converted into a normalized case form suitable for efficient matching. Use the CaseFold module for this.

In your build.zig:

exe.root_module.addImport("CaseFolding",zg.module("CaseFolding"));

In your code:

constCaseFolding=@import("CaseFolding");test"Caseless matching"{// We need Unicode case fold data.constcase_fold=tryCaseFolding.init(allocator);defercase_fold.deinit(allocator);// `compatCaselessMatch` provides the deepest level of caseless// matching because it decomposes fully to NFKD.consta="Héllo World! \u{3d3}";constb="He\u{301}llo World! \u{3a5}\u{301}";tryexpect(trycase_fold.compatCaselessMatch(allocator,a,b));constc="He\u{301}llo World! \u{3d2}\u{301}";tryexpect(trycase_fold.compatCaselessMatch(allocator,a,c));// `canonCaselessMatch` isn't as comprehensive as `compatCaselessMatch`// because it only decomposes to NFD. Naturally, it's faster because of this.tryexpect(!trycase_fold.canonCaselessMatch(allocator,a,b));tryexpect(trycase_fold.canonCaselessMatch(allocator,a,c));}

Case folding needs to use the Normalize module in order to produce the compatibility forms for comparison. If you are already using a Normalize for other purposes, CaseFolding can borrow it:

constCaseFolding=@import("CaseFolding");constNormalize=@import("Normalize");test"Initialize With a Normalize"{constnormalize=tryNormalize.init(allocator);// You're responsible for freeing this:defernormalize.deinit(allocator);constcase_fold=tryCaseFolding.initWithNormalize(allocator,normalize);// This will not free your normalize when it runs first.defercase_fold.deinit(allocator);}

This has a setupWithNormalize variant as well, note that this also takes a Normalize struct, and not a pointer to it.

Display Width of Characters and Strings

When displaying text with a fixed-width font on a terminal screen, it's very important to know exactly how many columns or cells each character should take. Most characters will use one column, but there are many, like emoji and East- Asian ideographs that need more space. The DisplayWidth module provides methods for this purpose. It also has methods that use the display width calculation to center, padLeft, padRight, and wrap text.

In your build.zig:

exe.root_module.addImport("DisplayWidth",zg.module("DisplayWidth"));

In your code:

constDisplayWidth=@import("DisplayWidth");test"Display width"{constdw=tryDisplayWidth.init(allocator);deferdw.deinit(allocator);// String display widthtryexpectEqual(@as(usize,5),dw.strWidth("Hello\r\n"));tryexpectEqual(@as(usize,8),dw.strWidth("Hello 😊"));tryexpectEqual(@as(usize,8),dw.strWidth("Héllo 😊"));tryexpectEqual(@as(usize,9),dw.strWidth("Ẓ̌á̲l͔̝̞̄̑͌g̖̘̘̔̔͢͞͝o̪̔T̢̙̫̈̍͞e̬͈͕͌̏͑x̺̍ṭ̓̓ͅ"));tryexpectEqual(@as(usize,17),dw.strWidth("슬라바 우크라이나"));// Centering textconstcentered=trydw.center(allocator,"w😊w",10,"-");deferallocator.free(centered);tryexpectEqualStrings("---w😊w---",centered);// Pad leftconstright_aligned=trydw.padLeft(allocator,"abc",9,"*");deferallocator.free(right_aligned);tryexpectEqualStrings("******abc",right_aligned);// Pad rightconstleft_aligned=trydw.padRight(allocator,"abc",9,"*");deferallocator.free(left_aligned);tryexpectEqualStrings("abc******",left_aligned);// Wrap textconstinput="The quick brown fox\r\njumped over the lazy dog!";constwrapped=trydw.wrap(allocator,input,10,3);deferallocator.free(wrapped);constwant=\\The quick\\brown fox\\jumped\\over the\\lazy dog!;tryexpectEqualStrings(want,wrapped);}

This module has build options. The first is cjk, which will consider ambiguous characters as double-width.

To choose this option, add it to the dependency like so:

constzg=b.dependency("zg",.{.cjk=true,});

The other options are c0_width and c1_width. The standard behavior is to treat C0 and C1 control codes as zero-width, except for delete and backspace, which are -1 (the logic ensures that a strWidth is always at least 0). If printing control codes with replacement characters, it's necessary to assign these a width, hence the options. When provided these values must fit in an i4, this allows for C1s to be printed as \u{80} if desired.

DisplayWidth uses the Graphemes module internally. If you already have one, it can be borrowed using DisplayWidth.initWithGraphemes(allocator, graphemes) in the same fashion as shown for CaseFolding and Normalize.

Scripts

Unicode categorizes code points by the Script in which they belong. A Script collects letters and other symbols that belong to a particular writing system. You can detect the Script for a code point with the Scripts module.

In your build.zig:

exe.root_module.addImport("Scripts",zg.module("Scripts"));

In your code:

constScripts=@import("Scripts");test"Scripts"{constscripts=tryScripts.init(allocator);deferscripts.deinit(allocator);// To see the full list of Scripts, look at the// `src/Scripts.zig` file. They are list in an enum.tryexpect(scripts.script('A')==.Latin);tryexpect(scripts.script('Ω')==.Greek);tryexpect(scripts.script('צ')==.Hebrew);}

Limits

Iterators, and fragment types such as CodePoint, Grapheme and Word, use a u32 to store the offset into a string, and the length of the fragment (CodePoint uses a u3 for length, actually).

4GiB is a lot of string. There are a few reasons to work with that much string, log files primarily, but fewer to bring it all into memory at once, and practically no reason at all to do anything to such a string without breaking it into smaller piece to work with.

Also, Zig compiles on 32 bit systems, where usize is a u32. Code running on such systems has no choice but to handle slices in smaller pieces. In general, if you want code to perform correctly when encountering multi-gigabyte strings, you'll need to code for that, at a level one or two steps above that in which you'll want to, for example, iterate some graphemes of that string.

That all said, zg modules can be passed the Boolean config option fat_offset, which will make all of those data structures use a u64 instead. I added this option not because you should use it, which you should not, but to encourage awareness that code operating on strings needs to pay attention to the size of those strings, and have a plan for when sizes get out of specification. What would your code do with a 1MiB region of string with no newline? There are many questions of this nature, and robust code must detect when data is out of the expected envelope, so it can respond accordingly.

Code which does pay attention to these questions has no need for u64 sized offsets, and code which does not will not be helped by them. But perhaps yours is an exception, in which case, by all means, configure accordingly.