[Unicode]  Technical Reports
 

DRAFT - L2/02-xxx

Proposed Draft
Unicode Technical Report #XX

Assessing Unicode Support

Version 4
Authors Mark Davis (mark.davis@us.ibm.com)
Date 2002-10-29
This Version n/a
Previous Version n/a
Latest Version n/a
Tracking Number 1

Summary

This document describes guidelines for testing programs and systems to see if they support Unicode, and the level of support that they offer.

Status

[Boilerplate here]

Contents

1 Introduction

In today's world, software components must interact with a wide variety of other components. Systems often consist of components running on different machines and different platforms, all communicating with one another in complex ways. Unicode is fundamental to providing seamless support of all world languages, and it will appear in many different products: from operating systems to databases, from digital cameras to online games. When assembling systems, it is crucial to be able to ensure that all the different components of a system support Unicode; otherwise weak links may degrade or disable the internationalization support offered by the system as a whole.

This document describes guidelines for testing programs and systems to see if they support Unicode, and if so, to determine the level of support that they offer. These guidelines explicitly do not test for general internationalization or localization capabilities; those are out of scope for this document.

Because Unicode is such a fundamental technology, any tests for Unicode capabilities must be tailored to the specific type of product. Moreover, many of the requirements for Unicode support are only applicable to particular products. BIDI conformance, for example, may not be applicable if the product never displays text, but only processes it. Thus all of the following guidelines can only be applied to products that support or require the relevant kinds of processing described in each section.

1.1 Conformance Issues

Any assessment of Unicode support must start with conformance to the Unicode Standard itself. The Unicode Standard is a very large and many-faceted standard. Because of this, and because of the nature of the material in the standard, it may not be clear how to test for conformance to the Unicode Standard.

A conformance test for the Unicode Standard is a list of data certified by the UTC to be "correct" in regard to some particular requirement for conformance to the standard. In some instances, as for example, the implementation of the bidirectional algorithm, producing a definitive list of correct results is difficult or impossible, and in such cases, a conformance test may itself consist of an implemented algorithm certified by the UTC to produce correct results for any pertinent input data. Conformance tests for the Unicode Standard are essentially benchmarks that someone can use to determine if their algorithm, API, etc., claiming to conform to some requirement of the standard, does in fact match the data that the UTC claims defines such conformance.

Some formal standards are developed once and then are essentially frozen and stable forever. For such standards, stability of content and the corresponding stability of conformance claims is not an issue. For a large, complex standard aimed at the universal encoding of characters, such as the Unicode Standard, such stability is not possible. The standard is necessarily evolving and expanding over time, to extend its coverage of all the writing systems of the world. And as experience in its implementation accumulates, further aspects of character processing also accrue to the formal content of the standard. This fundamentally dynamic quality of the Unicode Standard complicates issues of conformance, since the content to which conformance requirements pertain continually expands, both horizontally to more characters and scripts, and vertically to more aspects of character processing.

The Unicode Standard is regularly versioned, as new characters are added. A formal system of versioning is in place, involving major, minor, and update versions, all with carefully controlled rules for the type of documentation required, handling of the associated data files, and allowable types of change between versions. For more information about the details of Unicode versioning see [Versions]. Conformance claims clearly must be specific to versions of the Unicode Standard, but the level of specificity needed for a claim may vary according to the nature of the particular conformance claim being made.

Systems may also claim conformance to specifications that are outside of the Unicode Standard proper, such as to the Unicode Collation Algorithm, or to Unicode Regular Expression Level 1. Unicode-conformant software does not need to implement these features: however, if a system purports to support one of these features then they can be tested for. This document does include tests for such features, but they are marked as "optional", being not required for Unicode Standard conformance.

1.2 Guidelines

Because the criteria for conformance to the Unicode Standard apply to a wide range of possible systems, they sometimes do not require the same level of quality in behavior or display that one would require of a production system. For example, a badly-drawn low-resolution depiction of an 'a' is conformant, but would not be acceptable in practice.

In many cases below, precise tests cannot be formulated, since the types of processes are so varied. In such cases, for example for Protocols, examples are given of the types of tests that can be formulated.

In a number of cases, examples are formulated using code snippets. These snippets are only examples; there is no implication about the use of any particular programming language, nor that any particular syntax is better or worse than any other; these are only examples.

2 Basic Requirements

The most fundamental requirements for Unicode support are the following:

2.2 Basic Requirements Tests

Tests for Roundtrip testing of these are fairly straightforward with components that store and retrieve data, such as databases. Here is one example:

Build a small table, insert Unicode data, select the data from the table and compare the results. For instance, use the following SQL statements to create a table named "langs", insert data, select all data and search for one record:

SQL Statements Results
drop table langs; The SQL command completed successfully.
create table langs (L1 character(10), L2 varchar(18)); The SQL command completed successfully.
insert into langs values ('Russian', 'русский'); The SQL command completed successfully.
insert into langs values ('Spanish', 'Español'); The SQL command completed successfully.
insert into langs values ('Czech', 'čeština'); The SQL command completed successfully.
insert into langs values ('Greek', 'ελληνικά'); The SQL command completed successfully.
insert into langs values ('Japanese', '日本語'); The SQL command completed successfully.
insert into langs values ('Vietnamese', 'Tiểng Việt'); The SQL command completed successfully.
select * from langs; L1         L2
---------- ------------------
Russian   
русский
Spanish   
Español
Czech     
čeština
Greek     
ελληνικά
Japanese  
日本語
Vietnamese
Tiểng Việt

6 record(s) selected.

select * from langs where L2 like '%λη%'; L1         L2
---------- ------------------
Greek     
ελληνικά

1 record(s) selected.
 

3 Character Conversion

This section is optional, since not every product does--or needs to do--conversion. However, if a product does do conversion, here are the areas to test for:

3.1 Character Conversion Tests

For ISO/IEC 8859 tests, download the files in http://www.unicode.org/Public/MAPPINGS/ISO8859/. The files are of the following format, with two significant fields: the first is a byte, and the second is a code point.

0x00 0x0000 # NULL
...
0xFF 0x00FF # LATIN SMALL LETTER Y WITH DIAERESIS
  1. Confirm that the converter converts each of the bytes in the first field to the code point in the second field, and back.
  2. Generate a random selection of code points that are outside of the values in the second field, and convert them. This should at least include U+0212, U+FFFD, U+FFFF, U+10FFFD, and U+10FFFF. Confirm that these generate one of the following (depending on converter options):
    • a substitution character (e.g. 0x1A)
    • an escape (e.g. Ȓ)
    • an error
  3. If the converter distinguishes between illegal (source) values and unassigned values (in the target set), verify that the appropriate responses are generated:
    • unassigned: U+0212, U+FFFD, U+10FFFD
    • illegal: U+FFFF, U+10FFFF

For compression tests and UTF tests (and if CESU-8 is supported), for each converter:

  1. Verify that every code point from U+0000 to U+10FFFF converts to the UTF and back, returning the same results (e.g. round-trips).
  2. For real text samples, download the text of the files in http://www.unicode.org/standard/WhatIsUnicode.html. Verify that they also round-trip.
  3. Verify that 'code point' values from 0x110000 to 0xFFFFFFFF (incrementing by 0x12345) are treated as illegal.
  4. Verify that illegal code unit sequences treated as illegal.
    • UTF-32: the numeric values in #3
    • UTF-8: (based on Table 3.1B in UAX #28)
      • 1st bytes 80..C1, F5..FF
      • 2nd-4th bytes outside of the ranges given in 3.1B, according to each 1st byte.
  5. Verify correct conversion
    • UTF-8: for each range in the left side of 3.1B, the lowest range in the subsequent table is produced.
      • e.g. U+1000 => E1, 80
    • UTF-16*: U+10000 => D800, DC00; U+10FFFF => DBFF, DFFF
  6. BOM:
    • In the following table, convert the Bytes column according to the Encoding. The result should match the Code Points column. Note: There is no mechanism in Unicode to determine whether an EF BB BF at the start of UTF-8 represents a BOM or not; there is no different charset name to indicate that. The following test assumes that it does represent a BOM.
Bytes Encoding Code Points
EF BB BF E1 88 B4 UTF-8 1234
EF BB BF E1 88 B4 UTF-16/LE/BE EFBB BFE1 88B4
EF BB BF E1 88 B4 UTF-32/LE/BE error
FE FF 12 34 UTF-16 1234
FE FF 12 34 UTF-16BE FEFF 1234
FE FF 12 34 UTF-16LE/UTF-32* error
FF FE 34 12 UTF-16 1234
FF FE 34 12 UTF-16LE FEFF 1234
FF FE FF FE 34 12 UTF-16 FEFF 1234
FE FF FE FF 12 34 UTF-16 FEFF 1234

 

4 Protocol Guidelines

Unlike some of the other sections, no specific tests are available for this section. Moreover, only general guidelines can be described for protocols, including the following:

4.1 Protocol Guideline Sample Tests

SMTP (with/without MIME) is given as a simple example. For SMTP, there are sending and receiving clients easily available: email applications like Outlook Express and Netscape Messenger.

Proposed test scenario using such a client program:

  1. Create a new document/email.
  2. Set the format to either plain text to test SMTP just by itself or to rich text to test SMTP+MIME.
  3. Set the encoding of the email to UTF-8.
  4. Include in the email body characters from multiple scripts, e.g., Latin, Cyrillic, Arabic, Hindi, Hiragana, Han ideographs, Deseret.
  5. Send the email so that it is processed by the test object.

Sample text for the email body:

Latin: U+00FE ð
Cyrillic: U+0436 ж
Arabic: U+0628 ب
Hindi: U+0905 अ
Hiragana: U+3042 あ
Han Ideograph: U+4E0A 上
Deseret (plane 1): U+1040C А?
Han Ideograph (plane 2): U+20021 ࠀ?

Test of an SMTP server:

Verify that the email contents is preserved when stored+forwarded through this server.
Requires proper configuration of the email client/network.
In this case, as with some other protocols, the server will almost always just pass the contents through. The test will thus just verify that the server is 8-bit clean, which is almost always the case.

Test of an SMTP client:

Send the email to an address that is handled by a particular client program. Make sure that the text is fully preserved and displayed in a reasonable way (given available fonts etc.).
Example for email clients that are expected to have problems in this area: Eudora, Netscape 4.x (do not use Unicode internally, so must convert to subset-charsets).

Test of an email gateway:

Some email systems (Lotus Notes, X.500, VM) use other protocols than SMTP and transform emails between SMTP and their own formats. Send emails into such systems, forward/reply them back to a globalization-capable client and verify full roundtrip of the text. The following are examples of gateways/systems that are expected to have problems: VM (EBCDIC encodings have a subset of the Unicode repertoire).

Note: Lotus Notes is globalization-capable (should pass the test) because LMBCS can encapsulate Unicode; it will not fully roundtrip arbitrary MIME/HTML formatting, but this is out of scope for G11N certification. All of the characters should roundtrip.

Test of a non-SMTP email client:

A non-SMTP email client would have to get the test email through such a gateway. It is possible that the client may show an email with higher or lower fidelity compared with the roundtrip test into and out of the gateway. Higher if only the second part of the roundtrip were to lose information. Lower if the roundtrip can preserve some or all of the original contents in a form that is not displayed in the non-SMTP client.

Other ways to test protocols

With many IETF (Internet) protocols it is possible to test at least some of the protocol elements using a telnet client or a special-purpose client (e.g., Java application reading/writing to sockets) by reading and writing plain text streams directly, and using UTF-8 text for the contents.

Generally, it may be necessary to write custom test clients/servers to perform meaningful tests of a protocol at all or to automate such tests.

Some protocols (like HTTP) allow many more charsets in direct use than SMTP. "Direct use" means that UTF-16 is possible in SMTP emails only after a base64-transformation (or quoted-printable), while HTTP allows the contents to be encoded in UTF-16 directly in the byte stream.

SOAP Tests

SOAP is an XML vocabulary being defined by W3C. A SOAP message consists of SOAP envelope, SOAP header and SOAP body. The SOAP body contains user data which is used for RPC function.

<SOAP-ENV:Envelope>
 <SOAP-ENV:Header>
  Additional Information for SOAP message transmission
 </SOAP-ENV:Header>
 <SOAP-ENV:Body>
  Body data of SOAP-RPC message transmission
 </SOAP-ENV:Body>
</SOAP-ENV:Envelope>
Sample 1.

SOAP request between service requester and UDDI service provider

  1. All XML char MUST be encoded in UTF-8 in SOAP envelope because UDDI SOAP interface MUST use UTF-8.
  2. MIME MUST be "Content-type : text/xml; charset=UTF-8" as http header.
  3. Max name data is FIVE according to the UDDI V2.0 specification.
POST /uddisoap/publishapi HTTP/1.1
Host: abc.def.com
Content-Type: text/xml; charset=utf-8
Content-Length: nnn
SOAPAction: ""

<?xml version="1.0" encoding="UTF-8" ?>
<SOAP-ENV:Envelope xmlns:SOAP-ENV="http://schemas.xmlsoap.org/soap/envelope/">
  <SOAP-ENV:Body>  
    <save_business generic="2.0" xmlns="urn:uddi-org:api_v2">    
      <authInfo>uddiUser</authInfo>
      <businessEntity businessKey="">
         <name xml:lang="ru">русский</name>
         <name xml:lang="cs">čeština</name>
         <name xml:lang="el">ελληνικά</name>
         <name xml:lang="ja">日本語</name>
         <name xml:lang="vi">Tiểng Việt</name>
      </businessEntity>  
    </save_business>
  </SOAP-ENV:Body>
</SOAP-ENV:Envelope>
Sample 2.

The expected SOAP message from UDDI service provider for Example 1 above.

  1. All XML characters MUST be encoded in UTF-8 in SOAP envelope because the UDDI SOAP interface MUST use UTF-8.
  2. MIME MUST be "Content-type : text/xml; charset=UTF-8" as http header.
HTTP/1.1 200 OK
Server: ABC
Content-Type: text/xml; charset="utf-8"
Content-Length: nnnn
Connection: close

<?xml version="1.0" encoding="UTF-8" ?>
  <SOAP-ENV:Envelope xmlns:SOAP-ENV="http://schemas.xmlsoap.org/soap/envelope/">
    <SOAP-ENV:Body>
     <businessDetail generic="2.0" xmlns="urn:uddi-org:api_v2" operator="operator">
       <businessEntity businessKey="14821BDD-00EA-4398-8003-24BC35F0394A" operator="operator" authorizedName="uddiUser">
        <discoveryURLs>
          <discoveryURL useType="businessEntity">http://abc.def.com:9080/uddisoap/get?businessKey=14821BDD-00EA-4398-8003-24BC35F0394A
          </discoveryURL>
         </discoveryURLs>
         <name xml:lang="ru">русский</name>
         <name xml:lang="cs">čeština</name>
         <name xml:lang="el">ελληνικά</name>
         <name xml:lang="ja">日本語</name>
         <name xml:lang="vi">Tiểng Việt</name>
      </businessEntity>
    </businessDetail>
  </SOAP-ENV:Body>
</SOAP-ENV:Envelope>
Sample 3.

Java program (SaveBusinessExample.java) using SOAP interface in UDDI4J to generate sample 1.

5 Programming Support

Programming Language support includes both the basic programming language, and libraries that supplement the basic support with additional functionality. Thus, for example, even though the basic support in C for Unicode is fairly rudimentary, there are supplementary libraries that provide full-featured Unicode support. While it would be more efficient and more interoperable if C had the capabilities discussed in section 5.1 below, it is certainly possible to work around those limitations in providing Unicode support based on C. Thus this section is optional.

Typically, code units are treated as primitives (8/16/32-bit unsigned integers); the values are verified at certain defined points in processing. There are two broad strategies for handling this:

For more information on the distinction between Unicode strings and Unicode encoding forms, see Chapters 2 and 3 of The Unicode Standard.

5.1 Language Support

The fundamental requirements for good programming language support include the following. The names of the datatypes are not important; the ones given below are only examples.

UTF-32 datatype: A Unicode code point datatype whose value space encompasses the entire repertoire of Unicode code points, from U+0000 to U+10FFFF inclusive. For example:

UTF32_t  cp1 = '\U0001D434';
UTF32_t  cp2 = '\u1234';
UTF32_t* sp1 = "ABC \u5678 \U0002F884";

UTF-16 datatype: With the wide variety of Unicode libraries and operating system functions using 16-bit Unicode strings, for interoperability it is incumbent upon a programming language to also supply a UTF-16 datatype, one that contains 16 unsigned bits encompassing a range of integral values from 0x0000 to 0xFFFF inclusive. For example:

utf16_t  cu1 = '\u1234';
utf16_t* su2 = "ABC \u5678 \U0002F884";

Literals: Although not formally necessary (since any localizable text will be in resources), for full support a language will provide character literal and string literal representations with UTF32 and UTF16 literals, as in the above examples.

5.2 Standard and Supplemental Libraries

Testing for full internationalization support is beyond the scope of this document, but the language (supplemented by libraries) can be tested for the following.

5.3 Development Environments

Integrated Development Environments (IDEs) or the the tools that can be part of a development environment, are subject to the general  requirements Unicode GUIs. Particular features to watch for are:

5.4 Programming Support Tests

The StringTest.txt file contains machine-readable tests for code point operations. These can be used for iteration and extraction.

To test to see if all Unicode identifiers are supported, access the DerivedCoreProperties.txt file. Write a file that has each of the following strings in the context of an identifier. Verify that the resulting file (or files: they may need to be broken up to get past compiler memory limitations) can be successfully compiled and linked.

This is also a good test of string literals. For example:

utf16_t* a = "a"; // U+0061 ( a )
...
utf16_t* α = "α"; // U+03B1 ( α ) 
...
utf16_t* 串 = "串"; // U+4E32 ( 串 )
...
utf16_t*  = ""; // U+10400 (  )
...
utf16_t*  = ""; // U+0061 ( a ) U+0316 ( ◌̖ )
...

6 Analysis

Analysis includes character properties, regular expressions, and boundaries (grapheme cluster, word, line sentence breaks). In this area, typically the tests will check against the UCD properties, plus the guidelines for how those properties are used. The exact formulation of the test will depend on the API and language involved.

Caution: For the English language, functions such as isLetter() are sufficient for a variety of tasks, such as word-break. With the wide variety of languages, scripts, and types of characters supported by the Unicode Standard, this is not true. The presence of non-spacing marks in Arabic, for example, will cause any naïve use of such functions to give incorrect results. More sophisticated mechanisms must be used for determining such tasks.

Note: It is perfectly conformant to supply additional, tailored behavior (such as the results of property APIs, or different word breaks) that is different than the Unicode default behavior, as long as such behavior does not purport to follow the Unicode default specifications.

The main features to test for are the following.

6.1 Analysis Tests

For testing Unicode properties, a small test program should be written that for each property:

For regular expressions, UTR #18 provides 3 levels for regular expressions. The feature sets in these levels can be tested for explicitly. Note: the TR does not require any particular syntax, so any tests have to be adapted to the syntax of the regular expression engine.

For case detection, test with the following file [TBD]. In addition, verify that the functions respect canonical equivalence by applying all functions to each field in NormalizationTest.txt, and verifying that the same answer is produced.

For grapheme-cluster, word, line and sentence boundaries, the following tests can be used.

Common APIs will test a particular offset to see if it is a boundary, and also iterate (e.g. find the next boundary). Verify that both APIs provide the same results on all of the test cases, by iterating over each test case, and independently determining the boundaries one at a time, then comparing the two sets of results.

The Well-Formed tests, as mentioned above, are rarely worth testing for. However, if those features are important for a particular application, the following can be used.

7 Comparison

Comparison includes both binary comparison, and comparison based on UCA (UTS #10). In the latter case, it includes string comparison, string search, and sortkey generation. In the case of Collation (and the related StringSearch), only the default collation ordering can be tested, since there is no accepted repository of machine-readable tailorings for different languages.

Note: It is perfectly conformant to supply additional, tailored behavior (such as the results of collation ordering) that is different than the Unicode default behavior, as long as such behavior does not purport to follow the Unicode default specifications.

Binary comparison works by lexically comparing strings. The first string unit difference "wins". Unicode has three encoding forms for processing: UTF-8/16/32. Software needs to be able to perform either comparison regardless of its native Unicode encoding form to achieve the same binary order for sorted data structures (lists, trees, etc.) as other software in a connected system. For example, Java Servlets vs. UTF-8 database.

7.1 Comparison Tests

Binary Comparison: There are two common binary comparison orders: UTF-16 order and code point order. The test file BinaryComparisonTest.txt has the following format:
<string1> ; <string2> ; <code point relation> ; <UTF-16 relation>

For example:

0061; 0062; LESS; LESS;
FFFF; 10FFFF; LESS; GREATER;
FFFF; 10FFFF; LESS; GREATER;

Collation (UCA UTS #10): If the process purports to support the UCA, verify the default collation sequence using the test files in http://www.unicode.org/unicode/reports/tr10/#Test. If both sortkey generation and

String Search: Verify that the locale-sensitive string search functions follow the UCA, according to StringSearchTest.txt. Note: This needs to be fleshed out more.

Case-Insensitive Compare: Verify that that the results follow the guidelines in UAX #21

8 Transformations

Transformations are functions that take a string as input, and produce a (perhaps) modified string as output. They include case conversion and normalization. These are described in detail in UAX #21 and UAX #15.

Note: It is perfectly conformant to supply additional, tailored behavior (such as the results of case folding) that is different than the Unicode default behavior, as long as such behavior does not purport to follow the Unicode default specifications. However, it must be clear to programmers and end users that the default Unicode behavior is not being followed.

8.1 Transformation Tests

The following tests can be used.

References

[FAQ] Unicode Frequently Asked Questions
http://www.unicode.org/unicode/faq/
For answers to common questions on technical issues.
[Glossary] Unicode Glossary
http://www.unicode.org/glossary/
For explanations of terminology used in this and other documents.
[Reports] Unicode Technical Reports
http://www.unicode.org/unicode/reports/
For information on the status and development process for technical reports, and for a list of technical reports.
[U3.1] Unicode Standard Annex #27: Unicode 3.1
http://www.unicode.org/unicode/reports/tr27/
[Versions] Versions of the Unicode Standard
http://www.unicode.org/unicode/standard/versions/
For details on the precise contents of each version of the Unicode Standard, and how to cite them.

Acknowledgements

Thanks to Helena Shih Chapman, Julius Griffith, Markus Scherer, Baldev Soor, Akio Kido, Kentaroh Noji, Takaaki Shiratori, Xiao Hu Zhu, Geng Zheng, CP Chang, Matitiahu Allouche, Tarek Abou Aly, Ranat Thopunya, and Israel Gidali for their many contributions to this document.

Modifications

The following summarizes modifications from the previous version of this document.

  1. Incorporated feedback from UTC
  2. Reworded conformance sections
  3. Dropped keyboard input and rendering section

Previous Modifications

  1. Changed focus of document to assessing Unicode support, and added more qualifying language about tailorings and defaults.
  2. Added substantial new sections on:
  3. Added new programmatic tests, in particular:
  4. Added new draft rendering/input tests for Arabic, Hebrew, and Thai.
  5. Substantial rework of the text in general.
  6. Removed all conversions except UTFs and 8859-* from 3.1 Character Conversion Tests
  7. Broke 5 Programming_Support into separate sections, and added to the test section.