ACS 4 Globalization Requirements

by Henry Minsky, Yon Feldman, Lars Pind, others


I. Introduction

This document describes the requirements for functionality in the ACS platform to support globalization of the core and optional modules. The goal is to make it possible to support delivery of applications which work properly in multiple locales with the lowest development and maintenance cost.

Definitions

internationalization (i18n)

The provision within a computer program of the capability of making itself adaptable to the requirements of different native languages, local customs and coded character sets.

locale

The definition of the subset of a user's environment that depends on language and cultural conventions.

localization (L10n)

The process of establishing information within a computer system specific to the operation of particular native languages, local customs and coded character sets.

globalization

A product development approach which ensures that software products are usable in the worldwide markets through a combination of internationalization and localization.

II. Vision Statement

The Mozilla project suggests keeping two catchy phrases in mind when thinking about globalization:

Building an application often involves making a number of assumptions on the part of the developers which depend on their own culture. These include constant strings in the user interface and system error messages, names of countries, cities, order of given and family names for people, syntax of numeric and date strings and collation order of strings.

The ACS should be able to operate in languages and regions beyond US English. The goal of ACS Globalization is to provide a clean and efficient way to factor out the locale dependent functionality from our applications, in order to be able to easily swap in alternate localizations.

This in turn will reduce redundant, costly, and error prone rework when targeting the toolkit or applications built with the toolkit to another locale.

The cost of porting the ACS to another locale without some kind of globalization support would be large and ongoing, since without a mechanism to incorporate the locale-specific changes cleanly back into the code base, it would require making a new fork of the source code for each locale.

III. System/Application Overview

A globalized application will perform some or all of the following steps to handle a page request for a specific locale:

  1. Decide what the target locale is for an incoming page request

  2. Decide which character set encoding the output should be delivered in

  3. If a script file to handle the request needs to be loaded from disk, determine if a character set conversion needs to be performed when loading the script

  4. If needed, locale-specific resources are fetched. These can include text, graphics, or other resources that would vary with the target locale.

  5. If content data is fetched from the database, check for locale-specific versions of the data (e.g. country names).

  6. Source code should use a message catalog API to translate constant strings in the code to the target locale

  7. Perform locale-specific linguistic sorting on data if needed

  8. If the user submitted form input data, decide what character set encoding conversion if any is needed. Parse locale-specific quantities if needed (number formats, date formats).

  9. If templating is being used, select correct locale-specific template to merge with content

  10. Format output data quantities in locale-specific manner (date, time, numeric, currency). If templating is being used, this may be done either before and/or after merging the data with a template.

Since the internationalization APIs may potentially be used on every page in an application, the overhead for adding internationalization to a module or application must not cause a significant time delay in handling page requests.

In many cases there are facilities in Oracle to perform various localization functions, and also there are facilities in Java which we will want to move to. So the design to meet the requirements will tend to rely on these capabilities, or close approximations to them where possible, in order to make it easier to maintain Tcl and Java ACS versions.

IV. Use-cases and User-scenarios

Here are the cases that we need to be able to handle efficiently:

  1. A developer needs to author a web site/application in a language besides English, and possibly a character set besides ISO-8859-1. This includes the operation of the ACS itself, i.e., navigation, admin pages for modules, error messages, as well as additional modules or content supplied by the web site developer.

    What do they need to modify to make this work? Can their localization work be easily folded in to future releases of ACS?

  2. A developer needs to author a web site which operates in multiple languages simultaneously. For example, arsDigita.com with content and navigation in English, German, and Japanese.

    The site would have an end-user visible UI to support these languages, and the content management system must allow articles to be posted in these languages. In some cases it may be necessary to make the modules' admin UI's operate in more than one supported language, while in other cases the backend admin interface can operate in a single language.

  3. A developer is writing a new module, and wants to make it easy for someone to localize it. There should be a clear path to author the module so that future developers can easily add support for other locales. This would include support for creating resources such as message catalogs, non-text assets such as graphics, and use of templates which help to separate application logic from presentation.

Competitive Analysis

Other application servers: ATG Dyanmo, Broadvision, Vignette, ... ? Anyone know how they deal with i18n ?

V. Related Links

VI Requirements

Because the requirements for globalization affect many areas of the system, we will break up the requirements into phases, with a base required set of features, and then stages of increasing functionality.

VI.A Locales

10.0 A standard representation of locale will be used throughout the system. A locale refers to a language and territory, and is uniquely identified by a combination of ISO language and ISO country abbreviations.

See Content Repository Requirement 100.20

10.10 Provide a consistent representation and API for creating and referencing a locale

10.20 There will be a Tcl library of locale-aware formatting and parsing functions for numbers, dates and times. Note that Java has builtin support for these already.

10.30 For each locale there will be default date, number and currency formats.

VI.B Associating a Locale with a Request

20.0 The request processor must have a mechanism for associating a locale with each request. This locale is then used to select the appropriate template for a request, and will also be passed as the locale argument to the message catalog or locale-specific formatting functions.

20.10 The locale for a request should be computed by the following method, in descending order of priority:

  • get locale associated with subsite or package id
  • get locale from user preference
  • get locale from site wide default

    20.20 An API will be provided for getting the current request locale from the ad_conn structure.

  • VI.C Resource Bundles / Content Repository

    30.0 A mechanism must be provided for a developer to group a set of arbitrary content resources together, keyed by a unique identifier and a locale.

    For example, what approaches could be used to implement a localizable nav-bar mechanism for a site? A navigation bar might be made up of a set of text strings and graphics, where the graphics themselves are locale-specific, such as images of English or Japanese text (as on www.arsdigita.com). It should be easy to specify alternate configurations of text and graphics to lay out the page for different locales.

    Design note: Alternative mechanisms to implement this functionality might include using templates, Java ResourceBundles, content-item containers in the Content Repository, or some convention assigning a common prefix to key strings in the message catalog.

    VI.D Message Catalog for String Translation

    40.0 A message catalog facility will provide a database of translations for constant strings for multilingual applications. It must support the following:

    40.10 Each message will referenced via unique a key.

    40.20 The key for a message will have some hierarchical structure to it, so that sets of messages can be grouped with respect to a module name or package path.

    40.30 The API for lookup of a message will take a locale and message key as arguments, and return the appropriate translation of that message for the specifed locale.

    40.40 The API for lookup of a message will accept an optional default string which can be used if the message key is not found in the catalog. This lets the developer get code working and tested in a single language before having to initialize or update a message catalog.

    40.50 For use within templates, custom tags which invoke the message lookup API will be provided.

    40.60 Provide a method for importing and exporting a flat file of translation strings, in order to make it as easy as possible to create and modify message translations in bulk without having to use a web interface.

    40.70 Since translations may be in different character sets, there must be provision for writing and reading catalog files in different character sets. A mechanism must exist for identifying the character set of a catalog file before reading it.

    40.80 There should be a mechanism for tracking dependencies in the message catalog, so that if a string is modified, the other translations of that string can be flagged as needing update.

    40.90 The message lookup must be as efficient as possible so as not to slow down the delivery of pages.


    Design question: Is there any reason to implement the message catalog on top of the content repository as the underlying storage and retrieval service, with a layer of caching for performance? Would we get a nice user interface and version control almost for free?

    VI.E Character Set Encoding

    Character Sets

    50.0 A locale will have a primary associated character set which is used to encode text in the language. When given a locale, we can query the system for the associated character set to use.

    The assumption is that we are going to use Unicode in our database to hold all text data. Our current programming environments (Tcl/Oracle or Java/Oracle) operate on Unicode data internally. However, since Unicode is not yet commonly used in browsers and authoring tools, the system must be able to read and write other character sets. In particular, conversions to and from Unicode will need to be explicitly performed at the following times:


    Design question: Do we want to mandate that all template files be stored in UTF8? I don't think so, because most people don't have Unicode editors, or don't want to be bothered with an extra step to convert files to UTF8 and back when editing them in their favorite editor.

    Same question for script and template files, how do we know what language and character set they are authored in? Should we overload the filename suffix (e.g., '.shiftjis.adp', '.ja_JP.euc.adp')?

    The simplest design is probably just to assign a default mapping from each locale to character a set: e.g. ja_JP -> ShiftJIS, fr_FR -> ISO-8859-1. +++ (see new ACS/Java notes) +++

    Tcl Source File Character Set

    There are two classes of Tcl files loaded by the system; library files loaded at server startup, and page script files, which are run on each page request.


    Should we require all Tcl files be stored as UTF8? That seems too much of a burden on developers.

    50.10 Tcl library files can be authored in any character set. The system must have a way to determine the character set before loading the files, probably from the filename.

    50.20 Tcl page script files can be authored in any character set. The system must have a way to determine the character set before loading the files, probably from the filename.

    Submitted Form Data Character Set

    50.30 Data which is submitted with a HTTP request using a GET or POST method may be in any character set. The system must be able to determine the encoding of the form data and convert it to Unicode on demand.

    50.35 The developer must be able to override the default system choice of character set when parsing and validating user form data.

    50.30.10 Extra hair: In Japan and some other Asian languages where there are multiple character set encodings in common use, the server may need to attempt to do an auto-detection of the character set, because buggy browsers may submit form data in an unexpected alternate encoding.

    Output Character Set

    50.40 The output character set for a page request will be determined by default by the locale associated with the request (see requirement 20.0).

    50.50 It must be possible for a developer to manually override the output character set encoding for a request using an API function.

    VI.F ACS Kernel Issues

    60.10 All ACS error messages must use the message catalog and the request locale to generate error message for the appropriate locale.

    60.20 Web server error messages such as 404, 500, etc must also be delivered in the appropriate locale.

    60.30 Where files are written or read from disk, their filenames must use a character set and character values which are safe for the underlying operating system.

    VI.G Templates

    70.0 For a given abstract URL, the designer may create multiple locale-specific template files may be created (one per locale or language)

    70.10 For a given page request, the system must be able to select an approprate locale-specific template file to use. The request locale is computed as per (see requirement 20.0).

    Design note: this would probably be implemented by suffixing the locale or a locale abbreviation to the template filename, such as foo.ja.adp or foo.en_GB.adp.

    70.20A template file may be created for a partial locale (language only, without a territory), and the request processor should be able to find the closest match for the current request locale.

    70.30 A template file may be created in any character set. The system must have a way to know which character set a template file contains, so it can properly process it.

    Formatting Datasource Output in Templates

    70.50 The properties of a datasource column may include a datatype so that the templating system can format the output for the current locale. The datatype is defined by a standard ACS datatype plus a format token or format string, for example: a date column might be specified as 'current_date:date LONG,' or 'current_date:date "YYYY-Mon-DD"'

    Forms

    70.60 The forms API must support construction of locale-specific HTML form widgets, such as date entry widgets, and form validation of user input data for locale-specific data, such as dates or numbers.

    70.70 For forms which allow users to upload files, a standard method for a user to indicate the charset of a text file being uploaded must be provided.

    Design note: this presumably applies to uploading data to the content repository as well

    VI.H Sorting and Searching

    80.10 Support API for correct collation (sorting order) on lists of strings in locale-dependent way.

    80.20 For the Tcl API, we will say that locale-dependent sorting will use Oracle SQL operations (i.e., we won't provide a Tcl API for this). We require a Tcl API function to return the correct incantation of NLS_SORT to use for a given locale with ORDER BY clauses in queries.

    80.40 The system must handle full-text search in any supported language.

    VI.G Time Zones

    90.10 Provide API support for specifying a time zone

    90.20 Provide an API for computing time and date operations which are aware of timezones. So for example a calendar module can properly synchronize items inserted into a calendar from users in different time zones using their own local times.

    90.30 Store all dates and times in universal time zone, UTC.

    90.40 For a registered users, a time zone preference should be stored.

    90.50 For a non-registered user a time zone preference should be attached via a session or else UTC should be used to display every date and time.

    90.60 The default if we can't determine a time zone is to display all dates and times in some universal time zone such as GMT.

    VI.H Database

    100.10 Since UTF8 strings can use up to three (UCS2) or six (UCS4) bytes per character, make sure that column size declarations in the schema are large enough to accomodate required data (such as email addresses in Japanese).

    VI.I Email and Messaging

    When sending an email message, just as when delivering the content in web page over an HTTP connection, it is necessary to be able to specify what character set encoding to use.

    110.10 The email message sending API will allow for a character set encoding to be specified.

    110.20 The email accepting API will allow for character set to be parsed correctly (hopefully a well formatted message will have a MIME character set content type header)

    Implementation Notes

    Because globalization touches many different parts of the system, we want to reduce the implementation risk by breaking the implementation into phases.

    VII. Revision History

    Document Revision # Action Taken, Notes When? By Whom?
    0.1 Creation 11/08/2000 Henry Minsky
    0.2 Minor typos fixed, clarifications to wording 11/14/2000 Henry Minsky
    0.3 comments from Christian 1/14/2000 Henry Minsky

    hqm@arsdigita.com

    Last modified: $Date: 2001/04/20 20:51:09 $