%myvars; ]> By Peter Marklund and Lars Pind Stage Task Who Internationalization Package Developer uses the acs-lang tools to replace all visible text in a package with message keys. (More information) Package Developer Release Management The newly internationalized package is released. Package Developer The translation server is updated with the new package. Translation server maintainers Localization Translators work in their respective locales to write text for each message key. (More information) Translators Release Management The translated text in the database of the translation server is compared to the current translations in the OpenACS code base, conflicts are resolved, and the new text is written to catalog files on the translation server. Translation server maintainers The catalog files are committed to the OpenACS code base. Translation server maintainers A new version of OpenACS core and/or affected packages is released and published in the OpenACS.org repository. Release Manager Upgrading Site Administrators upgrade their OpenACS sites, either via the automatic upgrade from the Repository or via tarball or CVS Site Administrators Site Administrators import the new translations. Existing local translations, if they exist, are not overwritten. Site Administrators This document describes how to develop internationalized OpenACS packages, including writing new packages with internationalization and converting old packages. Text that users might see is "localizable text"; replacing monolingual text and single-locale date/time/money functions with generic functions is "internationalization"; translating first generation text into a specific language is "localization." At a minimum, all packages should be internationalized. If you do not also localize your package for different locales, volunteers may use a public "localization server" to submit suggested text. Otherwise, your package will not be usable for all locales. The main difference between monolingual and internationalized packages is that all user-visible text in the code of an internationalized package are coded as "message keys." The message keys correspond to a message catalog, which contains versions of the text for each available language. Script files (.adp and .tcl and .vuh), database files (.sql), and APM parameters are affected. Other differences include: all dates read or written to the database must use internationalized functions. All displayed dates must use internationalized functions. All displayed numbers must use internationalized functions. Localizable text must be handled in ADP files, in Tcl files, and in APM Parameters. OpenACS provides two approaches, message keys and localized ADP files. For ADP pages which are mostly code, replacing the message text with message key placeholders is simpler. This approach also allows new translation in the database, without affecting the filesystem. For ADP pages which are static and mostly text, it may be easier to create a new ADP page for each language. In this case, the pages are distinguished by a file naming convention. OpenACS does not have a general system for supporting multiple, localized versions of user-input content. This document currently refers only to internationalizing the text in the package user interface. If the request processor finds a file named filename.locale.adp, where locale matches the user's locale, it will process that file instead of filename.adp. For example, for a user with locale tl_PH, the file index.tl_PH.adp, if found, will be used instead of index.adp. The locale-specific file should thus contain text in the language appropriate for that locale. The code in the page, however, should still be in English. Message keys are processed normally. Internationalizing templates is about replacing human readable text in a certain language with internal message keys, which can then be dynamically replaced with real human language in the desired locale. Message keys themselves should be in ASCII English, as should all code. Three different syntaxes are possible for message keys. "Short" syntax is the recommended syntax and should be used for new development. When internationalizing an existing package, you can use the "temporary" syntax, which the APM can use to auto-generate missing keys and automatically translate to the short syntax. The "verbose" syntax is useful while developing, because it allows default text so that the page is usable before you have done localization. The short: #package_key.message_key# The advantage of the short syntax is that it's short. It's as simple as inserting the value of a variable. Example: #forum.title# The verbose: <trn key="package_key.message_key" locale="locale">default text</trn> The verbose syntax allows you to specify a default text in a certain language. This syntax is not recommended anymore, but it can be convenient for development, because it still works even if you haven't created the message in the message catalog yet, because what it'll do is create the message key with the default text from the tag as the localized message. Example: <trn key="forum.title" locale="en_US">Title</trn> The temporary: <#message_key original text#> This syntax has been designed to make it easy to internationalize existing pages. This is not a syntax that stays in the page. As you'll see later, it'll be replaced with the short syntax by a special feature of the APM. You may leave out the message_key by writing an underscore (_) character instead, in which case a message key will be auto-generated by the APM. Example: <_ Title> We recommend the short notation for new package development. In adp files message lookups are typically done with the syntax \#package_key.message_key\#. In Tcl files all message lookups *must* be on either of the following formats: Typical static key lookup: [_ package_key.message_key] - The message key and package key used here must be string literals, they can't result from variable evaluation. Static key lookup with nondefault locale: [lang::message::lookup $locale package_key.message_key] - The message key and package key used here must be string literals, they can't result from variable evaluation. Dynamic key lookup: [lang::util::localize $var_with_embedded_message_keys] - In this case the message keys in the variable var_with_embedded_message_keys must appear as string literals \#package_key.message_key\# somewhere in the code. Here is an example of a dynamic lookup: set message_key_array { dynamic_key_1 \#package_key.message_key1\# dynamic_key_2 \#package_key.message_key2\# } set my_text [lang::util::localize $message_key_array([get_dynamic_key])] Translatable texts in page Tcl scripts are often found in page titles, context bars, and form labels and options. Many times the texts are enclosed in double quotes. The following is an example of grep commands that can be used on Linux to highlight translatable text in Tcl files: # Find text in double quotes find -iname '*.tcl'|xargs egrep -i '"[a-z]' # Find untranslated text in form labels, options and values find -iname '*.tcl'|xargs egrep -i '\-(options|label|value)'|egrep -v '<#'|egrep -v '\-(value|label|options)[[:space:]]+\$[a-zA-Z_]+[[:space:]]*\\?[[:space:]]*$' # Find text in page titles and context bars find -iname '*.tcl'|xargs egrep -i 'set (title|page_title|context_bar) '|egrep -v '<#' # Find text in error messages find -iname '*.tcl'|xargs egrep -i '(ad_complain|ad_return_error)'|egrep -v '<#' You may mark up translatable text in Tcl library files and Tcl pages with temporary tags on the <#key text#> syntax. If you have a sentence or paragraph of text with variables and or procedure calls in it you should in most cases try to turn the whole text into one message in the catalog (remember that translators is made easier the longer the phrases to translate are). In those cases, follow these steps: For each message call in the text, decide on a variable name and replace the procedure call with a variable lookup on the syntax %var_name%. Remember to initialize a Tcl variable with the same name on some line above the text. If the text is in a Tcl file you must replace variable lookups (occurrences of $var_name or ${var_name}) with %var_name% You are now ready to follow the normal procedure and mark up the text using a temporary message tag (<#_ text_with_percentage_vars#>) and run the action replace tags with keys in the APM. The variable values in the message are usually fetched with upvar, here is an example from dotlrn: ad_return_complaint 1 "Error: A [parameter::get -parameter classes_pretty_name] must have <em>no</em>[parameter::get -parameter class_instances_pretty_plural] to be deleted" was replaced by: set subject [parameter::get -localize -parameter classes_pretty_name] set class_instances [parameter::get -localize -parameter class_instances_pretty_plural] ad_return_complaint 1 [_ dotlrn.class_may_not_be_deleted] This kind of interpolation also works in adp files where adp variable values will be inserted into the message. Alternatively, you may pass in an array list of the variable values to be interpolated into the message so that our example becomes: set msg_subst_list [list subject [parameter::get -localize -parameter classes_pretty_name] class_instances [parameter::get -localize -parameter class_instances_pretty_plural]] ad_return_complaint 1 [_ dotlrn.class_may_not_be_deleted $msg_subst_list] When we were done going through the Tcl files we ran the following commands to check for mistakes: # Message tags should usually not be in curly braces since then the message lookup may not be # executed then (you can usually replace curly braces with the list command). Find message tags # in curly braces (should return nothing, or possibly a few lines for inspection) find -iname '*.tcl'|xargs egrep -i '\{.*<#' # Check if you've forgotten space between default key and text in message tags (should return nothing) find -iname '*.tcl'|xargs egrep -i '<#_[^ ]' # Review the list of Tcl files with no message lookups for tcl_file in $(find -iname '*.tcl'); do egrep -L '(<#|\[_)' $tcl_file; done When you feel ready you may visit your package in the package manager and run the action "Replace tags with keys and insert into catalog" on the Tcl files that you've edited to replace the temporary tags with calls to the message lookup procedure. Most date, time, and number variables are calculated in Tcl files. Dates and times must be converted when stored in the database, when retrieved from the database, and when displayed. All dates are stored in the database in the server's timezone, which is an APM Parameter set at /acs-lang/admin/set-system-timezone and readable at lang::system::timezone.. When retrieved from the database and displayed, dates and times must be localized to the user's locale. Some parameters contain text that need to be localized. In this case, instead of storing the real text in the parameter, you should use message keys using the short notation above, i.e. #package_key.message_key#. In order to avoid clashes with other uses of the hash character, you need to tell the APM that the parameter value needs to be localized when retrieving it. You do that by saying: parameter::get -localize. Here are a couple of examples. Say we have the following two parameters, taken directly from the dotlrn package. Parameter Name Parameter Value class_instance_pages_csv #dotlrn.class_page_home_title#,Simple 2-Column;#dotlrn.class_page_calendar_title#,Simple 1-Column;#dotlrn.class_page_file_storage_title#,Simple 1-Column departments_pretty_name #departments_pretty_name# Then, depending on how we retrieve the value, here's what we get: Command used to retrieve Value Retrieved Value parameter::get -localize -parameter class_instances_pages_csv Kurs Startseite,Simple 2-Column;Kalender,Simple 1-Column;Dateien,Simple 1-Column parameter::get -localize -parameter departments_pretty_name Abteilung parameter::get -parameter departments_pretty_name #departments_pretty_name# The value in the rightmost column in the table above is the value returned by an invocation of parameter::get. Note that for localization to happen you must use the -localize flag. The locale used for the message lookup will be the locale of the current request, i.e. lang::conn::locale or ad_conn locale. Developers are responsible for creating the keys in the message catalog, which is available at /acs-lang/admin/ For multilingual websites we recommend using the UTF8 charset. In order for AOLserver to use utf8 you need to set the config parameters OutputCharset and URLCharset to utf-8 in your AOLserver config file (use the etc/config.tcl template file). This is the default for OpenACS 5.1 and later. For sites running on Oracle you need to make sure that AOLserver is running with the NLS_LANG environment variable set to .UTF8. You should set this variable in the nsd-oracle run script (use the acs-core-docs/www/files/nds-oracle.txt template file). From/acs-admin/apm/, select a package and then click on Internationalization, then Convert ADP, Tcl, and SQL files to using the message catalog.. This pass only changes the adp files; it does not affect catalog files or the catalog in the database. You will now be walked through all of the selected adp pages. The UI shows you the intended changes and lets you edit or cancel them key by key. From the same Convert ADP ... page in /acs-admin/apm as in the last step, repeat the process but deselect Find human language text ... and select Replace <# ... #> tags ... and click OK. This step replaces all of the temporary tags with "short" message lookups, inserts the message keys into the database message catalog, and then writes that catalog out to an XML file. Examine all of the Tcl files in the packages for human-readable text and replace it with temporary tags. The temporary tags in Tcl are slightly different from those in ADP. If the first character in the temporary tag is an underscore (_), then the message keys will be auto-generated from the original message text. Here is an unmodified Tcl file: set title "Messages for $a(name) in $b(label)" set context [list [list . "SimPlay"] \ [list [export_vars -base case-admin { case_id }] \ "Administer $a(name)"] \ "Messages for $a(name)"] ... and here is the same file after temporary message tags have been manually added: set title <#admin_title Messages for %a.name% in %b.label%#> set context [list [list . <#_ SimPlay#>] \ [list [export_vars -base case-admin { case_id }] \ <#_ Administer %a.name%#>] \ <#_ Messages for %a.name%#>] Note that the message key case_admin_page_title was manually selected, because an autogenerated key for this text, with its substitute variables, would have been very confusing Repeat step 2 for Tcl files. Here is the example Tcl file after conversion: set title [_ simulation.admin_title] set context [list [list . [_ simulation.SimPlay]] \ [list [export_vars -base case-admin { case_id }] \ [_ simulation.lt_Administer_name_gt]] \ [_ simulation.lt_Messages_for_role_pre]] If there is any user-visible Tcl code in the .sql or .xql files, internationalize that the same way as for the Tcl files. See Find datetime in .xql files. Use command line tools to find suspect SQL code: grep -r "to_char.*H" * grep -r "to_date.*H" * In SQL statements, replace the format string with the ANSI standard format, YYYY-MM-DD HH24:MI:SS and change the field name to *_ansi so that it cannot be confused with previous, improperly formatting fields. For example, to_char(timestamp,'MM/DD/YYYY HH:MI:SS') as foo_date_pretty becomes to_char(timestamp,'YYYY-MM-DD HH24:MI:SS') as foo_date_ansi In Tcl files where the date fields are used, convert the datetime from local server timezone, which is how it's stored in the database, to the user's timezone for display. Do this with the localizing function lc_time_system_to_conn: set foo_date_ansi [lc_time_system_to_conn $foo_date_ansi] When a datetime will be written to the database, first convert it from the user's local time to the server's timezone with lc_time_conn_to_system. When a datetime field will be displayed, format it using the localizing function lc_time_fmt. lc_time_fmt takes two parameters, datetime and format code. Several format codes are usable for localization; they are placeholders that format dates with the appropriate codes for the user's locale. These codes are: %x, %X, %q, %Q, and %c. set foo_date_pretty [lc_time_fmt $foo_date_ansi "%x %X"] Use the _pretty version in your ADP page. %c: Long date and time (Mon November 18, 2002 12:00 AM) %x: Short date (11/18/02) %X: Time (12:00 AM) %q: Long date without weekday (November 18, 2002) %Q: Long date with weekday (Monday November 18, 2002) The "q" format strings are OpenACS additions; the rest follow unix standards (see man strftime). To internationalize numbers, use lc_numeric $value, which formats the number using the appropriate decimal point and thousand separator for the locale. When coding forms, remember to use message keys for each piece of text that is user-visible, including form option labels and button labels. This section describes how to check that the set of keys used in message lookups in tcl, adp, and info files and the set of keys in the catalog file are identical. The scripts below assume that message lookups in adp and info files are on the format \#package_key.message_key\#, and that message lookups in Tcl files are always is done with one of the valid lookups described above. The script further assumes that you have perl installed and in your path. Run the script like this: acs-lang/bin/check-catalog.sh package_key where package_key is the key of the package that you want to test. If you don't provide the package_key argument then all packages with catalog files will be checked. The script will run its checks primarily on en_US XML catalog files. When writing in one language, it is possible to create clever code to make correct text. In English, for example, you can put an if command at the end of a word which adds "s" if a count is anything but 1. This pluralizes nouns correctly based on the data. However, it is confusing to read and, when internationalized, may result in message keys that are both confusing and impossible to set correctly in some languages. While internationalizing, watch out that the automate converter does not create such keys. Also, refactor compound text as you encounter it. The automated system can easily get confused by tags within message texts, so that it tries to create two or three message keys for one long string with a tag in the middle. In these cases, uncheck those keys during the conversion and then edit the files directly. For example, this code: <p class="form-help-text"><b>Invitations</b> are sent, when this wizard is completed and casting begins.</p> has a bold tag which confuses the converter into thinking there are two message keys for the text beginning "Invitations ..." where there should be one: Instead, we cancel those keys, edit the file manually, and put in a single temporary message tag: <p class="form-help-text"> <#Invitations_are_sent <b>Invitations</b> are sent, when this wizard is completed and casting begins.#> </p> Complex if statements may produce convoluted message keys that are very hard to localize. Rewrite these if statements. For example: Select which case <if @simulation.casting_type@ eq "open">and role</if> to join, or create a new case for yourself. If you do not select a case <if @simulation.casting_type@ eq "open">and role</if> to join, you will be automatically assigned to a case <if @simulation.casting_type@ eq "open">and role</if> when the simulation begins. ... can be rewritten: <if @simulation.casting_type@ eq "open"> Select which case and role to join, or create a new case for yourself. If you do not select a case and role to join, you will be automatically assigned to a case and role when the simulation begins. </if> <else> Select which case to join, or create a new case for yourself. If you do not select a case to join, you will be automatically assigned to a case when the simulation begins. </else> Another example, where bugs are concatenated with a number: <if @components.view_bugs_url@ not nil> <a href="@components.view_bugs_url@" title="View the @pretty_names.bugs@ for this component"> </if> @components.num_bugs@ <if @components.num_bugs@ eq 1> @pretty_names.bug@ </if> <else> @pretty_names.bugs@ </else> <if @components.view_bugs_url@ not nil> </a> </if> <if @components.view_bugs_url@ not nil> <a href="@components.view_bugs_url@" title="#­bug-tracker.View_the_bug_fo_component#"> </if> @components.num_bugs@ <if @components.num_bugs@ eq 1> @pretty_names.bug@ </if> <else> @pretty_names.bugs@ </else> <if @components.view_bugs_url@ not nil> </a> </if> It would probably be better to do this as something like: <if @components.view_bugs_url@ not nil> <if @components.num_bugs@ eq 1> <a href="@components.view_bugs_url@" title="#­bug-tracker.View_the_bug_fo_component#">#­bug-tracker.one_bug#</a> </if><else> <a href="@components.view_bugs_url@" title="#­bug-tracker.View_the_bug_fo_component#">#­bug-tracker.N_bugs#</a> </else> </if> Converting a phrase from one language to another is usually more complicated than simply replacing each word with an equivalent. When several keys are concatenated, the resulting word order will not be correct for every language. Different languages may use expressions or idioms that don't match the phrase key-for-key. Create complete, distinct keys instead of building text from several keys. For example: Original code: multirow append links "New [bug_tracker::conn Bug]" Problematic conversion: multirow append links "[_ bug-tracker.New] [bug_tracker::conn Bug]" Better conversion: set bug_label [bug_tracker::conn Bug] multirow append links "[_ bug-tracker.New_Bug]" "${url_prefix}bug-add" ... and include the variable in the key: "New %bug_label%". This gives translators more control over the phrase. In this example of bad i18n, full name is created by concatenating first and last name (admittedly this is pervasive in the toolkit): <a href="@past_version.maintainer_url@" title="#­bug-tracker.Email# @past_version.maintainer_email@"> @past_version.maintainer_first_names@ @past_version.maintainer_last_name@</a> When phrases are exactly the same in several places, use a single key. For common words such as Yes and No, you can use a library of keys at acs-kernel. For example, instead of using myfirstpackage.Yes, you can use acs-kernel.Yes. You can also use the Message Key Search facility to find duplicates. Be careful, however, building up sentences from keys because grammar and other elements may not be consistent across different locales. Additional discussion: Re: Bug 961 ("Control Panel" displayed instead of "Administer"), Translation server upgraded, and Localization questions. Many packages use code words or key words, such as "open" and "closed", which will never be shown to the user. They may match key values in the database, or be used in a switch or if statement. Don't change these. For example, the original code is workflow::case::add_log_data \ -entry_id $entry_id \ -key "resolution" \ -value [db_string select_resolution_code {}] This is incorrectly internationalized to workflow::case::add_log_data \ -entry_id $entry_id \ -key "[_ bug-tracker.resolution]" \ -value [db_string select_resolution_code {}] But resolution is a keyword in a table and in the code, so this breaks the code. It should not have been internationalized at all. Here's another example of text that should not have been internationalized: {show_patch_status "open"} It is broken if changed to {show_patch_status "[_ bug-tracker.open]"} The automatic converter may create unique but crytic message keys. Watch out for these and replace them with more descriptive keys. For example: <msg key="You">You can filter by this %component_name% by viisting %filter_url_string%</msg> <msg key="You_1">You do not have permission to map this patch to a bug. Only the submitter of the patch and users with write permission on this Bug Tracker project (package instance) may do so.</msg> <msg key="You_2">You do not have permission to edit this patch. Only the submitter of the patch and users with write permission on the Bug Tracker project (package instance) may do so.</msg> These would be more useful if they were, "you_can_filter", "you_do_not_have_permission_to_map_this_patch", and "you_do_not_have_permission_to_edit_this_patch". Don't worry about exactly matching the english text, because that might change; instead try to capture the meaning of the phrase. Ask yourself, if I was a translator and didn't know how this application worked, would this key and text make translation easy for me? Sometimes the automatic converter creates keys that don't semantically match their text. Fix these: <msg key="Fix">for version</msg> <msg key="Fix_1">for</msg> <msg key="Fix_2">for Bugs</msg> Another example: Bug-tracker component maintainer was converted to [_ bug-tracker.Bug-tracker]. Instead, it should be bug_tracker_component_maintainer. Translations may need to differ depending on the context in which the message appears. Different languages create plurals differently. Try to avoid keys which will change based on the value of a number. OpenACS does not currently support internationalization of plurals. If you use two different keys, a plural and a singular form, your application will not localize properly for locales which use different rules or have more than two forms of plurals. Watch out for quoting and escaping when editing text that is also code. For example, the original string set title "Patch \"$patch_summary\" is nice." breaks if the message text retains all of the escaping that was in the Tcl command: <msg>Patch \"$patch_summary\" is nice.</msg> When it becomes a key, it should be: <msg>Patch "$patch_summary" is nice.</msg> Also, some keys had %var;noquote%, which is not needed since those variables are not quoted (and in fact the variable won't even be recognized so you get the literal %var;noquote% in the output). Code within curly brackets isn't evaluated. Tcl uses curly brackets as an alternative way to build lists. But Tcl also uses curly brackets as an alternative to quotation marks for quoting text. So this original code array set names { key "Pretty" ...} ... if converted to array set names { key "[_bug-tracker.Pretty]" ...} ... won't work since the _ func will not be called. Instead, it should be array set names [list key [_bug-tracker.Pretty] ...] User locale is a property of ad_conn, ad_conn locale. The request processor sets this by calling lang::conn::locale, which looks for the following in order of precedence: Use user preference for this package (stored in ad_locale_user_prefs) Use system preference for the package (stored in apm_packages) Use user's general preference (stored in user_preferences) Use Browser header (Accept-Language HTTP header) Use system locale (an APM parameter for acs_lang) default to en_US For ADP pages, message key lookup occurs in the templating engine. For Tcl pages, message key lookup happens with the _ function. In both cases, if the requested locale is not found but a locale which is the default for the language which matches your locale's language is found, then that locale is offered instead. Most translators use the OpenACS Public Translation Server, because the process of getting new message keys onto the server and getting new translations back into the distribution are handled by the maintainers of that machine. You can also do translation work on your own OpenACS site; this makes your own translations more readily available to you but also means that your work will not be shared with other users unless you take extra steps (contacting an OpenACS core developer or submitting a patch) to get your work back to the OpenACS core. The basic steps for translators: Go to the Localization page and choose the locale that you are translating to. If the locale is not present you need to visit Administration of Localization and create the locale. To translate messages in the pages they appear, Toggle Translator Mode and then browse to the page you want to translate. Untranslated messages will have a yellow background and a red star that you click to translate the message. Translated messages have a green star next to them that is a hyperlink to editing your translation. There is a history mechanism that allows you to see previous translations in case you would want to revert a translation. While in Translator mode, a list of all message keys appears at the bottom of each page. To translate many messages at once, go to Administration of Localization, click on the locale to translate, then click on a package, and then click Batch edit these messages. When creating a new locale based on an existing one, such as creating the Guatemalan version of Spanish, you can copy the existing locale's catalog files using the script /packages/acs-core-docs/www/files/create-new-catalog.sh.