Globalization HOWTO
Release: Pegasus 2.3
Author: Chuck Carmack (carmack@us.ibm.com)
December 1, 2003
Change History:
01/12/03 |
carmack |
Section 2.2.2. Changed how
the package name parameter should be used. It should no longer be
used as part of the table name inside the bundle. |
08/04/06 |
Marek Szermutzky |
Section 2.2.5. Added information how to write platform specific messages. |
01/23/07 |
Sushma Fernandes |
Section 2.2.5. Added information on special considerations while creating a new message. |
As part of the Pegasus 2.3 release, functions were added for
globalization support. Globalization involves two major
aspects: internationalization and localization.
Internationalization is the process of writing a program that is
locale-neutral. In other words, the program should be able to run
in any locale without change. There are several categories in a
locale, including the language of message strings, date format, time
format, etc. For release 2.3, the Pegasus server is concerned with
the language of the message strings it returns to its clients.
To support internationalization, a program is designed to do the
following:
Support character sets that can represent customer data in any language. Typically, the program supports some variation of Unicode for internal data. There is usually some conversion between the supported character sets for external data, and the internal character set. Since Unicode covers all characters, and usually has converters on the platform, it is a good choice for the 'normalized' internal character set. The most 'interoperable' solution for external data is to support UTF-8 (eg. network and file system data). The internal data is usually UTF-16 (or UCS-2, but that is deprecated).
Extract locale-sensitive resources, such as message strings, from the code to external resource files. Typically, the resources are loaded based on the locale requested by the end-user, and returned to the end-user for display.
Localization is the process of customizing a software product to
support particular locales. For example, a product that is
internationalized might want to only localize for certain
countries. This would mean that the localized resources (eg.
message files) would only be translated and shipped for the countries
that the product supports. Since the code for the product is
locale-neutral, it will be easy to drop in new translations as more
countries are supported.
The Pegasus 2.3 release added support for globalization. At a
high-level, the following additions were made to Pegasus 2.3:
Please refer to PEPs 56 and 58 for details about the globalization
design in Pegasus 2.3.
This document provides a HOWTO guide to be used by developers to
globalize code that is being added to Pegasus. The audience for
this document are:
The quickest way to approach this document is to read the General
section, and then the developer section that relates to what you are
doing.
Pegasus 2.3 supports Unicode throughout the processing of
requests. External data to Pegasus is encoded in UTF-8.
Internal data is encoded in UTF-16.
UTF-8 support for external data includes the CIM-XML messages passed
over the network, and the repository files. Note: UTF-8
support was NOT added to the MOF Compiler for MOF files in release
2.3. For the CIM-XML messages, Pegasus follows section 4.8 of
the CIM-HTTP
specification Specifically, Pegasus supports the
"utf-8" setting for the charset parameter of the Content-Type header and
the XML encoding attribute. If no charset is specified, the 7-bit
ASCII is assumed.
The internal support of UTF-16 is encapsulated in the Pegasus String class. This class has been updated to contain UTF-16 characters. Specifically, the Char16 objects inside the String contain UTF-16 characters. Note: a UTF-16 surrogate pair is contained in two consecutive Char16 objects. To keep backwards compatibilty, the methods on the String class have not changed. New methods have been added as needed. The following describes this in more detail:
Pegasus 2.3 supports clients and providers that wish to localize.
There are two areas to be localized: ERROR
elements in the CIM-XML; and Object
Definition elements in the CIM-XML. Clients can
request the server to return error messages and CIM objects in a
set of languages of their choosing. Clients can also tag a
language to the CIM objects they are sending to the server.
Providers and the server can return error messages and CIM objects that
are tagged with one of languages requested by the client.
The localization design is based on section 4.8 of the CIM-HTTP
specification , which refers to RFC 2616.
The method used to tag a language to the CIM-XML is through the
Accept-Language and Content-Language HTTP headers. These headers
are basically lists of language tags. An HTTP request can contain
an Accept-Language header, which indicates the list of preferred
languages that the client wants in the response. This list can be
prioritized by using the quality numbers. An HTTP request or
response can contain a Content-Language header, which indicates the
language(s) of the content in the message. In the Pegasus case,
this would be the CIM-XML. Note that the Content-Language header
is a list of language tags. This allows the content of an HTTP
message to contain more than one translation. However, in the
Pegasus case, there is only one CIM-XML document in the HTTP message,
and thus one translation.
CIM clients may use the Accept-Language HTTP header to specify the
languages they wish to be returned in the CIM response message.
CIM clients may also use the Content-Language header to tag the language
of any CIM objects they are sending to the server in the CIM request
message. The server, and providers, should attempt to return
error messages and CIM objects in one of the accept languages requested
by the client. The server and providers should set the
Content-Language header in the CIM response message to indicate which of
the requested languages they are returning.
NOTE: Localization support was not added for the MOF files and
repository in Pegasus 2.3. The #pragma locale, #pragma
instancelocale, and translatable qualifier flavor are not supported in
the Pegasus 2.3 MOF compiler. From the client perspective,
classes, qualifiers, and instances stored in the repository are not
tagged with a language. The Accept-Language and Content-Language
headers will be ignored for repository operations. However, since
the repository will support UTF-8, characters for any language may
be stored there.
NOTE: Since the Content-Language header applies to the entire
HTTP message, it applies to the entire CIM-XML document. This
includes all the objects in the document, including enumerated objects,
and all the values in the objects. This is a limitation that will
remain until the CIM standard has been updated to support language tags
tied to individual CIM values. From the client perspective, it is
possible for Pegasus to send a CIM response with NO Content-Language,
even if the client had sent Accept-Language. This can happen
if Pegasus does not know the language of the response. An example
is a request that was sent to a Pegasus 2.2 provider. Another
example is an enumerated response where each provider returned a
different language. Please refer to PEP58 for details on these
provider scenarios.
The Accept-Language and Content-Language headers are encapsulated
in AcceptLanguageList and ContentLanguageList classes, respectively.
These classes contain LanguageTag objects. The AcceptLanguageList class
keeps its LanguageTags prioritized based on quality,
according to RFC 2616.
AcceptLanguageList and ContentLanguageList are the objects used by code
throughout the request/response processing, from the client to the
server to the providers and back. The server handles the creation
of these objects from the HTTP headers. Code at each point in the
process will have access to these objects.
Please refer to the following files for details on the Pegasus
language interfaces.
See the sections below for details on how to write clients and
providers to use these classes.
One of the goals of globalization for Pegasus 2.3 is the extraction of
hardcoded messages into translated message files, loading
translated messages from those files, and returning those messages to
the client. The topics to be discussed here are: how to
create message files, how to compile message files, and how to load
messages into Pegasus.
At the time of writing, the message loading function in Pegasus 2.3
used the International Components for Unicode (ICU) libraries. This
is expected to be the future direction for Pegasus. ICU uses a resource bundle
format for their message files. In order to load the
messages, ICU requires that the resource bundles are compiled into a
binary form (.res file) using their genrb tool.
Platform Maintainers Note: Please refer to PEP 58 for
information about how to build Pegasus to use the ICU libraries.
The documentation for ICU resource bundles is in the Resource
Management section of the ICU User Guide
. This section will tell you how to create and organize your
resource bundles for different languages. Note: your
resource bundles should be organized in a tree structure similiar to
the one shown in the Resource Management section, including the empty
bundles in the tree.
It is recommended that you ship a root resource bundle to be used as
the fallback in case the client requests a language that you are not
supporting. The Pegasus make files are set up to automatically
create and compile a root resource bundle for you. For Pegasus
2.3, the make will use your "en" bundle, upper case all the messages,
and then put the uppercased messages into the root bundle. The
uppercasing of the messages is necessary to create a "fallback" root
bundle that contains invariant characters across all EBCDIC and
ASCII codepages.
NOTE: When creating your resource bundles, the name of the
table resource should not
contain the package name. For example, if you
have a bundle with a package name of "xyz", then the "en" bundle should
start like this:
en:table {
..... messages here
}
not like this:
xyz_en:table {
..... messages here
}
This is needed because the package name (-p) option is used by the
Pegasus make files on the call to genrb.
NOTE: Pegasus 2.3 only supports simple string resources in the
ICU resource bundles. String resources may only be loaded by
key. Tables, arrays, and other complex resource types, are not
supported.
In order to compile your resource bundles, support has been added to
the Pegasus make files to run genrb. A new make target,
"messages", has been added that will call genrb and put the compiled
bundles (.res) in a directory of your choosing. An example of ICU
resource bundles and the make files to compile them are located in:
NOTE: At the time of writing, only the Linux make files have been
updated to compile ICU resource bundles.
It is important to place the compiled resource bundles in a
directory where your code can find them . The make files above
compile the resource bundles into
$PEGASUS_HOME/msg/provider/localizedProvider. The code that loads
these messages uses the MessageLoader class (next section) to load
messages from this directory.
Code that needs to load a message in Pegasus does not call ICU
directly. Two message loading classes were added for Pegasus
2.3: MessageLoader and MessageLoaderParms. These classes are
abstractions designed to hide of the actual loader used (but note that
at the time of writing, only ICU is supported). The
MessageLoader is used to load a message using a list of preferrred
languages. The parameters to MessageLoader are encapsulated in a
MessageLoaderParms object.
The MessageLoader is the place where the Accept-Language header,
Content-Language header, and the ICU resource bundles, join up.
The MessageLoader class is designed to receive an AcceptLanguageList
object, and a set of parameters indicating the bundle base-name and
message ID to use. The AcceptLanguageList object contains the list of
requested languages sent by the client. The MessageLoader
searches for the message in the set of bundles named with the base-name,
using the AcceptLanguageList for the list of specific translated bundles
to search. The MessageLoader returns the message that it found,
along with a ContentLanguageList object indicating the language of the
message. The ContentLanguageList object should be used to indicate
the language of the response sent back to the client.
The MessageLoaderParms object contains the parameters to load the
message. There are many parameters, but many can be allowed to
default. Here is a description of the parameters:
String msg_id; | Input. Required. |
Message ID of the message to load from the resource bundle. This is the key that ICU will use to load the message. |
String default_msg; | Input. Required |
Message to return if the no message can be loaded for msg_id
from any resource bundle. Note: The args parameters below
are substituted into this string. Note: For the args into this string, use the Pegasus '$' form, as described in pegasus/src/Pegasus/Common/Formatter.h. Don't use the ICU substitution format for the default message string. |
String msg_src_path; | Input. Optional Default: $PEGASUS_HOME/msg/pegasus/pegasusServer |
Path to the resource bundle file which contains the
msg_id. Note: Only specify the path down to the bundle base-name. Do not append a language tag, such as "_root" or "_en". Do not append a file extension. Note: relative paths start at $PEGASUS_HOME/msg. Note: defaults to the bundle containing the Pegasus server messages. |
AcceptLanguageList acceptlanguages; | Input. Optional Default: AcceptLanguageList() |
Contains the list of preferred languages, in priority order. This is combined with msg_src_path to determine which resource bundles to search for for the msg_id. If not empty, overrides useThreadLocale and useProcessLocale. |
ContentLanguageList contentlanguages; | Output | Contains the language that MessageLoader found for the msg_id. |
Boolean useProcessLocale; | Input Optional Default = false |
If true, MessageLoader will use the default locale of the process. If true, overrides useThreadLocale. |
Boolean useThreadLocale; | Input Optional Default = true |
If true, MessageLoader will use the AcceptLanguageList set by Pegasus into the caller's Thread. See the Note below for details. |
Boolean useICUfallback | Input Optional Default = false |
If true, use ICU's fallback mechnism to search more general resource bundles if the msg_id cannot be found. Note: the recommended setting is false if you are using an AcceptLanguageList from a CIM client. The Accept-Languages HTTP header from the client contains the fallback specifications. Using ICU's fallback in this case may lead to returning a language that the client didn't ask for. |
Formatter::Arg arg0; Formatter::Arg arg1; Formatter::Arg arg2; Formatter::Arg arg3; Formatter::Arg arg4; Formatter::Arg arg5; Formatter::Arg arg6; Formatter::Arg arg7; Formatter::Arg arg8; Formatter::Arg arg9; |
Input Optional Default: Formatter::Arg( ) // empty arg |
These are the substitution variables, using the Pegasus Formatter::Arg class. |
Notes:
The "useThreadLocale" parameter defaults to true. This flag
indicates to use the AcceptLanguageList object set by Pegasus into the
Pegasus Thread in which the caller's code is running. This
AcceptLanguageList object reflects the languages requested by the
client. This is useful for code that may not have access to the
AcceptLanguageList from the client. Pegasus sets this AcceptLanguageList
object into the Thread of providers and internal Pegasus code.
For this reason, it is recommended that provider and internal Pegasus
code use the "useThreadLocale" flag instead of explicity passing in an
AcceptLanguageList object. See the Provider Developer and Pegasus
Developer sections for details.
The "useProcessLocale" flag can be used to tell MessageLoader to use
the default locale of the process, as determined by ICU. This is
useful for situations where the caller is not localizing for a client
request. The caller may itself be a client (eg. cimconfig), or may
need to log messages to the system log in the locale of the Pegasus
server process. See the CLI Messages and Logger Messages sections
below.
"Master switch"
The MessageLoader class has a public static Boolean variable called
_useProcessLocale that may be used to override all the AcceptLanguageList
and useThreadLocale settings in the MessageLoaderParms objects passed
in. This is useful for CLI code (eg cimconfig) that needs to
localize its messages based on the locale of its process, which refects
the locale set by the user running the CLI (eg. $LANG on Unix).
The CLI code may call Pegasus APIs that are coded to use the Thread's
AcceptLanguageList, which will not be set in this case. The
_useProcessLocale static variable tells the MessageLoader to ignore the
AcceptLanguageList, useThreadLocale, and useProcessLocale settings in
MessageLoaderParms that it gets. The MessageLoader will use the
default process locale, as determined by ICU, in this case.
Important Note: The MessageLoader defaults to not use
the "fallback" mechanism described in the ICU Resource Management
section. This is because the Accept-Language header itself
describes the fallback that the client wants. However, the
MessageLoader does "fallback" to the root resource bundle if none of the
languages in AcceptLanguageList can be found. If the root resource
bundle cannot be found, then the default_msg is returned. The
"useICUFallback" flag can be set to have MessageLoader use ICU fallback
on all message load attempts. However, usage of this flag for
client requests may lead to incorrect results. For example, a
client sets Accept-Language to french, german, and spanish, in that
order, but there is no french resource bundle. A call to
MessageLoader with useICUfallback == true would cause the root resource
bundle string to be returned on the attempt to load from the french
bundle. But the client requested german to be the fallback after
french.
Please refer to the following files for details on the new Pegasus
classes.
The following example shows how a message may be loaded using the
classes described above. Note: this a generic example. Each
of the developer sections below have 'real-life' examples that are
better suited to each type of code.
// Build an AcceptLanguageList with some language elements
AcceptLanguageList acceptLangs;
acceptLangs.insert(LanguageTag("fr"), 0.5);
acceptLangs.insert(LanguageTag("de"), 0.8);
acceptLangs.insert(LanguageTag("es"), 0.4);
// Construct a MessageLoaderParms
MessageLoaderParms parms("msgID", "default message");
parms. msg_src_path = "/my_msg_dir/my_bundle";
parms.acceptlanguages = acceptLangs;
// Note: If you have args, set them into MessageLoaderParms
// Load the localized String
String localizedMsg = MessageLoader::getMessage(parms);
Here are some basic rules for writing messages:
When do I create a new message ?
A new message should be created if a message is needed with a content not described by any existing message.
A new message should be created if the number or placement of substitution parameters of an existing message would require an update.
It is not necessary to create a new message if just the text of the message is changed, while the meaning is kept. For instance if the event(error,warning,whatever) is described more precisely by the new message text, it is not necessary to create a new message, but the existing one should be updated.
Are there any special considerations while creating a new message ?
Example:
Server.CIMOperationRequestAuthorizer.NOT_IN_AUTHORIZED_GRP: string {"PGS05202: User '{0}' is not authorized to access CIM data."}
Processed message:
User {0} is not authorized to access CIM data.
Example:
Server.CIMOperationRequestAuthorizer.NOT_IN_AUTHORIZED_GRP: string {"PGS05202: User ''{0}'' is not authorized to access CIM data."}
Processed message:
User 'wbemuser' is not authorized to access CIM data.
Example:
ControlProviders.ProviderRegistrationProvider.ProviderRegistrationProvider. UNSUPPORTED_USERCONTEXT_VALUE:string {"PGS03029: Unsupported UserContext value: \"{0}\"."}
Processed message:
Unsupported UserContext value: "10".
How do I write a platform specific message ?
Platform specific messages generate in a non-platform specific source file should be formatted with a .<platform> or .STANDARD suffix.
Example:
Compiler.cmdline.cimmof.cmdline.MENU.PEGASUS_OS_HPUX
Compiler.cmdline.cimmof.cmdline.MENU.PEGASUS_OS_OS40
Compiler.cmdline.cimmof.cmdline.MENU.STANDARD
Where should I place platform specific messages ?
As described in the message bundle file pegasusServer_en.txt messages belong into the section corresponding the file they are created in. This does account the same to platform specific messages.
If a message is generated inside a source file not specific to a single platform, the message should be part of the message bundle section of that source file.
If a new platform specific message is generated inside a platform specific source file, the message belongs to the platform specific section of the message bundle file.
Examples:
ProviderManager.ProviderAgent.ProviderAgent.UNINITIALIZED_SECURITY_SETUP.PEGASUS_OS_ZOS - this message is and should be part of the section for the ProviderAgent as it is generated inside the provider agent and not a z/OS platform specific file
Common.safCheckzOS_inline.BAD_WBEM_SECURITY_SETUP - this message does and should reside inside the platform specific section as the message is generated in a z/OS platform only file
The base Exception class, and derived classes, have been updated to
support localization. Constructors have been added that take a
MessageLoaderParms object. These constructors will use the
MessageLoaderParms object to call the MessageLoader to load the
localized exception message. The localized message is saved in the
Exception. The ContentLanguageList object returned by MessageLoader
is also saved in the Exception. This indicates the language of
the message. The ContentLanguageList object is used later to set the
Content-Language header in the HTTP message to the client.
The old Exception constructors that take a String will remain.
These should be used in cases where the code throwing the exception is
not localized, or the String is not localized (for example, a file
name). Also, there are several exceptions in Pegasus where the
String parameter is meant to be a non-localized substitution in a
localized message owned by the Exception (see InternalException.h,
ClassNotResolved for an example). The old constructors for these
have been kept.
Providers that wish to globalize should consider the following in their
design:
To help providers handle the situations described above, Pegasus 2.3
will pass the Accept-Language received from the client to the
provider. The provider should load strings from its resource
bundle based on the client's Accept-Language. The client's
Accept-Language is passed to the provider in two ways:
The OperationContext will also contain a ContentLanguageList object that
is set from the Content-Language in the client request. This is
the language of the CIM objects being passed to the provider on that
request. A localized provider should store the content language
along with the data from the CIM objects. This will allow the
client to use Accept-Language later to retreive the data in that
language.
The provider should indicate the language of CIM objects it is
returning by calling setContext( ) on the ResponseHandler. This
will be used to set the Content-Language in the CIM response message
sent back to the client. If setContext( ) is not called, then no
Content-Language will be returned to the client. The setContext( )
function should only be called once per response.
The following sample code shows a localized getInstance( ) where the
instance returned is localized based on the Accept-Language of the
client request. Note that this example also throws a localized
exception.
void LocalizedProvider::getInstance(
const OperationContext & context,
const CIMObjectPath & instanceReference,
const Boolean includeQualifiers,
const Boolean includeClassOrigin,
const CIMPropertyList & propertyList,
InstanceResponseHandler & handler)
{
// convert a potential fully qualified
reference into a local reference
// (class name and keys only).
CIMObjectPath localReference = CIMObjectPath(
String(),
String(),
instanceReference.getClassName(),
instanceReference.getKeyBindings());
// begin processing the request
handler.processing();
// Find the instance to be returned.
Uint32 i;
Uint32 n = _instances.size();
for (i = 0; i < n; i++)
{
if(localReference == _instanceNames[i])
{
// We found the instance to return
// Build the parameters for loading the localized string property.
// We are going to let the message loader parameters default to use the
// AcceptLanguageList that Pegasus set into our thread.
// (this equals the AcceptLanguageList requested by the client)
// Note: This parms object could be constructed once and
// reused.
MessageLoaderParms parms("myMsgID", "myDefaultString");
parms.msg_src_path = "/myprovider/msg/myResourceBundle";
// Load the string for the localized property from the resource bundle
String localizedString = MessageLoader::getMessage(parms);
// Remove the old property from the instance to be returned
Uint32 index = instances[i].findProperty("myProperty");
if (index != PEG_NOT_FOUND)
{
_instances[i].removeProperty(index);
}
// Add the localized string property to the instance
instances[i].addProperty(CIMProperty("myProperty", localizedString));
// The MessageLoader set the contentlanguages member
// of parms to the language that it found for the message.
ContentLanguageList rtnLangs = parms.contentlanguages;
// We need to tag the instance we are returning with the
// the
content language.
OperationContext context;
context.insert(ContentLanguageListContainer(rtnLangs));
handler.setContext(context);
break;
} // end if
} //
end for
// throw an exception if
the instance wasn't found
if (i == n)
{
// Build the parameters for loading the localized error message.
// We are going to let the message loader parameters default to use the
// AcceptLanguageList that Pegasus set into our thread.
// (this equals the AcceptLanguageList requested by the client)
// Note: This parms object could be constructed once and
// reused.
MessageLoaderParms errParms("myErrorMsgID", "myErrorDefaultString");
errParms.msg_src_path = "/myprovider/msg/myResourceBundle";
// Note: the exception calls MessageLoader::getMessage( )
// Note: no need to call handler.setContext( ) in this case
throw CIMObjectNotFoundException(errParms);
}
// complete processing
the request
handler.complete();
}
NOTE: A sample provider has been written that fully demonstates the
design issues described above. This provider is located at:
This sample provider also demonstrates how some of the special issues
can be handled. The special issues are caused by having a
read/only localized property and a read/write localized property.
What happens if the client sets the read/write property with a
Content-Language that is not one of the supported languages for the
read/only property? This provider allows the client to set any
language into the read/write property, and get that property back in the
same language. This becomes an issue when the client does a
getInstance( ) later, because the Content-Language on the returned
instance applies to all the properties. A related issue is what to
return for Content-Language when the client does enumerateInstances,
but the instances have different languages. Recall that
Content-Language applies to the entire response (a limitation in the CIM
specification).
NOTE: Indication Providers have other special considerations
for language support. Please refer to PEP58.
NOTE: The CMPI interface has been updated for language
support. Please refer to the CMPI documentation for details.
NOTE: SPECIAL ISSUES FOR OS/400 PROVIDERS:
Methods have been added to CIMClient to set the Accept-Language and
Content-Language on the request, and retrieve Content-Language on the
response. The language tags in the Accept-Language header must
meet the ISO-639 and ISO-3166 standards.
Please refer to
Here is a code fragment that uses the new methods on CIMClient
//
// Get a localized instance in French
//
// Language priority is martian, pig-latin, and
french. We should
// get french back, even though its the lowest priority
AcceptLanguageList acceptLangs;
acceptLangs.insert(LanguageTag("x-martian"), 1.0);
acceptLangs.insert(LanguageTag("fr"), 0.1);
acceptLangs.insert(LanguageTag("x-pig-latin"), 0.4);
// Set the requested languages into the CIMClient
client.setRequestAcceptLanguages(acceptLangs);
// Get the instance
CIMInstance instance = client.getInstance(
NAMESPACE,
cimNInstances[0].buildPath(sampleClass),
localOnly,
includeQualifiers,
includeClassOrigin);
// Get the string property that should be french
String returnedString;
instance.getProperty (
instance.findProperty("myProp")).
getValue().
get(returnedString);
// Check that we got back french
ContentLanguageList CL_FR();
CL_FR.append(LanguageTag("fr"));
String expectedFRString = "oui";
PEGASUS_ASSERT(CL_FR == client.getResponseContentLanguages());
PEGASUS_ASSERT(expectedFRString == returnedString);
//
// Create an instance in French
//
String oui = "Oui";
CIMInstance frInstance(CLASSNAME);
frInstance.addProperty(CIMProperty(
CIMName("myProp"),
oui));
CIMObjectPath frInstanceName = frInstance.buildPath(sampleClass);
client.setRequestContentLanguages(CL_FR);
client.createInstance(NAMESPACE, frInstance);
Also, refer to
NOTE: Consideration should be given for converting the UTF-16
characters in the String objects passed over the CIMClient interface to
a platform codepage. This is especially needed for EBCDIC
platforms. See the Provider developer section for details of the
EBCDIC considerations.
A method has been added to CIMClient to set the Accept-Language for the
requests based on the default locale of the process, as determined by
ICU. If ICU is installed on the client system then CIMClient will
set the Accept-Language from the default ICU process locale. If
ICU is not installed then the caller is required to set an
AcceptLanguageList into CIMClient that meets the ISO-639 and IS0-3166
standards. Note: this is useful for local clients, such as
the Pegasus CLIs, where ICU would be installed on both the client and
server sides.
The design for Pegasus releases beyond 2.3 is to avoid using hardcoded
messages. All new messages should be loaded from a Pegasus
resource bundle. This section describes the process to follow if
you are creating a new message. The process depends on where you
are in the code.
Place any new Pegasus messages into one of the following resource
bundles:
The make messages target will compile these resource bundles.
Note: As described above, the resource bundle path in
MessageLoaderParms defaults to the server resource bundle. For CLI
messages, you will need to specify the bundle for your CLI.
For messages returned from one of the services in the Pegasus server
(eg. CIMOperationRequestDispatcher, or ProviderManagerService), the goal
is to make it easy for any code in the call chain to throw an exception
with a localized error string. The code throwing the exception
will not need to know the Accept-Language that the client
requested. To understand how this works, some design points need
to described:
Server Design Points:
The CIMMessage object has been expanded to include an
AcceptLanguageList object and a ContentLanguageList object in its
OperationContext member. For
CIMRequestMessage, these objects contain the Accept-Language and
Content-Language headers that were built from the client request.
For CIMResponseMessage, the ContentLanguageList object is used to build the
Content-Language header associated with the CIM objects in the
response message. The AcceptLanguageList object in the
CIMResponseMessage is ignored.
The localization of the cimException object in the
CIMResponseMessage is handled separately from the CIM objects. The
message string in the cimException object is assumed to have been
localized by the time it is built into the XML. For this reason,
the localization of the exception is the responsibility of the code
throwing the exception. (The goal of the design is to make that
easy - see below). The ContentLanguageList object in the
CIMResponseMessage has NO relation to this exception. The
cimException object keeps its own localization information once it is
created.
To enable exceptions to be localized, the ability was added to set a
global language for all the code running from a Pegasus Thread
object. The top level code for a Thread can set a global
AcceptLanguageList object that can be accessed by all the low-level
functions that it calls. This will allow an exception thrown by
the low-level function to be localized based on this global
AcceptLanguageList object. Note: This applies only to Threads
that are managed by a ThreadPool.
Each service in the request path of the Pegasus server sets the
AcceptLanguageList into its Thread from the AcceptLanguageList in the
CIMRequestMessage object that it dequeues. This sets the global
langauge for all the functions in the same thread that are called below
handleEnqueue. If you are writing a new service that processes
requests, or discover a request service that was missed, please do
this. The CIMOperationRequestDispatcher service is an example.
How to Throw a Localized Exception from Server code:
With all that background, here is how code running in a Pegasus
service can throw a localized exception:
This example assumes that the top-level code in the service had set the
global thread AcceptLanguageList beforehand. As described above,
every service in Pegasus should do that. The code here may be
buried several layers deep in the call chain, but does not need to know
the AcceptLanguagList of the current client request.
// First, construct a MessageLoaderParms
//
// Notes:
// 1) The errorMessageID must be in the Pegasus server resource
bundle.
// 2) The default message is the old "hardcoded" message.
// 3) The MessageLoaderParms will default to use the Pegasus
server resource bundle
// 4) The MessageLoaderParms will default to use the
AcceptLanguageList set into the current Thread. Don't change this!
// 5) You might need to set the arguments for the message into
the MessageLoaderParms
MessageLoaderParms parms("errorMessageID", "default message");
// Second, throw the Exception
// Note: this applies to all the derived classes from Exception,
including the CIMException's
throw new Exception(parms);
NOTE: If you are throwing an Exception with un-localized data,
use the constructor that takes a String. An example of this would
be an Exception where you are passing in a file name. Most of the
"non-CIM" exceptions defined in Exception.h and InternalException.h take
un-localized data.
The Exception Macros
There are many spots in the server code that use the PEGASUS_CIM_EXCEPTION macro to throw a TraceableCIMException. The use of this macro in the code like the following example presented a design problem:
....
} catch (Exception & e)
{
throw PEGASUS_CIM_EXCEPTION(CIM_ERR_FAILED,
e.getMessage());
}
This type of code would have lost the ContentLanguageList saved in "e",
so that the Content-Language would not be set in HTTP response to the
client.
For Pegasus 2.3, these types of macro calls can stay. The
TraceableCIMException constructed by the macro will "re-localize".
That is, the "CIM" part of the message (the part based on the error
code) will be localized at throw time, and the ContentLanguageList
re-established. A key is to avoid a "language mismatch" problem
between the CIM part of the message and the extra part of the
message. The design point here is that all internal exceptions
thrown by Pegasus code are localized using the global AcceptLanguageList
of the Thread...see above.
In the future, it will be safer and more maintainable to use of
the new "localized" flavors of the macro. For example:
When the message from a caught Exception needs to be become the extra message in a thrown CIMException:
....
} catch (Exception & e)
{
throw
PEGASUS_CIM_EXCEPTION_LANG(e.getContentLanguages( ),
CIM_ERR_FAILED,
e.getMessage( ));
}
This guarantees that the ContentLanguageList in "e" is copied to the
newly created TraceableCIMException.
In the case where the extra message for the CIMException is
determined by the throwing code:
throw PEGASUS_CIM_EXCEPTION_L(CIM_ERR_FAILED,
MessageLoaderParms("Repository.CIMRepository.COMPACT_FAILED",
"compact failed"));
(example from CIMRepository.cpp)
This uses a MessageLoaderParms object to localize the extra message in
the newly created TraceableCIMException.
New methods have been added to Logger to take a message ID of a message
to be loaded from the Pegasus server resource bundle. The caller
is only required to pass in the message ID, the old "hardcoded" message,
and the args. The Logger will use MessageLoader to load the
message in the locale of the Pegasus server process, using the
hardcoded message as the default string. Please refer to
pegasus/src/Pegasus/Logger.h.
Note: Messages sent to the "logs", whether the system logs or
the Pegasus log file, are converted to UTF-8 before being sent.
The goal for messages returned by the Pegasus CLIs is to localize in
the locale of the user running the CLI. This should be automatic
-- the user should not be required to tell the CLI what the locale
is. For the CLIs that are CIM clients (cimconfing,
cimprovider) there are two sets of messages to localize --
messages generated in the CLI process itself, and messages returned from
the Pegasus server . For CLIs that are directly linked into
Pegasus (cimmofl), all the messages are generated in the CLI's process,
but the CLI may call Pegasus APIs that are coded to localize based on a
client's requested languages.
Code in the client side of the client/server CLIs (eg. cimconfig,
cimmof), or in directly linked CLIs (cimmofl), should use the
_useProcessLocale "master switch" described in the Message Loading
section. This will cause all messages, including exceptions thrown
by Pegasus APIs, to be loaded in the locale based on the
environment in which the program is running. This locale can be
set by the user before running the program.
Code in the client side of the client/server CLIs need to send an
Accept-Language to the Pegasus server that reflects the default locale
of the CLI's process. See the Client Developer section for
details.
An example of these considerations can be seen in the source code
for cimconfig.
Copyright (c) 2000, 2001, 2002 BMC Software; Hewlett-Packard Development Company, L.P.; IBM Corp.; The Open Group; Tivoli Systems. Copyright (c) 2003 BMC Software; Hewlett-Packard Development Company, L.P.; IBM Corp.; EMC Corporation, The Open Group. Copyright (c) 2004 BMC Software; Hewlett-Packard Development Company, L.P.; IBM Corp.; EMC Corporation; VERITAS Software Corporation; The Open Group. Copyright (c) 2005 Hewlett-Packard Development Company, L.P.; IBM Corp.; EMC Corporation; VERITAS Software Corporation; The Open Group. Copyright (c) 2006 Hewlett-Packard Development Company, L.P.; IBM Corp.; EMC Corporation; Symantec Corporation; The Open Group.
Permission is hereby granted, free of charge, to any person obtaining a copy of this software and associated documentation files (the "Software"), to deal in the Software without restriction, including without limitation the rights to use, copy, modify, merge, publish, distribute, sublicense, and/or sell copies of the Software, and to permit persons to whom the Software is furnished to do so, subject to the following conditions:
THE ABOVE COPYRIGHT NOTICE AND THIS PERMISSION NOTICE SHALL BE INCLUDED IN
ALL COPIES OR SUBSTANTIAL PORTIONS OF THE SOFTWARE. THE SOFTWARE IS PROVIDED
"AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR IMPLIED, INCLUDING BUT NOT
LIMITED TO THE WARRANTIES OF MERCHANTABILITY, FITNESS FOR A PARTICULAR
PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE AUTHORS OR COPYRIGHT
HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN
ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.