This document describes how messages will be input, stored and formatted
by a Win32 application.

1.  Message Input

    Messages are input as ASCII text in a text file.  The format of this
    file supports specifying multiple versions of the same message text,
    one for each language supported.  It also supports automatic assignment
    of code numbers to each message, along with the generation of a C
    language include file for use by the application for accessing the
    messages using symbolic constants.  The purpose of the message text
    file is to define all of the messages needed by an application, in a
    format that makes it easy to support multiple languages with the same
    image file.

    Message text files are converted into binary resource files by the
    MC program.  These binary resource files are then input to the RC
    compiler which will put them in the resource table for an
    application or DLL.

    The format of the message text file (default extension is .mc).
    Basic syntax is Keyword=Value, where spaces around the equal sign
    are ignored, and the value is delimited by white space from the next
    keyword=value pair.  Case is ignored when comparing against keyword
    names.  The value portion can either be a numeric integer constant,
    {NUMBER}, using C syntax; a symbol name, {NAME}, that follows the
    rules for C identifier names; or a file name that follows the
    rules for the FAT file system (8 characters or less, no periods).

    Comment lines are allowed in the message text file.  The comment
    syntax is the same as for WIN.INI, namely a semicolon begins a
    comment which is terminated by the end of the line.  Comments that
    exist by themselves on a line are copied as is to the output .h
    file, with the semicolon converted to the C end of line comment
    syntax (//).

    The overall structure of a message text file consists of a header
    section which contains zero or more of the following keywords:

        MessageIdTypedef={NAME}
        SeverityNames=({NAME}={NUMBER}:{NAME})
        FacilityNames=({NAME}={NUMBER}:{NAME})
        LanguageNames=({NAME}={NUMBER}:{FILENAME})

    These keywords have the following meaning:

        MessageIdTypedef - gives a symbolic name that is output as the
            typedef name for each numeric MessageId value.  The default
            value for this is NULL, which means there will be no type
            cast output when defining symbolic names for a MessageId.

        SeverityNames - defines the set of names that are allowed as the
            value of the Severity keyword in the message definition.
            The set is delimited by left and right parenthesis.
            Associated with each severity name is a number that, when
            shifted left by 30, gives the bit pattern to logically OR
            with the Facility value and MessageId to come up with the
            full 32-bit message code.  The default value of this keyword
            is:

                SeverityNames=(Success=0x0
                               Informational=0x1
                               Warning=0x2
                               Error=0x3
                              )

            Severity values occupy the high two bits of a 32-bit message
            code.  Any severity value that does not fit in two bits is
            an error.  The severity codes can be given symbolic names
            by following each value with :{NAME}

        FacilityNames - defines the set of names that are allowed as the
            value of the Facility keyword in the message definition.
            The set is delimited by left and right parenthesis.
            Associated with each facility name is a number that, when
            shift it left by 16 bits, gives the bit pattern to logically
            OR with the Severity value and MessageId to come up with the
            full 32-bit message code.  The default value of this keyword
            is:

                FacilityNames=(System=0x0FF
                               Application=0xFFF
                              )


            Facility codes occupy the low order 12 bits of the high
            order 16-bits of a 32-bit message code.  Any facility code
            that does not fit in 12 bits is an error.  This allows for
            4096 facility codes.  The first 256 are reserved for
            use by the system software.

            The facility codes can be given symbolic names by following
            each value with :{NAME}

        LanguageNames - defines the set of names that are allowed as the
            value of the Language keyword in the message definition.
            The set is delimited by left and right parenthesis.
            Associated with each language name is a number and a file
            name that will be used to name the binary output file that
            will contain all of the message text for that language.  The
            number corresponds to the Language Id tag to use in the
            resource table.  The number is separate from the file name
            with a colon.  The initial value of this keyword is:

                LanguageNames=(English=1:MSG00001)

            Any new names that an application defines in its .mc file
            which don't override any of the builtin names will be added
            to the list of valid languages.  This allows an application
            to support private languages with descriptive names.

    Following the header section are zero or more message definitions.
    Each message definition begins with one or more of the following
    keywords.

        MessageId={|{NUMBER}|+{NUMBER}}
        Severity={SEVERITY_NAME}
        Facility={FACILITY_NAME}
        SymbolicName={NAME}

    The MessageId keyword is required to mark the beginning of the
    message definition, although its value is optional.  If no value is
    specified, then the value used will be the last value used for the
    facility, plus one.  If the value is specified as +{NUMBER} then
    the value used will be the last value used for the facility, plus
    the number after the plus sign.  Otherwise if a numeric value is
    given, that will be value used.  Any MessageId value that does not
    fit in 16 bits is an error.

    Severity and Facility are optional keywords that can specify
    additional bits to OR into the final 32-bit message code.  If either
    of these are not specified they default to the value last specified
    for a message definition.  The initial values of these prior to
    processing the first message definition is:

        Severity=Success
        Facility=Application

    The value associated with these keywords must match one of the names
    given to the FacilityNames and SeverityNames keywords.  The SymbolicName
    keyword allows the ISV to associate a C symbolic constant name with the
    final 32-bit message code that is a result of ORing together the
    MessageId, Severity and Facility bits.  The constant definition is
    output to the generated .h file with the following format:

        //
        // {MESSAGETEXT}
        //

        #define CONSTANT_SYMBOL_NAME ((MessageIdTypedef) 0x12345678)

    where the comment before the definition is a copy of the message
    text for the first language specified in the message definition.
    The CONSTANT_SYMBOL_NAME is the value of the SymbolicName keyword.
    The MessageIdTypedef is not output if it is NULL, the default value.

    After the message definition keywords, comes one or more message text
    definitions.  Each message text definition begins with the Language
    keyword that identifies which binary output file this message text
    is to be output to.  Beginning on the very next line is the first
    line of the message text.  The message text is terminated by a line
    containing a single period at the beginning of the line, immediately
    followed by a new line.  No spaces allowed around keyword.  Within
    the message text, blank lines and white space are preserved as part
    of the message.

        Language={LANGUAGE_NAME}
        {MESSAGETEXT}
        .

    Within the message text, several escape sequences are supported for
    dynamically formatting the message.  The percent sign character (%)
    begins all escape sequences.


        %0 - This terminates a message text line without a trailing
             newline.  This can be used to build up long lines or to
             terminate the message itself without a trailing newline,
             which is useful for prompt messages.

        %n!printf format string!  - This identifies an insert.  The
             value of n can be between 1 and 99.  The printf format
             string must be bracketed by exclamation marks.  It is
             optional and defaults to !s! if not specified.

             The printf format string can contain the * specifier for
             either the precision or width components, and if so, they
             will consume inserts %n+1 and %n+2 for their values at run
             time.  MC will print a warning message if an explicit
             reference is made to these inserts elsewhere in the message
             text.

    Inserts must reference a parameter passed to the FormatMessage API
    call.  It will return an error if a message refers to an insert that
    was not passed to the FormatMessage API call.

    Any other character following a percent sign, other than a digit will
    be formatted in the output message without the percent sign.  Some
    examples:

        %% - will output a single percent sign in the formatted message text.

        %n - will output a hard line break when it occurs at the end of a
             a line.  Useful when FormatMessage is supplying normal line
             breaks so the message fits in a certain width.

        %r - will output a hard carriage return, without a trailing newline.

        %b - will output a space in the formatted message text.  This
             can be used to insure there are the appropriate number of
             trailing spaces in a message text line.

        %t - will output a tab in the formatted message text.

        %. - will output a single period in the formatted message text.
             This can be used to get a single period at the beginning of
             a line without terminating the message text definition.

        %! - will output a single exclamation point in the formatted
             message text.  This can be used to get an exclamation point
             immediately after an insert without it being mistaken for
             the beginning of a printf format string.

    Unicode support is not understood yet.  If the input file is ASCII
    text, do we need an escape sequence to allow input of Unicode values?
    Or do we just let them use DBCS in the text file, assuming they have
    a text editor that can do this.

2.  Message Compiler (MC)

    This program converts .mc message text files into binary files
    suitable for inclusion into a .RC file by the resource compiler.

    Command line syntax:

        MC [-v] [-w] [-s] [-h DirSpec] [-r DirSpec] filename[.mc] ...

    where:

        -v - generates verbose output to stderr.

        -w - generates a warning message whenever an insert escape
             sequence is seen that is a superset of the type supported
             by OS/2 mkmsgf (i.e.  anything other than %0 and %n).
             Useful for converting old OS/2 message files to this
             format.

        -s - Add an extra line to the beginning of each message that is
             the symbolic name associated with the message id.

        -h DirSpec - specifies the target directory of the generated
             .h file.  The file name is the name of the .mc file with a
             .h extension.

        -r DirSpec - specifies the target directory of the generated
             .rc file.  The file name is the name of the .mc file with a
             .rc extension.

        filename.mc - specifes one or more input message files that
                      will be compiled into one or more binary resource
                      files, one for each language that the message
                      files contain message text for.

    The message compiler reads the .mc file and generates a .h file
    containing all the symbolic name definitions.  For each LanguageId
    that was used to specify message text, it outputs a binary file
    containing a message table resource.  It also outputs a single .rc
    file that contains the appropriate RC syntax to include each binary
    file output as a resource with the appropriate name and type ids.

3.  Message Win32 API Calls

DWORD
APIENTRY
FormatMessage(
    DWORD dwFlags,
    LPVOID lpSource,
    DWORD dwMessageId,
    DWORD dwLanguageId,
    LPSTR lpBuffer,
    DWORD nSize,
    va_list Arguments
    )

Routine Description:

    This function formats a message string.  Input to this function is a
    message definition.  It can come from a buffer passed into this
    function.  It can come from a message table resource in a module
    already loaded.  Or the caller can ask this function to search the
    system message table resource(s) for the message.  This function
    finds the message definition based on the Message Id and the
    Language Id and copies the message text to the output buffer,
    processing any imbedded insert sequences if requested.

Arguments:

    dwFlags - Specifies options to the formatting process along with how
        to interpret the lpSource parameter.  The low order 16bits of
        this parameter are the maximum width of a line, in characters.
        Possible values are:

        FORMAT_MESSAGE_ALLOCATE_BUFFER - the lpBuffer is a PVOID * and
            nSize is the minimum size to allocate.  This function will
            then allocate a buffer large enought to hold the formatted
            message and store the pointer to the buffer in the location
            pointed to by lpBuffer.   Caller should free the buffer
            with LocalFree when they are done using it.

        FORMAT_MESSAGE_IGNORE_INSERTS - insert sequences in the message
            definition will be ignored and passed through to the output
            buffer as is.  Useful for fetching a message for later
            formatting.  If this flag is set, the lpArguments parameter
            is ignored.

        FORMAT_MESSAGE_FROM_STRING - lpSource is a pointer to a null
            terminated message definition.  It can contain insert
            sequences just as the message text in the .mc file can.

        FORMAT_MESSAGE_FROM_HMODULE - lpSource is a module handle that
            contains the message table resource(s) to search.  If this
            handle is NULL, then the current process's application
            image file will be searched.

        FORMAT_MESSAGE_FROM_SYSTEM - If the requested message was not
            found in lpSource or if lpSource was not examined (i.e. neither
            of the preceeding two flags was specified), then this function
            will search the system message table resource(s).

        FORMAT_MESSAGE_ARGUMENT_ARRAY - If set, specifies that the passed
            Arguments parameter is NOT a va_list structure but instead is
            just a pointer to an array of 32-bit values that represent
            the arguments.

        FORMAT_MESSAGE_MAX_WIDTH_MASK - The low order 8 bits specify the
            maximum width of each line formatted into the output buffer.
            A maximum width of zero, means that no restrictions are
            placed on the width, and only the line breaks in the message
            definition will be placed in the output buffer.  If a non-zero
            value is specified, then line breaks in the message definition
            text are ignored, and instead line breaks are calculated based
            on the maximum width, with white space delimited strings never
            being split across a line break.  Hard coded line breaks in
            the message definition text, that are identified by the %n
            escape sequence, are always output to the output buffer.

            If the width specified is FORMAT_MESSAGE_MAX_WIDTH_MASK, then
            line breaks in the message file are ignored and only hard coded
            line breaks are kept and none are generated.

    lpSource - specifies where to retrieve the message definition from.
        The type of this parameter depends upon the settings in the
        dwFlags parameter.

            FORMAT_MESSAGE_FROM_HMODULE - lpSource is an hModule of the
                module that contains the message table to search.

            FORMAT_MESSAGE_FROM_STRING - lpSource is an LPSTR that points
                to unformatted message text.  It will be scanned for
                inserts and formatted accordingly.

        If neither of these options is specified, then this parameter is
        ignored.

    dwMessageId - specifices the 32-bit message identifier that identifies
        the message being requested.  This parameter is ignored if the
        FORMAT_MESSAGE_FROM_STRING flag is specified.

    dwLanguageId - specifices the 32-bit language identifier that
        identifies the language of the message being requested.  This
        parameter is ignored if the FORMAT_MESSAGE_FROM_STRING flag is
        specified.

    lpBuffer - specifies a pointer to a buffer where the formatted message
        is to be written.  A terminating null byte will also be written.
        If the FORMAT_MESSAGE_ALLOCATE_BUFFER flag was specified, then
        this parameter points to a 32-bit pointer value that is filled in
        by this call with a pointer allocated via LocalAlloc to contain
        the text of the message.

    nSize - specifies the maximum number of bytes that can be stored in
        the output buffer.  This parameter is ignore if the
        FORMAT_MESSAGE_ALLOCATE_BUFFER flag is set.

    Arguments - specifies a pointer to variable number of arguments.
        These arguments are used to satisfy insert requests in the
        format string.  Thus %1 in the format string specifies the
        first argument in the variable number of arguments described
        by the Arguments parameter; %3 would specify the third, etc.

        The interpretation of each 32-bit arguments value depends upon
        the formatting information associated with the insert in the
        message definition.  The default is to treat each pointer as a
        pointer to a null terminated string.

        By default the Arguments parameter is of type va_list, which is
        a language and implementation specific data type for describing
        a variable number of arguments.  If you do not have a pointer of
        type va_list, then specify the FORMAT_MESSAGE_ARGUMENT_ARRAY
        flag and pass a pointer to an array of 32-bit values that are
        are input to the message formatted as the insert values.
Return Value:

    DwORD - Returns the number of bytes actually stored in the output
        buffer, excluding the terminating null character.  Returns 0 if
        an error occurred.  Extended error status is available via the
        GetLastError API.
