Add composite-type attributes to information_schema.element_types view

[pg-rex/syncrep.git] / doc / src / sgml / nls.sgml

<!-- doc/src/sgml/nls.sgml -->

<chapter id="nls">
 <chapterinfo>
  <author>
   <firstname>Peter</firstname>
   <surname>Eisentraut</surname>
  </author>
 </chapterinfo>

 <title>Native Language Support</title>

 <sect1 id="nls-translator">
  <title>For the Translator</title>

  <para>
   <productname>PostgreSQL</>
   programs (server and client) can issue their messages in
   your favorite language &mdash; if the messages have been translated.
   Creating and maintaining translated message sets needs the help of
   people who speak their own language well and want to contribute to
   the <productname>PostgreSQL</> effort.  You do not have to be a
   programmer at all
   to do this.  This section explains how to help.
  </para>

  <sect2>
   <title>Requirements</title>

   <para>
    We won't judge your language skills &mdash; this section is about
    software tools.  Theoretically, you only need a text editor.  But
    this is only in the unlikely event that you do not want to try out
    your translated messages.  When you configure your source tree, be
    sure to use the <option>--enable-nls</option> option.  This will
    also check for the <application>libintl</application> library and the
    <filename>msgfmt</filename> program, which all end users will need
    anyway.  To try out your work, follow the applicable portions of
    the installation instructions.
   </para>

   <para>
    If you want to start a new translation effort or want to do a
    message catalog merge (described later), you will need the
    programs <filename>xgettext</filename> and
    <filename>msgmerge</filename>, respectively, in a GNU-compatible
    implementation.  Later, we will try to arrange it so that if you
    use a packaged source distribution, you won't need
    <filename>xgettext</filename>.  (If working from Git, you will still need
    it.)  <application>GNU Gettext 0.10.36</application> or later is currently recommended.
   </para>

   <para>
    Your local gettext implementation should come with its own
    documentation.  Some of that is probably duplicated in what
    follows, but for additional details you should look there.
   </para>
  </sect2>

  <sect2>
   <title>Concepts</title>

   <para>
    The pairs of original (English) messages and their (possibly)
    translated equivalents are kept in <firstterm>message
    catalogs</firstterm>, one for each program (although related
    programs can share a message catalog) and for each target
    language.  There are two file formats for message catalogs:  The
    first is the <quote>PO</quote> file (for Portable Object), which
    is a plain text file with special syntax that translators edit.
    The second is the <quote>MO</quote> file (for Machine Object),
    which is a binary file generated from the respective PO file and
    is used while the internationalized program is run.  Translators
    do not deal with MO files; in fact hardly anyone does.
   </para>

   <para>
    The extension of the message catalog file is to no surprise either
    <filename>.po</filename> or <filename>.mo</filename>.  The base
    name is either the name of the program it accompanies, or the
    language the file is for, depending on the situation.  This is a
    bit confusing.  Examples are <filename>psql.po</filename> (PO file
    for psql) or <filename>fr.mo</filename> (MO file in French).
   </para>

   <para>
    The file format of the PO files is illustrated here:
<programlisting>
# comment

msgid "original string"
msgstr "translated string"

msgid "more original"
msgstr "another translated"
"string can be broken up like this"

...
</programlisting>
    The msgid's are extracted from the program source.  (They need not
    be, but this is the most common way.)  The msgstr lines are
    initially empty and are filled in with useful strings by the
    translator.  The strings can contain C-style escape characters and
    can be continued across lines as illustrated.  (The next line must
    start at the beginning of the line.)
   </para>

   <para>
    The # character introduces a comment.  If whitespace immediately
    follows the # character, then this is a comment maintained by the
    translator.  There can also be automatic comments, which have a
    non-whitespace character immediately following the #.  These are
    maintained by the various tools that operate on the PO files and
    are intended to aid the translator.
<programlisting>
#. automatic comment
#: filename.c:1023
#, flags, flags
</programlisting>
    The #. style comments are extracted from the source file where the
    message is used.  Possibly the programmer has inserted information
    for the translator, such as about expected alignment.  The #:
    comment indicates the exact location(s) where the message is used
    in the source.  The translator need not look at the program
    source, but he can if there is doubt about the correct
    translation.  The #, comments contain flags that describe the
    message in some way.  There are currently two flags:
    <literal>fuzzy</literal> is set if the message has possibly been
    outdated because of changes in the program source.  The translator
    can then verify this and possibly remove the fuzzy flag.  Note
    that fuzzy messages are not made available to the end user.  The
    other flag is <literal>c-format</literal>, which indicates that
    the message is a <function>printf</function>-style format
    template.  This means that the translation should also be a format
    string with the same number and type of placeholders.  There are
    tools that can verify this, which key off the c-format flag.
   </para>
  </sect2>

  <sect2>
   <title>Creating and Maintaining Message Catalogs</title>

   <para>
    OK, so how does one create a <quote>blank</quote> message
    catalog?  First, go into the directory that contains the program
    whose messages you want to translate.  If there is a file
    <filename>nls.mk</filename>, then this program has been prepared
    for translation.
   </para>

   <para>
    If there are already some <filename>.po</filename> files, then
    someone has already done some translation work.  The files are
    named <filename><replaceable>language</replaceable>.po</filename>,
    where <replaceable>language</replaceable> is the
    <ulink url="http://www.loc.gov/standards/iso639-2/php/English_list.php">
    ISO 639-1 two-letter language code (in lower case)</ulink>, e.g.,
    <filename>fr.po</filename> for French.  If there is really a need
    for more than one translation effort per language then the files
    can also be named
    <filename><replaceable>language</replaceable>_<replaceable>region</replaceable>.po</filename>
    where <replaceable>region</replaceable> is the
    <ulink url="http://www.iso.org/iso/english_country_names_and_code_elements">
    ISO 3166-1 two-letter country code (in upper case)</ulink>,
    e.g.,
    <filename>pt_BR.po</filename> for Portuguese in Brazil.  If you
    find the language you wanted you can just start working on that
    file.
   </para>

   <para>
    If you need to start a new translation effort, then first run the
    command:
<programlisting>
gmake init-po
</programlisting>
    This will create a file
    <filename><replaceable>progname</replaceable>.pot</filename>.
    (<filename>.pot</filename> to distinguish it from PO files that
    are <quote>in production</quote>. The <literal>T</> stands for
    <quote>template</>.)
    Copy this file to
    <filename><replaceable>language</replaceable>.po</filename> and
    edit it.  To make it known that the new language is available,
    also edit the file <filename>nls.mk</filename> and add the
    language (or language and country) code to the line that looks like:
<programlisting>
AVAIL_LANGUAGES := de fr
</programlisting>
    (Other languages can appear, of course.)
   </para>

   <para>
    As the underlying program or library changes, messages might be
    changed or added by the programmers.  In this case you do not need
    to start from scratch.  Instead, run the command:
<programlisting>
gmake update-po
</programlisting>
    which will create a new blank message catalog file (the pot file
    you started with) and will merge it with the existing PO files.
    If the merge algorithm is not sure about a particular message it
    marks it <quote>fuzzy</quote> as explained above.  The new PO file
    is saved with a <filename>.po.new</filename> extension.
   </para>
  </sect2>

  <sect2>
   <title>Editing the PO Files</title>

   <para>
    The PO files can be edited with a regular text editor.  The
    translator should only change the area between the quotes after
    the msgstr directive, add comments, and alter the fuzzy flag.
    There is (unsurprisingly) a PO mode for Emacs, which I find quite
    useful.
   </para>

   <para>
    The PO files need not be completely filled in.  The software will
    automatically fall back to the original string if no translation
    (or an empty translation) is available.  It is no problem to
    submit incomplete translations for inclusions in the source tree;
    that gives room for other people to pick up your work.  However,
    you are encouraged to give priority to removing fuzzy entries
    after doing a merge.  Remember that fuzzy entries will not be
    installed; they only serve as reference for what might be the right
    translation.
   </para>

   <para>
    Here are some things to keep in mind while editing the
    translations:
    <itemizedlist>
     <listitem>
      <para>
       Make sure that if the original ends with a newline, the
       translation does, too.  Similarly for tabs, etc.
      </para>
     </listitem>

     <listitem>
      <para>
       If the original is a <function>printf</> format string, the translation
       also needs to be.  The translation also needs to have the same
       format specifiers in the same order.  Sometimes the natural
       rules of the language make this impossible or at least awkward.
       In that case you can modify the format specifiers like this:
<programlisting>
msgstr "Die Datei %2$s hat %1$u Zeichen."
</programlisting>
       Then the first placeholder will actually use the second
       argument from the list.  The
       <literal><replaceable>digits</replaceable>$</literal> needs to
       follow the % immediately, before any other format manipulators.
       (This feature really exists in the <function>printf</function>
       family of functions.  You might not have heard of it before because
       there is little use for it outside of message
       internationalization.)
      </para>
     </listitem>

     <listitem>
      <para>
       If the original string contains a linguistic mistake, report
       that (or fix it yourself in the program source) and translate
       normally.  The corrected string can be merged in when the
       program sources have been updated.  If the original string
       contains a factual mistake, report that (or fix it yourself)
       and do not translate it.  Instead, you can mark the string with
       a comment in the PO file.
      </para>
     </listitem>

     <listitem>
      <para>
       Maintain the style and tone of the original string.
       Specifically, messages that are not sentences (<literal>cannot
       open file %s</literal>) should probably not start with a
       capital letter (if your language distinguishes letter case) or
       end with a period (if your language uses punctuation marks).
       It might help to read <xref linkend="error-style-guide">.
      </para>
     </listitem>

     <listitem>
      <para>
       If you don't know what a message means, or if it is ambiguous,
       ask on the developers' mailing list.  Chances are that English
       speaking end users might also not understand it or find it
       ambiguous, so it's best to improve the message.
      </para>
     </listitem>

    </itemizedlist>
   </para>
  </sect2>

 </sect1>


 <sect1 id="nls-programmer">
  <title>For the Programmer</title>

  <sect2 id="nls-mechanics">
   <title>Mechanics</title>

  <para>
   This section describes how to implement native language support in a
   program or library that is part of the
   <productname>PostgreSQL</> distribution.
   Currently, it only applies to C programs.
  </para>

  <procedure>
   <title>Adding NLS Support to a Program</title>

   <step>
    <para>
     Insert this code into the start-up sequence of the program:
<programlisting>
#ifdef ENABLE_NLS
#include &lt;locale.h&gt;
#endif

...

#ifdef ENABLE_NLS
setlocale(LC_ALL, "");
bindtextdomain("<replaceable>progname</replaceable>", LOCALEDIR);
textdomain("<replaceable>progname</replaceable>");
#endif
</programlisting>
     (The <replaceable>progname</replaceable> can actually be chosen
     freely.)
    </para>
   </step>

   <step>
    <para>
     Wherever a message that is a candidate for translation is found,
     a call to <function>gettext()</function> needs to be inserted.  E.g.:
<programlisting>
fprintf(stderr, "panic level %d\n", lvl);
</programlisting>
     would be changed to:
<programlisting>
fprintf(stderr, gettext("panic level %d\n"), lvl);
</programlisting>
     (<symbol>gettext</symbol> is defined as a no-op if NLS support is
     not configured.)
    </para>

    <para>
     This tends to add a lot of clutter.  One common shortcut is to use:
<programlisting>
#define _(x) gettext(x)
</programlisting>
     Another solution is feasible if the program does much of its
     communication through one or a few functions, such as
     <function>ereport()</function> in the backend.  Then you make this
     function call <function>gettext</function> internally on all
     input strings.
    </para>
   </step>

   <step>
    <para>
     Add a file <filename>nls.mk</filename> in the directory with the
     program sources.  This file will be read as a makefile.  The
     following variable assignments need to be made here:

     <variablelist>
      <varlistentry>
       <term><varname>CATALOG_NAME</varname></term>

       <listitem>
        <para>
         The program name, as provided in the
         <function>textdomain()</function> call.
        </para>
       </listitem>
      </varlistentry>

      <varlistentry>
       <term><varname>AVAIL_LANGUAGES</varname></term>

       <listitem>
        <para>
         List of provided translations &mdash; initially empty.
        </para>
       </listitem>
      </varlistentry>

      <varlistentry>
       <term><varname>GETTEXT_FILES</varname></term>

       <listitem>
        <para>
         List of files that contain translatable strings, i.e., those
         marked with <function>gettext</function> or an alternative
         solution.  Eventually, this will include nearly all source
         files of the program.  If this list gets too long you can
         make the first <quote>file</quote> be a <literal>+</literal>
         and the second word be a file that contains one file name per
         line.
        </para>
       </listitem>
      </varlistentry>

      <varlistentry>
       <term><varname>GETTEXT_TRIGGERS</varname></term>

       <listitem>
        <para>
         The tools that generate message catalogs for the translators
         to work on need to know what function calls contain
         translatable strings.  By default, only
         <function>gettext()</function> calls are known.  If you used
         <function>_</function> or other identifiers you need to list
         them here.  If the translatable string is not the first
         argument, the item needs to be of the form
         <literal>func:2</literal> (for the second argument).
         If you have a function that supports pluralized messages,
         the item should look like <literal>func:1,2</literal>
         (identifying the singular and plural message arguments).
        </para>
       </listitem>
      </varlistentry>
     </variablelist>
    </para>
   </step>

  </procedure>

  <para>
   The build system will automatically take care of building and
   installing the message catalogs.
  </para>
  </sect2>

  <sect2 id="nls-guidelines">
   <title>Message-writing Guidelines</title>

  <para>
   Here are some guidelines for writing messages that are easily
   translatable.

   <itemizedlist>
    <listitem>
     <para>
      Do not construct sentences at run-time, like:
<programlisting>
printf("Files were %s.\n", flag ? "copied" : "removed");
</programlisting>
      The word order within the sentence might be different in other
      languages.  Also, even if you remember to call <function>gettext()</> on
      each fragment, the fragments might not translate well separately.  It's
      better to duplicate a little code so that each message to be
      translated is a coherent whole.  Only numbers, file names, and
      such-like run-time variables should be inserted at run time into
      a message text.
     </para>
    </listitem>

    <listitem>
     <para>
      For similar reasons, this won't work:
<programlisting>
printf("copied %d file%s", n, n!=1 ? "s" : "");
</programlisting>
      because it assumes how the plural is formed.  If you figured you
      could solve it like this:
<programlisting>
if (n==1)
    printf("copied 1 file");
else
    printf("copied %d files", n):
</programlisting>
      then be disappointed.  Some languages have more than two forms,
      with some peculiar rules.  It's often best to design the message
      to avoid the issue altogether, for instance like this:
<programlisting>
printf("number of copied files: %d", n);
</programlisting>
     </para>

     <para>
      If you really want to construct a properly pluralized message,
      there is support for this, but it's a bit awkward.  When generating
      a primary or detail error message in <function>ereport()</>, you can
      write something like this:
<programlisting>
errmsg_plural("copied %d file",
              "copied %d files",
              n,
              n)
</programlisting>
      The first argument is the format string appropriate for English
      singular form, the second is the format string appropriate for
      English plural form, and the third is the integer control value
      that determines which plural form to use.  Subsequent arguments
      are formatted per the format string as usual.  (Normally, the
      pluralization control value will also be one of the values to be
      formatted, so it has to be written twice.)  In English it only
      matters whether <replaceable>n</> is 1 or not 1, but in other
      languages there can be many different plural forms.  The translator
      sees the two English forms as a group and has the opportunity to
      supply multiple substitute strings, with the appropriate one being
      selected based on the run-time value of <replaceable>n</>.
     </para>

     <para>
      If you need to pluralize a message that isn't going directly to an
      <function>errmsg</> or <function>errdetail</> report, you have to use
      the underlying function <function>ngettext</>.  See the gettext
      documentation.
     </para>
    </listitem>

    <listitem>
     <para>
      If you want to communicate something to the translator, such as
      about how a message is intended to line up with other output,
      precede the occurrence of the string with a comment that starts
      with <literal>translator</literal>, e.g.:
<programlisting>
/* translator: This message is not what it seems to be. */
</programlisting>
      These comments are copied to the message catalog files so that
      the translators can see them.
     </para>
    </listitem>
   </itemizedlist>
  </para>
  </sect2>
 </sect1>

</chapter>