Locale::Maketext(3) User Contributed Perl Documentation Locale::Maketext(3)
NAME
Locale::Maketext - framework for localization
SYNOPSIS
package MyProgram;
use strict;
use MyProgram::L10N;
# ...which inherits from Locale::Maketext
my $lh = MyProgram::L10N->get_handle() || die "What language?";
...
# And then any messages your program emits, like:
warn $lh->maketext( "Can't open file [_1]: [_2]\n", $f, $! );
...
DESCRIPTION
It is a common feature of applications (whether run directly, or via
the Web) for them to be "localized" -- i.e., for them to a present an
English interface to an English-speaker, a German interface to a
German-speaker, and so on for all languages it's programmed with.
Locale::Maketext is a framework for software localization; it provides
you with the tools for organizing and accessing the bits of text and
text-processing code that you need for producing localized
applications.
In order to make sense of Maketext and how all its components fit
together, you should probably go read Locale::Maketext::TPJ13, and then
read the following documentation.
You may also want to read over the source for "File::Findgrep" and its
constituent modules -- they are a complete (if small) example
application that uses Maketext.
QUICK OVERVIEW
The basic design of Locale::Maketext is object-oriented, and
Locale::Maketext is an abstract base class, from which you derive a
"project class". The project class (with a name like
"TkBocciBall::Localize", which you then use in your module) is in turn
the base class for all the "language classes" for your project (with
names "TkBocciBall::Localize::it", "TkBocciBall::Localize::en",
"TkBocciBall::Localize::fr", etc.).
A language class is a class containing a lexicon of phrases as class
data, and possibly also some methods that are of use in interpreting
phrases in the lexicon, or otherwise dealing with text in that
language.
An object belonging to a language class is called a "language handle";
it's typically a flyweight object.
The normal course of action is to call:
use TkBocciBall::Localize; # the localization project class
$lh = TkBocciBall::Localize->get_handle();
# Depending on the user's locale, etc., this will
# make a language handle from among the classes available,
# and any defaults that you declare.
die "Couldn't make a language handle??" unless $lh;
From then on, you use the "maketext" function to access entries in
whatever lexicon(s) belong to the language handle you got. So, this:
print $lh->maketext("You won!"), "\n";
...emits the right text for this language. If the object in $lh
belongs to class "TkBocciBall::Localize::fr" and
%TkBocciBall::Localize::fr::Lexicon contains "("You won!" => "Tu as
gagne!")", then the above code happily tells the user "Tu as gagne!".
METHODS
Locale::Maketext offers a variety of methods, which fall into three
categories:
o Methods to do with constructing language handles.
o "maketext" and other methods to do with accessing %Lexicon data for
a given language handle.
o Methods that you may find it handy to use, from routines of yours
that you put in %Lexicon entries.
These are covered in the following section.
Construction Methods
These are to do with constructing a language handle:
o $lh = YourProjClass->get_handle( ...langtags... ) || die "lg-
handle?";
This tries loading classes based on the language-tags you give
(like "("en-US", "sk", "kon", "es-MX", "ja", "i-klingon")", and for
the first class that succeeds, returns
YourProjClass::language->new().
If it runs thru the entire given list of language-tags, and finds
no classes for those exact terms, it then tries "superordinate"
language classes. So if no "en-US" class (i.e.,
YourProjClass::en_us) was found, nor classes for anything else in
that list, we then try its superordinate, "en" (i.e.,
YourProjClass::en), and so on thru the other language-tags in the
given list: "es". (The other language-tags in our example list:
happen to have no superordinates.)
If none of those language-tags leads to loadable classes, we then
try classes derived from YourProjClass->fallback_languages() and
then if nothing comes of that, we use classes named by
YourProjClass->fallback_language_classes(). Then in the (probably
quite unlikely) event that that fails, we just return undef.
o $lh = YourProjClass->get_handle() || die "lg-handle?";
When "get_handle" is called with an empty parameter list, magic
happens:
If "get_handle" senses that it's running in program that was
invoked as a CGI, then it tries to get language-tags out of the
environment variable "HTTP_ACCEPT_LANGUAGE", and it pretends that
those were the languages passed as parameters to "get_handle".
Otherwise (i.e., if not a CGI), this tries various OS-specific ways
to get the language-tags for the current locale/language, and then
pretends that those were the value(s) passed to "get_handle".
Currently this OS-specific stuff consists of looking in the
environment variables "LANG" and "LANGUAGE"; and on MSWin machines
(where those variables are typically unused), this also tries using
the module Win32::Locale to get a language-tag for whatever
language/locale is currently selected in the "Regional Settings"
(or "International"?) Control Panel. I welcome further
suggestions for making this do the Right Thing under other
operating systems that support localization.
If you're using localization in an application that keeps a
configuration file, you might consider something like this in your
project class:
sub get_handle_via_config {
my $class = $_[0];
my $chosen_language = $Config_settings{'language'};
my $lh;
if($chosen_language) {
$lh = $class->get_handle($chosen_language)
|| die "No language handle for \"$chosen_language\""
. " or the like";
} else {
# Config file missing, maybe?
$lh = $class->get_handle()
|| die "Can't get a language handle";
}
return $lh;
}
o $lh = YourProjClass::langname->new();
This constructs a language handle. You usually don't call this
directly, but instead let "get_handle" find a language class to
"use" and to then call ->new on.
o $lh->init();
This is called by ->new to initialize newly-constructed language
handles. If you define an init method in your class, remember that
it's usually considered a good idea to call $lh->SUPER::init in it
(presumably at the beginning), so that all classes get a chance to
initialize a new object however they see fit.
o YourProjClass->fallback_languages()
"get_handle" appends the return value of this to the end of
whatever list of languages you pass "get_handle". Unless you
override this method, your project class will inherit
Locale::Maketext's "fallback_languages", which currently returns
"('i-default', 'en', 'en-US')". ("i-default" is defined in RFC
2277).
This method (by having it return the name of a language-tag that
has an existing language class) can be used for making sure that
"get_handle" will always manage to construct a language handle
(assuming your language classes are in an appropriate @INC
directory). Or you can use the next method:
o YourProjClass->fallback_language_classes()
"get_handle" appends the return value of this to the end of the
list of classes it will try using. Unless you override this
method, your project class will inherit Locale::Maketext's
"fallback_language_classes", which currently returns an empty list,
"()". By setting this to some value (namely, the name of a
loadable language class), you can be sure that "get_handle" will
always manage to construct a language handle.
The "maketext" Method
This is the most important method in Locale::Maketext:
$text = $lh->maketext(I<key>, ...parameters for this phrase...);
This looks in the %Lexicon of the language handle $lh and all its
superclasses, looking for an entry whose key is the string key.
Assuming such an entry is found, various things then happen, depending
on the value found:
If the value is a scalarref, the scalar is dereferenced and returned
(and any parameters are ignored).
If the value is a coderef, we return &$value($lh, ...parameters...).
If the value is a string that doesn't look like it's in Bracket
Notation, we return it (after replacing it with a scalarref, in its
%Lexicon).
If the value does look like it's in Bracket Notation, then we compile
it into a sub, replace the string in the %Lexicon with the new coderef,
and then we return &$new_sub($lh, ...parameters...).
Bracket Notation is discussed in a later section. Note that trying to
compile a string into Bracket Notation can throw an exception if the
string is not syntactically valid (say, by not balancing brackets
right.)
Also, calling &$coderef($lh, ...parameters...) can throw any sort of
exception (if, say, code in that sub tries to divide by zero). But a
very common exception occurs when you have Bracket Notation text that
says to call a method "foo", but there is no such method. (E.g., "You
have [quatn,_1,ball]." will throw an exception on trying to call
$lh->quatn($_[1],'ball') -- you presumably meant "quant".) "maketext"
catches these exceptions, but only to make the error message more
readable, at which point it rethrows the exception.
An exception may be thrown if key is not found in any of $lh's %Lexicon
hashes. What happens if a key is not found, is discussed in a later
section, "Controlling Lookup Failure".
Note that you might find it useful in some cases to override the
"maketext" method with an "after method", if you want to translate
encodings, or even scripts:
package YrProj::zh_cn; # Chinese with PRC-style glyphs
use base ('YrProj::zh_tw'); # Taiwan-style
sub maketext {
my $self = shift(@_);
my $value = $self->maketext(@_);
return Chineeze::taiwan2mainland($value);
}
Or you may want to override it with something that traps any
exceptions, if that's critical to your program:
sub maketext {
my($lh, @stuff) = @_;
my $out;
eval { $out = $lh->SUPER::maketext(@stuff) };
return $out unless $@;
...otherwise deal with the exception...
}
Other than those two situations, I don't imagine that it's useful to
override the "maketext" method. (If you run into a situation where it
is useful, I'd be interested in hearing about it.)
$lh->fail_with or $lh->fail_with(PARAM)
$lh->failure_handler_auto
These two methods are discussed in the section "Controlling Lookup
Failure".
Utility Methods
These are methods that you may find it handy to use, generally from
%Lexicon routines of yours (whether expressed as Bracket Notation or
not).
$language->quant($number, $singular)
$language->quant($number, $singular, $plural)
$language->quant($number, $singular, $plural, $negative)
This is generally meant to be called from inside Bracket Notation
(which is discussed later), as in
"Your search matched [quant,_1,document]!"
It's for quantifying a noun (i.e., saying how much of it there is,
while giving the correct form of it). The behavior of this method
is handy for English and a few other Western European languages,
and you should override it for languages where it's not suitable.
You can feel free to read the source, but the current
implementation is basically as this pseudocode describes:
if $number is 0 and there's a $negative,
return $negative;
elsif $number is 1,
return "1 $singular";
elsif there's a $plural,
return "$number $plural";
else
return "$number " . $singular . "s";
#
# ...except that we actually call numf to
# stringify $number before returning it.
So for English (with Bracket Notation) "...[quant,_1,file]..." is
fine (for 0 it returns "0 files", for 1 it returns "1 file", and
for more it returns "2 files", etc.)
But for "directory", you'd want "[quant,_1,directory,directories]"
so that our elementary "quant" method doesn't think that the plural
of "directory" is "directorys". And you might find that the output
may sound better if you specify a negative form, as in:
"[quant,_1,file,files,No files] matched your query.\n"
Remember to keep in mind verb agreement (or adjectives too, in
other languages), as in:
"[quant,_1,document] were matched.\n"
Because if _1 is one, you get "1 document were matched". An
acceptable hack here is to do something like this:
"[quant,_1,document was, documents were] matched.\n"
$language->numf($number)
This returns the given number formatted nicely according to this
language's conventions. Maketext's default method is mostly to
just take the normal string form of the number (applying sprintf
"%G" for only very large numbers), and then to add commas as
necessary. (Except that we apply "tr/,./.,/" if
$language->{'numf_comma'} is true; that's a bit of a hack that's
useful for languages that express two million as "2.000.000" and
not as "2,000,000").
If you want anything fancier, consider overriding this with
something that uses Number::Format, or does something else
entirely.
Note that numf is called by quant for stringifying all quantifying
numbers.
$language->numerate($number, $singular, $plural, $negative)
This returns the given noun form which is appropriate for the
quantity $number according to this language's conventions.
"numerate" is used internally by "quant" to quantify nouns. Use it
directly -- usually from bracket notation -- to avoid "quant"'s
implicit call to "numf" and output of a numeric quantity.
$language->sprintf($format, @items)
This is just a wrapper around Perl's normal "sprintf" function.
It's provided so that you can use "sprintf" in Bracket Notation:
"Couldn't access datanode [sprintf,%10x=~[%s~],_1,_2]!\n"
returning...
Couldn't access datanode Stuff=[thangamabob]!
$language->language_tag()
Currently this just takes the last bit of "ref($language)", turns
underscores to dashes, and returns it. So if $language is an
object of class Hee::HOO::Haw::en_us, $language->language_tag()
returns "en-us". (Yes, the usual representation for that language
tag is "en-US", but case is never considered meaningful in
language-tag comparison.)
You may override this as you like; Maketext doesn't use it for
anything.
$language->encoding()
Currently this isn't used for anything, but it's provided (with
default value of "(ref($language) && $language->{'encoding'})) or
"iso-8859-1"" ) as a sort of suggestion that it may be
useful/necessary to associate encodings with your language handles
(whether on a per-class or even per-handle basis.)
Language Handle Attributes and Internals
A language handle is a flyweight object -- i.e., it doesn't
(necessarily) carry any data of interest, other than just being a
member of whatever class it belongs to.
A language handle is implemented as a blessed hash. Subclasses of
yours can store whatever data you want in the hash. Currently the only
hash entry used by any crucial Maketext method is "fail", so feel free
to use anything else as you like.
Remember: Don't be afraid to read the Maketext source if there's any
point on which this documentation is unclear. This documentation is
vastly longer than the module source itself.
LANGUAGE CLASS HIERARCHIES
These are Locale::Maketext's assumptions about the class hierarchy
formed by all your language classes:
o You must have a project base class, which you load, and which you
then use as the first argument in the call to
YourProjClass->get_handle(...). It should derive (whether directly
or indirectly) from Locale::Maketext. It doesn't matter how you
name this class, although assuming this is the localization
component of your Super Mega Program, good names for your project
class might be SuperMegaProgram::Localization,
SuperMegaProgram::L10N, SuperMegaProgram::I18N,
SuperMegaProgram::International, or even
SuperMegaProgram::Languages or SuperMegaProgram::Messages.
o Language classes are what YourProjClass->get_handle will try to
load. It will look for them by taking each language-tag (skipping
it if it doesn't look like a language-tag or locale-tag!), turning
it to all lowercase, turning dashes to underscores, and appending
it to YourProjClass . "::". So this:
$lh = YourProjClass->get_handle(
'en-US', 'fr', 'kon', 'i-klingon', 'i-klingon-romanized'
);
will try loading the classes YourProjClass::en_us (note
lowercase!), YourProjClass::fr, YourProjClass::kon,
YourProjClass::i_klingon and YourProjClass::i_klingon_romanized.
(And it'll stop at the first one that actually loads.)
o I assume that each language class derives (directly or indirectly)
from your project class, and also defines its @ISA, its %Lexicon,
or both. But I anticipate no dire consequences if these
assumptions do not hold.
o Language classes may derive from other language classes (although
they should have "use Thatclassname" or "use base
qw(...classes...)"). They may derive from the project class. They
may derive from some other class altogether. Or via multiple
inheritance, it may derive from any mixture of these.
o I foresee no problems with having multiple inheritance in your
hierarchy of language classes. (As usual, however, Perl will
complain bitterly if you have a cycle in the hierarchy: i.e., if
any class is its own ancestor.)
ENTRIES IN EACH LEXICON
A typical %Lexicon entry is meant to signify a phrase, taking some
number (0 or more) of parameters. An entry is meant to be accessed by
via a string key in $lh->maketext(key, ...parameters...), which should
return a string that is generally meant for be used for "output" to the
user -- regardless of whether this actually means printing to STDOUT,
writing to a file, or putting into a GUI widget.
While the key must be a string value (since that's a basic restriction
that Perl places on hash keys), the value in the lexicon can currently
be of several types: a defined scalar, scalarref, or coderef. The use
of these is explained above, in the section 'The "maketext" Method',
and Bracket Notation for strings is discussed in the next section.
While you can use arbitrary unique IDs for lexicon keys (like
"_min_larger_max_error"), it is often useful for if an entry's key is
itself a valid value, like this example error message:
"Minimum ([_1]) is larger than maximum ([_2])!\n",
Compare this code that uses an arbitrary ID...
die $lh->maketext( "_min_larger_max_error", $min, $max )
if $min > $max;
...to this code that uses a key-as-value:
die $lh->maketext(
"Minimum ([_1]) is larger than maximum ([_2])!\n",
$min, $max
) if $min > $max;
The second is, in short, more readable. In particular, it's obvious
that the number of parameters you're feeding to that phrase (two) is
the number of parameters that it wants to be fed. (Since you see _1
and a _2 being used in the key there.)
Also, once a project is otherwise complete and you start to localize
it, you can scrape together all the various keys you use, and pass it
to a translator; and then the translator's work will go faster if what
he's presented is this:
"Minimum ([_1]) is larger than maximum ([_2])!\n",
=> "", # fill in something here, Jacques!
rather than this more cryptic mess:
"_min_larger_max_error"
=> "", # fill in something here, Jacques
I think that keys as lexicon values makes the completed lexicon entries
more readable:
"Minimum ([_1]) is larger than maximum ([_2])!\n",
=> "Le minimum ([_1]) est plus grand que le maximum ([_2])!\n",
Also, having valid values as keys becomes very useful if you set up an
_AUTO lexicon. _AUTO lexicons are discussed in a later section.
I almost always use keys that are themselves valid lexicon values. One
notable exception is when the value is quite long. For example, to get
the screenful of data that a command-line program might return when
given an unknown switch, I often just use a brief, self-explanatory key
such as "_USAGE_MESSAGE". At that point I then go and immediately to
define that lexicon entry in the ProjectClass::L10N::en lexicon (since
English is always my "project language"):
'_USAGE_MESSAGE' => <<'EOSTUFF',
...long long message...
EOSTUFF
and then I can use it as:
getopt('oDI', \%opts) or die $lh->maketext('_USAGE_MESSAGE');
Incidentally, note that each class's %Lexicon inherits-and-extends the
lexicons in its superclasses. This is not because these are special
hashes per se, but because you access them via the "maketext" method,
which looks for entries across all the %Lexicon hashes in a language
class and all its ancestor classes. (This is because the idea of
"class data" isn't directly implemented in Perl, but is instead left to
individual class-systems to implement as they see fit..)
Note that you may have things stored in a lexicon besides just phrases
for output: for example, if your program takes input from the
keyboard, asking a "(Y/N)" question, you probably need to know what the
equivalent of "Y[es]/N[o]" is in whatever language. You probably also
need to know what the equivalents of the answers "y" and "n" are. You
can store that information in the lexicon (say, under the keys
"~answer_y" and "~answer_n", and the long forms as "~answer_yes" and
"~answer_no", where "~" is just an ad-hoc character meant to indicate
to programmers/translators that these are not phrases for output).
Or instead of storing this in the language class's lexicon, you can
(and, in some cases, really should) represent the same bit of knowledge
as code in a method in the language class. (That leaves a tidy
distinction between the lexicon as the things we know how to say, and
the rest of the things in the lexicon class as things that we know how
to do.) Consider this example of a processor for responses to French
"oui/non" questions:
sub y_or_n {
return undef unless defined $_[1] and length $_[1];
my $answer = lc $_[1]; # smash case
return 1 if $answer eq 'o' or $answer eq 'oui';
return 0 if $answer eq 'n' or $answer eq 'non';
return undef;
}
...which you'd then call in a construct like this:
my $response;
until(defined $response) {
print $lh->maketext("Open the pod bay door (y/n)? ");
$response = $lh->y_or_n( get_input_from_keyboard_somehow() );
}
if($response) { $pod_bay_door->open() }
else { $pod_bay_door->leave_closed() }
Other data worth storing in a lexicon might be things like filenames
for language-targetted resources:
...
"_main_splash_png"
=> "/styles/en_us/main_splash.png",
"_main_splash_imagemap"
=> "/styles/en_us/main_splash.incl",
"_general_graphics_path"
=> "/styles/en_us/",
"_alert_sound"
=> "/styles/en_us/hey_there.wav",
"_forward_icon"
=> "left_arrow.png",
"_backward_icon"
=> "right_arrow.png",
# In some other languages, left equals
# BACKwards, and right is FOREwards.
...
You might want to do the same thing for expressing key bindings or the
like (since hardwiring "q" as the binding for the function that quits a
screen/menu/program is useful only if your language happens to
associate "q" with "quit"!)
BRACKET NOTATION
Bracket Notation is a crucial feature of Locale::Maketext. I mean
Bracket Notation to provide a replacement for the use of sprintf
formatting. Everything you do with Bracket Notation could be done with
a sub block, but bracket notation is meant to be much more concise.
Bracket Notation is a like a miniature "template" system (in the sense
of Text::Template, not in the sense of C++ templates), where normal
text is passed thru basically as is, but text in special regions is
specially interpreted. In Bracket Notation, you use square brackets
("[...]"), not curly braces ("{...}") to note sections that are
specially interpreted.
For example, here all the areas that are taken literally are underlined
with a "^", and all the in-bracket special regions are underlined with
an X:
"Minimum ([_1]) is larger than maximum ([_2])!\n",
^^^^^^^^^ XX ^^^^^^^^^^^^^^^^^^^^^^^^^^ XX ^^^^
When that string is compiled from bracket notation into a real Perl
sub, it's basically turned into:
sub {
my $lh = $_[0];
my @params = @_;
return join '',
"Minimum (",
...some code here...
") is larger than maximum (",
...some code here...
")!\n",
}
# to be called by $lh->maketext(KEY, params...)
In other words, text outside bracket groups is turned into string
literals. Text in brackets is rather more complex, and currently
follows these rules:
o Bracket groups that are empty, or which consist only of whitespace,
are ignored. (Examples: "[]", "[ ]", or a [ and a ] with
returns and/or tabs and/or spaces between them.
Otherwise, each group is taken to be a comma-separated group of
items, and each item is interpreted as follows:
o An item that is "_digits" or "_-digits" is interpreted as
$_[value]. I.e., "_1" becomes with $_[1], and "_-3" is interpreted
as $_[-3] (in which case @_ should have at least three elements in
it). Note that $_[0] is the language handle, and is typically not
named directly.
o An item "_*" is interpreted to mean "all of @_ except $_[0]".
I.e., @_[1..$#_]. Note that this is an empty list in the case of
calls like $lh->maketext(key) where there are no parameters (except
$_[0], the language handle).
o Otherwise, each item is interpreted as a string literal.
The group as a whole is interpreted as follows:
o If the first item in a bracket group looks like a method name, then
that group is interpreted like this:
$lh->that_method_name(
...rest of items in this group...
),
o If the first item in a bracket group is "*", it's taken as
shorthand for the so commonly called "quant" method. Similarly, if
the first item in a bracket group is "#", it's taken to be
shorthand for "numf".
o If the first item in a bracket group is the empty-string, or "_*"
or "_digits" or "_-digits", then that group is interpreted as just
the interpolation of all its items:
join('',
...rest of items in this group...
),
Examples: "[_1]" and "[,_1]", which are synonymous; and
""[,ID-(,_4,-,_2,)]"", which compiles as "join "", "ID-(", $_[4],
"-", $_[2], ")"".
o Otherwise this bracket group is invalid. For example, in the group
"[!@#,whatever]", the first item "!@#" is neither the empty-string,
"_number", "_-number", "_*", nor a valid method name; and so
Locale::Maketext will throw an exception of you try compiling an
expression containing this bracket group.
Note, incidentally, that items in each group are comma-separated, not
"/\s*,\s*/"-separated. That is, you might expect that this bracket
group:
"Hoohah [foo, _1 , bar ,baz]!"
would compile to this:
sub {
my $lh = $_[0];
return join '',
"Hoohah ",
$lh->foo( $_[1], "bar", "baz"),
"!",
}
But it actually compiles as this:
sub {
my $lh = $_[0];
return join '',
"Hoohah ",
$lh->foo(" _1 ", " bar ", "baz"), # note the <space> in " bar "
"!",
}
In the notation discussed so far, the characters "[" and "]" are given
special meaning, for opening and closing bracket groups, and "," has a
special meaning inside bracket groups, where it separates items in the
group. This begs the question of how you'd express a literal "[" or
"]" in a Bracket Notation string, and how you'd express a literal comma
inside a bracket group. For this purpose I've adopted "~" (tilde) as
an escape character: "~[" means a literal '[' character anywhere in
Bracket Notation (i.e., regardless of whether you're in a bracket group
or not), and ditto for "~]" meaning a literal ']', and "~," meaning a
literal comma. (Altho "," means a literal comma outside of bracket
groups -- it's only inside bracket groups that commas are special.)
And on the off chance you need a literal tilde in a bracket expression,
you get it with "~~".
Currently, an unescaped "~" before a character other than a bracket or
a comma is taken to mean just a "~" and that character. I.e., "~X"
means the same as "~~X" -- i.e., one literal tilde, and then one
literal "X". However, by using "~X", you are assuming that no future
version of Maketext will use "~X" as a magic escape sequence. In
practice this is not a great problem, since first off you can just
write "~~X" and not worry about it; second off, I doubt I'll add lots
of new magic characters to bracket notation; and third off, you aren't
likely to want literal "~" characters in your messages anyway, since
it's not a character with wide use in natural language text.
Brackets must be balanced -- every openbracket must have one matching
closebracket, and vice versa. So these are all invalid:
"I ate [quant,_1,rhubarb pie."
"I ate [quant,_1,rhubarb pie[."
"I ate quant,_1,rhubarb pie]."
"I ate quant,_1,rhubarb pie[."
Currently, bracket groups do not nest. That is, you cannot say:
"Foo [bar,baz,[quux,quuux]]\n";
If you need a notation that's that powerful, use normal Perl:
%Lexicon = (
...
"some_key" => sub {
my $lh = $_[0];
join '',
"Foo ",
$lh->bar('baz', $lh->quux('quuux')),
"\n",
},
...
);
Or write the "bar" method so you don't need to pass it the output from
calling quux.
I do not anticipate that you will need (or particularly want) to nest
bracket groups, but you are welcome to email me with convincing (real-
life) arguments to the contrary.
AUTO LEXICONS
If maketext goes to look in an individual %Lexicon for an entry for key
(where key does not start with an underscore), and sees none, but does
see an entry of "_AUTO" => some_true_value, then we actually define
$Lexicon{key} = key right then and there, and then use that value as if
it had been there all along. This happens before we even look in any
superclass %Lexicons!
(This is meant to be somewhat like the AUTOLOAD mechanism in Perl's
function call system -- or, looked at another way, like the AutoLoader
module.)
I can picture all sorts of circumstances where you just do not want
lookup to be able to fail (since failing normally means that maketext
throws a "die", although see the next section for greater control over
that). But here's one circumstance where _AUTO lexicons are meant to
be especially useful:
As you're writing an application, you decide as you go what messages
you need to emit. Normally you'd go to write this:
if(-e $filename) {
go_process_file($filename)
} else {
print qq{Couldn't find file "$filename"!\n};
}
but since you anticipate localizing this, you write:
use ThisProject::I18N;
my $lh = ThisProject::I18N->get_handle();
# For the moment, assume that things are set up so
# that we load class ThisProject::I18N::en
# and that that's the class that $lh belongs to.
...
if(-e $filename) {
go_process_file($filename)
} else {
print $lh->maketext(
qq{Couldn't find file "[_1]"!\n}, $filename
);
}
Now, right after you've just written the above lines, you'd normally
have to go open the file ThisProject/I18N/en.pm, and immediately add an
entry:
"Couldn't find file \"[_1]\"!\n"
=> "Couldn't find file \"[_1]\"!\n",
But I consider that somewhat of a distraction from the work of getting
the main code working -- to say nothing of the fact that I often have
to play with the program a few times before I can decide exactly what
wording I want in the messages (which in this case would require me to
go changing three lines of code: the call to maketext with that key,
and then the two lines in ThisProject/I18N/en.pm).
However, if you set "_AUTO => 1" in the %Lexicon in,
ThisProject/I18N/en.pm (assuming that English (en) is the language that
all your programmers will be using for this project's internal message
keys), then you don't ever have to go adding lines like this
"Couldn't find file \"[_1]\"!\n"
=> "Couldn't find file \"[_1]\"!\n",
to ThisProject/I18N/en.pm, because if _AUTO is true there, then just
looking for an entry with the key "Couldn't find file \"[_1]\"!\n" in
that lexicon will cause it to be added, with that value!
Note that the reason that keys that start with "_" are immune to _AUTO
isn't anything generally magical about the underscore character -- I
just wanted a way to have most lexicon keys be autoable, except for
possibly a few, and I arbitrarily decided to use a leading underscore
as a signal to distinguish those few.
READONLY LEXICONS
If your lexicon is a tied hash the simple act of caching the compiled
value can be fatal.
For example a GDBM_File GDBM_READER tied hash will die with something
like:
gdbm store returned -1, errno 2, key "..." at ...
All you need to do is turn on caching outside of the lexicon hash
itself like so:
sub init {
my ($lh) = @_;
...
$lh->{'use_external_lex_cache'} = 1;
...
}
And then instead of storing the compiled value in the lexicon hash it
will store it in $lh->{'_external_lex_cache'}
CONTROLLING LOOKUP FAILURE
If you call $lh->maketext(key, ...parameters...), and there's no entry
key in $lh's class's %Lexicon, nor in the superclass %Lexicon hash, and
if we can't auto-make key (because either it starts with a "_", or
because none of its lexicons have "_AUTO => 1,"), then we have failed
to find a normal way to maketext key. What then happens in these
failure conditions, depends on the $lh object's "fail" attribute.
If the language handle has no "fail" attribute, maketext will simply
throw an exception (i.e., it calls "die", mentioning the key whose
lookup failed, and naming the line number where the calling
$lh->maketext(key,...) was.
If the language handle has a "fail" attribute whose value is a coderef,
then $lh->maketext(key,...params...) gives up and calls:
return $that_subref->($lh, $key, @params);
Otherwise, the "fail" attribute's value should be a string denoting a
method name, so that $lh->maketext(key,...params...) can give up with:
return $lh->$that_method_name($phrase, @params);
The "fail" attribute can be accessed with the "fail_with" method:
# Set to a coderef:
$lh->fail_with( \&failure_handler );
# Set to a method name:
$lh->fail_with( 'failure_method' );
# Set to nothing (i.e., so failure throws a plain exception)
$lh->fail_with( undef );
# Get the current value
$handler = $lh->fail_with();
Now, as to what you may want to do with these handlers: Maybe you'd
want to log what key failed for what class, and then die. Maybe you
don't like "die" and instead you want to send the error message to
STDOUT (or wherever) and then merely "exit()".
Or maybe you don't want to "die" at all! Maybe you could use a handler
like this:
# Make all lookups fall back onto an English value,
# but only after we log it for later fingerpointing.
my $lh_backup = ThisProject->get_handle('en');
open(LEX_FAIL_LOG, ">>wherever/lex.log") || die "GNAARGH $!";
sub lex_fail {
my($failing_lh, $key, $params) = @_;
print LEX_FAIL_LOG scalar(localtime), "\t",
ref($failing_lh), "\t", $key, "\n";
return $lh_backup->maketext($key,@params);
}
Some users have expressed that they think this whole mechanism of
having a "fail" attribute at all, seems a rather pointless
complication. But I want Locale::Maketext to be usable for software
projects of any scale and type; and different software projects have
different ideas of what the right thing is to do in failure conditions.
I could simply say that failure always throws an exception, and that if
you want to be careful, you'll just have to wrap every call to
$lh->maketext in an eval { }. However, I want programmers to reserve
the right (via the "fail" attribute) to treat lookup failure as
something other than an exception of the same level of severity as a
config file being unreadable, or some essential resource being
inaccessible.
One possibly useful value for the "fail" attribute is the method name
"failure_handler_auto". This is a method defined in the class
Locale::Maketext itself. You set it with:
$lh->fail_with('failure_handler_auto');
Then when you call $lh->maketext(key, ...parameters...) and there's no
key in any of those lexicons, maketext gives up with
return $lh->failure_handler_auto($key, @params);
But failure_handler_auto, instead of dying or anything, compiles $key,
caching it in
$lh->{'failure_lex'}{$key} = $complied
and then calls the compiled value, and returns that. (I.e., if $key
looks like bracket notation, $compiled is a sub, and we return
&{$compiled}(@params); but if $key is just a plain string, we just
return that.)
The effect of using "failure_auto_handler" is like an AUTO lexicon,
except that it 1) compiles $key even if it starts with "_", and 2) you
have a record in the new hashref $lh->{'failure_lex'} of all the keys
that have failed for this object. This should avoid your program dying
-- as long as your keys aren't actually invalid as bracket code, and as
long as they don't try calling methods that don't exist.
"failure_auto_handler" may not be exactly what you want, but I hope it
at least shows you that maketext failure can be mitigated in any number
of very flexible ways. If you can formalize exactly what you want, you
should be able to express that as a failure handler. You can even make
it default for every object of a given class, by setting it in that
class's init:
sub init {
my $lh = $_[0]; # a newborn handle
$lh->SUPER::init();
$lh->fail_with('my_clever_failure_handler');
return;
}
sub my_clever_failure_handler {
...you clever things here...
}
HOW TO USE MAKETEXT
Here is a brief checklist on how to use Maketext to localize
applications:
o Decide what system you'll use for lexicon keys. If you insist, you
can use opaque IDs (if you're nostalgic for "catgets"), but I have
better suggestions in the section "Entries in Each Lexicon", above.
Assuming you opt for meaningful keys that double as values (like
"Minimum ([_1]) is larger than maximum ([_2])!\n"), you'll have to
settle on what language those should be in. For the sake of
argument, I'll call this English, specifically American English,
"en-US".
o Create a class for your localization project. This is the name of
the class that you'll use in the idiom:
use Projname::L10N;
my $lh = Projname::L10N->get_handle(...) || die "Language?";
Assuming you call your class Projname::L10N, create a class
consisting minimally of:
package Projname::L10N;
use base qw(Locale::Maketext);
...any methods you might want all your languages to share...
# And, assuming you want the base class to be an _AUTO lexicon,
# as is discussed a few sections up:
1;
o Create a class for the language your internal keys are in. Name
the class after the language-tag for that language, in lowercase,
with dashes changed to underscores. Assuming your project's first
language is US English, you should call this Projname::L10N::en_us.
It should consist minimally of:
package Projname::L10N::en_us;
use base qw(Projname::L10N);
%Lexicon = (
'_AUTO' => 1,
);
1;
(For the rest of this section, I'll assume that this "first
language class" of Projname::L10N::en_us has _AUTO lexicon.)
o Go and write your program. Everywhere in your program where you
would say:
print "Foobar $thing stuff\n";
instead do it thru maketext, using no variable interpolation in the
key:
print $lh->maketext("Foobar [_1] stuff\n", $thing);
If you get tired of constantly saying "print $lh->maketext",
consider making a functional wrapper for it, like so:
use Projname::L10N;
use vars qw($lh);
$lh = Projname::L10N->get_handle(...) || die "Language?";
sub pmt (@) { print( $lh->maketext(@_)) }
# "pmt" is short for "Print MakeText"
$Carp::Verbose = 1;
# so if maketext fails, we see made the call to pmt
Besides whole phrases meant for output, anything language-dependent
should be put into the class Projname::L10N::en_us, whether as
methods, or as lexicon entries -- this is discussed in the section
"Entries in Each Lexicon", above.
o Once the program is otherwise done, and once its localization for
the first language works right (via the data and methods in
Projname::L10N::en_us), you can get together the data for
translation. If your first language lexicon isn't an _AUTO
lexicon, then you already have all the messages explicitly in the
lexicon (or else you'd be getting exceptions thrown when you call
$lh->maketext to get messages that aren't in there). But if you
were (advisedly) lazy and are using an _AUTO lexicon, then you've
got to make a list of all the phrases that you've so far been
letting _AUTO generate for you. There are very many ways to
assemble such a list. The most straightforward is to simply grep
the source for every occurrence of "maketext" (or calls to wrappers
around it, like the above "pmt" function), and to log the following
phrase.
o You may at this point want to consider whether your base class
(Projname::L10N), from which all lexicons inherit from
(Projname::L10N::en, Projname::L10N::es, etc.), should be an _AUTO
lexicon. It may be true that in theory, all needed messages will
be in each language class; but in the presumably unlikely or
"impossible" case of lookup failure, you should consider whether
your program should throw an exception, emit text in English (or
whatever your project's first language is), or some more complex
solution as described in the section "Controlling Lookup Failure",
above.
o Submit all messages/phrases/etc. to translators.
(You may, in fact, want to start with localizing to one other
language at first, if you're not sure that you've properly
abstracted the language-dependent parts of your code.)
Translators may request clarification of the situation in which a
particular phrase is found. For example, in English we are
entirely happy saying "n files found", regardless of whether we
mean "I looked for files, and found n of them" or the rather
distinct situation of "I looked for something else (like lines in
files), and along the way I saw n files." This may involve
rethinking things that you thought quite clear: should "Edit" on a
toolbar be a noun ("editing") or a verb ("to edit")? Is there
already a conventionalized way to express that menu option,
separate from the target language's normal word for "to edit"?
In all cases where the very common phenomenon of quantification
(saying "N files", for any value of N) is involved, each translator
should make clear what dependencies the number causes in the
sentence. In many cases, dependency is limited to words adjacent
to the number, in places where you might expect them ("I found
the-?PLURAL N empty-?PLURAL directory-?PLURAL"), but in some cases
there are unexpected dependencies ("I found-?PLURAL ..."!) as well
as long-distance dependencies "The N directory-?PLURAL could not be
deleted-?PLURAL"!).
Remind the translators to consider the case where N is 0: "0 files
found" isn't exactly natural-sounding in any language, but it may
be unacceptable in many -- or it may condition special kinds of
agreement (similar to English "I didN'T find ANY files").
Remember to ask your translators about numeral formatting in their
language, so that you can override the "numf" method as
appropriate. Typical variables in number formatting are: what to
use as a decimal point (comma? period?); what to use as a thousands
separator (space? nonbreaking space? comma? period? small middot?
prime? apostrophe?); and even whether the so-called "thousands
separator" is actually for every third digit -- I've heard reports
of two hundred thousand being expressible as "2,00,000" for some
Indian (Subcontinental) languages, besides the less surprising
"200 000", "200.000", "200,000", and "200'000". Also, using a set
of numeral glyphs other than the usual ASCII "0"-"9" might be
appreciated, as via "tr/0-9/\x{0966}-\x{096F}/" for getting digits
in Devanagari script (for Hindi, Konkani, others).
The basic "quant" method that Locale::Maketext provides should be
good for many languages. For some languages, it might be useful to
modify it (or its constituent "numerate" method) to take a plural
form in the two-argument call to "quant" (as in "[quant,_1,files]")
if it's all-around easier to infer the singular form from the
plural, than to infer the plural form from the singular.
But for other languages (as is discussed at length in
Locale::Maketext::TPJ13), simple "quant"/"numf" is not enough. For
the particularly problematic Slavic languages, what you may need is
a method which you provide with the number, the citation form of
the noun to quantify, and the case and gender that the sentence's
syntax projects onto that noun slot. The method would then be
responsible for determining what grammatical number that numeral
projects onto its noun phrase, and what case and gender it may
override the normal case and gender with; and then it would look up
the noun in a lexicon providing all needed inflected forms.
o You may also wish to discuss with the translators the question of
how to relate different subforms of the same language tag,
considering how this reacts with "get_handle"'s treatment of these.
For example, if a user accepts interfaces in "en, fr", and you have
interfaces available in "en-US" and "fr", what should they get?
You may wish to resolve this by establishing that "en" and "en-US"
are effectively synonymous, by having one class zero-derive from
the other.
For some languages this issue may never come up (Danish is rarely
expressed as "da-DK", but instead is just "da"). And for other
languages, the whole concept of a "generic" form may verge on being
uselessly vague, particularly for interfaces involving voice media
in forms of Arabic or Chinese.
o Once you've localized your program/site/etc. for all desired
languages, be sure to show the result (whether live, or via
screenshots) to the translators. Once they approve, make every
effort to have it then checked by at least one other speaker of
that language. This holds true even when (or especially when) the
translation is done by one of your own programmers. Some kinds of
systems may be harder to find testers for than others, depending on
the amount of domain-specific jargon and concepts involved -- it's
easier to find people who can tell you whether they approve of your
translation for "delete this message" in an email-via-Web
interface, than to find people who can give you an informed opinion
on your translation for "attribute value" in an XML query tool's
interface.
SEE ALSO
I recommend reading all of these:
Locale::Maketext::TPJ13 -- my The Perl Journal article about Maketext.
It explains many important concepts underlying Locale::Maketext's
design, and some insight into why Maketext is better than the plain old
approach of having message catalogs that are just databases of sprintf
formats.
File::Findgrep is a sample application/module that uses
Locale::Maketext to localize its messages. For a larger
internationalized system, see also Apache::MP3.
I18N::LangTags.
Win32::Locale.
RFC 3066, Tags for the Identification of Languages, as at
http://sunsite.dk/RFC/rfc/rfc3066.html
RFC 2277, IETF Policy on Character Sets and Languages is at
http://sunsite.dk/RFC/rfc/rfc2277.html -- much of it is just things of
interest to protocol designers, but it explains some basic concepts,
like the distinction between locales and language-tags.
The manual for GNU "gettext". The gettext dist is available in
"ftp://prep.ai.mit.edu/pub/gnu/" -- get a recent gettext tarball and
look in its "doc/" directory, there's an easily browsable HTML version
in there. The gettext documentation asks lots of questions worth
thinking about, even if some of their answers are sometimes wonky,
particularly where they start talking about pluralization.
The Locale/Maketext.pm source. Obverse that the module is much shorter
than its documentation!
COPYRIGHT AND DISCLAIMER
Copyright (c) 1999-2004 Sean M. Burke. All rights reserved.
This library is free software; you can redistribute it and/or modify it
under the same terms as Perl itself.
This program is distributed in the hope that it will be useful, but
without any warranty; without even the implied warranty of
merchantability or fitness for a particular purpose.
AUTHOR
Sean M. Burke "sburke AT cpan.org"
perl v5.16.3 2012-11-27 Locale::Maketext(3)