This manual is still in DRAFT state. Some sections are still empty, or almost. We keep merging material from other sources (essentially e-mail folders) while the proper integration of this material is delayed.
In this manual, we use he when speaking of the programmer or
maintainer, she when speaking of the translator, and they
when speaking of the installers or end users of the translated program.
This is only a convenience for clarifying the documentation. It is
absolutely not meant to imply that some roles are more appropriate
to males or females. Besides, as you might guess, GNU gettext
is meant to be useful for people using computers, whatever their sex,
race, religion or nationality!
This chapter explains the goals sought in the creation
of GNU gettext and the free Translation Project.
Then, it explains a few broad concepts around
Native Language Support, and positions message translation with regard
to other aspects of national and cultural variance, as they apply to
to programs. It also surveys those files used to convey the
translations. It explains how the various tools interact in the
initial generation of these files, and later, how the maintenance
cycle should usually operate.
Please send suggestions and corrections to:
Internet address:
[email protected]
Please include the manual's edition number and update date in your messages.
gettextUsually, programs are written and documented in English, and use English at execution time to interact with users. This is true not only of GNU software, but also of a great deal of commercial and free software. Using a common language is quite handy for communication between developers, maintainers and users from all countries. On the other hand, most people are less comfortable with English than with their own native language, and would prefer to use their mother tongue for day to day's work, as far as possible. Many would simply love to see their computer screen showing a lot less of English, and far more of their own language.
However, to many people, this dream might appear so far fetched that they may believe it is not even worth spending time thinking about it. They have no confidence at all that the dream might ever become true. Yet some have not lost hope, and have organized themselves. The Translation Project is a formalization of this hope into a workable structure, which has a good chance to get all of us nearer the achievement of a truly multi-lingual set of programs.
GNU gettext is an important step for the Translation Project,
as it is an asset on which we may build many other steps. This package
offers to programmers, translators and even users, a well integrated
set of tools and documentation. Specifically, the GNU gettext
utilities are a set of tools that provides a framework within which
other free packages may produce multi-lingual messages. These tools
include a set of conventions about how programs should be written to
support message catalogs, a directory and file naming organization for the
message catalogs themselves, a runtime library supporting the retrieval of
translated messages, and a few stand-alone programs to massage in various
ways the sets of translatable strings, or already translated strings.
A special mode for GNU Emacs also helps ease interested parties into
preparing these sets, or bringing them up to date.
GNU gettext is designed to minimize the impact of
internationalization on program sources, keeping this impact as small
and hardly noticeable as possible. Internationalization has better
chances of succeeding if it is very light weighted, or at least,
appear to be so, when looking at program sources.
The Translation Project also uses the GNU gettext
distribution as a vehicle for documenting its structure and methods.
This goes beyond the strict technicalities of documenting the GNU gettext
proper. By so doing, translators will find in a single place, as
far as possible, all they need to know for properly doing their
translating work. Also, this supplemental documentation might also
help programmers, and even curious users, in understanding how GNU
gettext is related to the remainder of the Translation
Project, and consequently, have a glimpse at the big picture.
Two long words appear all the time when we discuss support of native language in programs, and these words have a precise meaning, worth being explained here, once and for all in this document. The words are internationalization and localization. Many people, tired of writing these long words over and over again, took the habit of writing i18n and l10n instead, quoting the first and last letter of each word, and replacing the run of intermediate letters by a number merely telling how many such letters there are. But in this manual, in the sake of clarity, we will patiently write the names in full, each time...
By internationalization, one refers to the operation by which a
program, or a set of programs turned into a package, is made aware of and
able to support multiple languages. This is a generalization process,
by which the programs are untied from calling only English strings or
other English specific habits, and connected to generic ways of doing
the same, instead. Program developers may use various techniques to
internationalize their programs. Some of these have been standardized.
GNU gettext offers one of these standards. See section The Programmer's View.
By localization, one means the operation by which, in a set of programs already internationalized, one gives the program all needed information so that it can adapt itself to handle its input and output in a fashion which is correct for some native language and cultural habits. This is a particularisation process, by which generic methods already implemented in an internationalized program are used in specific ways. The programming environment puts several functions to the programmers disposal which allow this runtime configuration. The formal description of specific set of cultural habits for some country, together with all associated translations targeted to the same native language, is called the locale for this language or country. Users achieve localization of programs by setting proper values to special environment variables, prior to executing those programs, identifying which locale should be used.
In fact, locale message support is only one component of the cultural data that makes up a particular locale. There are a whole host of routines and functions provided to aid programmers in developing internationalized software and which allow them to access the data stored in a particular locale. When someone presently refers to a particular locale, they are obviously referring to the data stored within that particular locale. Similarly, if a programmer is referring to "accessing the locale routines", they are referring to the complete suite of routines that access all of the locale's information.
One uses the expression Native Language Support, or merely NLS, for speaking of the overall activity or feature encompassing both internationalization and localization, allowing for multi-lingual interactions in a program. In a nutshell, one could say that internationalization is the operation by which further localizations are made possible.
Also, very roughly said, when it comes to multi-lingual messages, internationalization is usually taken care of by programmers, and localization is usually taken care of by translators.
For a totally multi-lingual distribution, there are many things to translate beyond output messages.
gettext offers a complete toolset for
translating messages output by C programs. Perl scripts and shell
scripts will also need to be translated. Even if there are today some hooks
by which this can be done, these hooks are not integrated as well as they
should be.
autoconf or bison, are able
to produce other programs (or scripts). Even if the generating
programs themselves are internationalized, the generated programs they
produce may need internationalization on their own, and this indirect
internationalization could be automated right from the generating
program. In fact, quite usually, generating and generated programs
could be internationalized independently, as the effort needed is
fairly orthogonal.
recode is able to reconstruct at execution.
Since these descriptions are extracted from the RFC by mechanical means,
translating them properly would require a prior translation of the RFC
itself.
gcc to allow diacriticized characters in identifiers or use
translated keywords; `rm -i' might accept something else than
`y' or `n' for replies, etc. Even if the program will
eventually make most of its output in the foreign languages, one has
to decide whether the input syntax, option values, etc., are to be
localized or not.
As we already stressed, translation is only one aspect of locales.
Other internationalization aspects are not currently handled by GNU
gettext, but perhaps may be handled in future versions. There
are many attributes that are needed to define a country's cultural
conventions. These attributes include beside the country's native
language, the formatting of the date and time, the representation of
numbers, the symbols for currency, etc. These local rules are
termed the country's locale. The locale represents the knowledge
needed to support the country's native attributes.
There are a few major areas which may vary between countries and
hence, define what a locale must describe. The following list helps
putting multi-lingual messages into the proper context of other tasks
related to locales, and also presents some other areas which GNU
gettext might eventually tackle, maybe, one of these days.
12,345.67 English 12.345,67 French 1,2345.67 AsiaSome programs could go further and use different unit systems, like English units or Metric units, or even take into account variants about how numbers are spelled in full.
gettext provides the means for developers and users to
easily change the language that the software uses to communicate to
the user.
In the near future we see no chance that components of locale outside of
message handling will be made available for use in other
packages. The reason for this is that most modern systems provide
a more or less reasonable support for at least some of the missing
components. Another point is that the GNU libc and Linux will get
a new and complete implementation of the whole locale functionality
which could be adopted by system lacking a reasonable locale support.
The letters PO in `.po' files means Portable Object, to distinguish it from `.mo' files, where MO stands for Machine Object. This paradigm, as well as the PO file format, is inspired by the NLS standard developed by Uniforum, and implemented by Sun in their Solaris system.
PO files are meant to be read and edited by humans, and associate each
original, translatable string of a given package with its translation
in a particular target language. A single PO file is dedicated to
a single target language. If a package supports many languages,
there is one such PO file per language supported, and each package
has its own set of PO files. These PO files are best created by
the xgettext program, and later updated or refreshed through
the msgmerge program. Program xgettext extracts all
marked messages from a set of C files and initializes a PO file with
empty translations. Program msgmerge takes care of adjusting
PO files between releases of the corresponding sources, commenting
obsolete entries, initializing new ones, and updating all source
line references. Files ending with `.pot' are kind of base
translation files found in distributions, in PO file format, and
`.pox' files are often temporary PO files.
MO files are meant to be read by programs, and are binary in nature.
A few systems already offer tools for creating and handling MO files
as part of the Native Language Support coming with the system, but the
format of these MO files is often different from system to system,
and non-portable. They do not necessary use `.mo' for file
extensions, but since system libraries are also used for accessing
these files, it works as long as the system is self-consistent about
it. If GNU gettext is able to interface with the tools already
provided with systems, it will consequently let these provided tools
take care of generating the MO files. Or else, if such tools are not
found or do not seem usable, GNU gettext will use its own ways
and its own format for MO files. Files ending with `.gmo' are
really MO files, when it is known that these files use the GNU format.
gettext
The following diagram summarizes the relation between the files
handled by GNU gettext and the tools acting on these files.
It is followed by a somewhat detailed explanations, which you should
read while keeping an eye on the diagram. Having a clear understanding
of these interrelations would surely help programmers, translators
and maintainers.
Original C Sources ---> PO mode ---> Marked C Sources ---.
|
.---------<--- GNU gettext Library |
.--- make <---+ |
| `---------<--------------------+-----------'
| |
| .-----<--- PACKAGE.pot <--- xgettext <---' .---<--- PO Compendium
| | | ^
| | `---. |
| `---. +---> PO mode ---.
| +----> msgmerge ------> LANG.pox --->--------' |
| .---' |
| | |
| `-------------<---------------. |
| +--- LANG.po <--- New LANG.pox <----'
| .--- LANG.gmo <--- msgfmt <---'
| |
| `---> install ---> /.../LANG/PACKAGE.mo ---.
| +---> "Hello world!"
`-------> install ---> /.../bin/PROGRAM -------'
The indication `PO mode' appears in two places in this picture, and you may safely read it as merely meaning "hand editing", using any editor of your choice, really. However, for those of you being the lucky users of GNU Emacs, PO mode has been specifically created for providing a cozy environment for editing or modifying PO files. While editing a PO file, PO mode allows for the easy browsing of auxiliary and compendium PO files, as well as for following references into the set of C program sources from which PO files have been derived. It has a few special features, among which are the interactive marking of program strings as translatable, and the validatation of PO files with easy repositioning to PO file lines showing errors.
As a programmer, the first step to bringing GNU gettext
into your package is identifying, right in the C sources, those strings
which are meant to be translatable, and those which are untranslatable.
This tedious job can be done a little more comfortably using emacs PO
mode, but you can use any means familiar to you for modifying your
C sources. Beside this some other simple, standard changes are needed to
properly initialize the translation library. See section Preparing Program Sources, for
more information about all this.
For newly written software the strings of course can and should be
marked while writing the it. The gettext approach makes this
very easy. Simply put the following lines at the beginning of each file
or in a central header file:
#define _(String) (String) #define N_(String) (String) #define textdomain(Domain) #define bindtextdomain(Package, Directory)
Doing this allows you to prepare the sources for internationalization.
Later when you feel ready for the step to use the gettext library
simply remove these definitions, include `libintl.h' and link
against `libintl.a'. That is all you have to change.
Once the C sources have been modified, the xgettext program
is used to find and extract all translatable strings, and create an
initial PO file out of all these. This `package.pot' file
contains all original program strings. It has sets of pointers to
exactly where in C sources each string is used. All translations
are set to empty. The letter t in `.pot' marks this as
a Template PO file, not yet oriented towards any particular language.
See section Invoking the xgettext Program, for more details about how one calls the
xgettext program. If you are really lazy, you might
be interested at working a lot more right away, and preparing the
whole distribution setup (see section The Maintainer's View). By doing so, you
spare yourself typing the xgettext command, as make
should now generate the proper things automatically for you!
The first time through, there is no `lang.po' yet, so the
msgmerge step may be skipped and replaced by a mere copy of
`package.pot' to `lang.pox', where lang
represents the target language.
Then comes the initial translation of messages. Translation in itself is a whole matter, still exclusively meant for humans, and whose complexity far overwhelms the level of this manual. Nevertheless, a few hints are given in some other chapter of this manual (see section The Translator's View). You will also find there indications about how to contact translating teams, or becoming part of them, for sharing your translating concerns with others who target the same native language.
While adding the translated messages into the `lang.pox' PO file, if you do not have GNU Emacs handy, you are on your own for ensuring that your efforts fully respect the PO file format, and quoting conventions (see section The Format of PO Files). This is surely not an impossible task, as this is the way many people have handled PO files already for Uniforum or Solaris. On the other hand, by using PO mode in GNU Emacs, most details of PO file format are taken care of for you, but you have to acquire some familiarity with PO mode itself. Besides main PO mode commands (see section Main PO mode Commands), you should know how to move between entries (see section Entry Positioning), and how to handle untranslated entries (see section Untranslated Entries).
If some common translations have already been saved into a compendium PO file, translators may use PO mode for initializing untranslated entries from the compendium, and also save selected translations into the compendium, updating it (see section Using Translation Compendiums). Compendium files are meant to be exchanged between members of a given translation team.
Programs, or packages of programs, are dynamic in nature: users write bug reports and suggestion for improvements, maintainers react by modifying programs in various ways. The fact that a package has already been internationalized should not make maintainers shy of adding new strings, or modifying strings already translated. They just do their job the best they can. For the Translation Project to work smoothly, it is important that maintainers do not carry translation concerns on their already loaded shoulders, and that translators be kept as free as possible of programmatic concerns.
The only concern maintainers should have is carefully marking new
strings as translatable, when they should be, and do not otherwise
worry about them being translated, as this will come in proper time.
Consequently, when programs and their strings are adjusted in various
ways by maintainers, and for matters usually unrelated to translation,
xgettext would construct `package.pot' files which are
evolving over time, so the translations carried by `lang.po'
are slowly fading out of date.
It is important for translators (and even maintainers) to understand that package translation is a continuous process in the lifetime of a package, and not something which is done once and for all at the start. After an initial burst of translation activity for a given package, interventions are needed once in a while, because here and there, translated entries become obsolete, and new untranslated entries appear, needing translation.
The msgmerge program has the purpose of refreshing an already
existing `lang.po' file, by comparing it with a newer
`package.pot' template file, extracted by xgettext
out of recent C sources. The refreshing operation adjusts all
references to C source locations for strings, since these strings
move as programs are modified. Also, msgmerge comments out as
obsolete, in `lang.pox', those already translated entries
which are no longer used in the program sources (see section Obsolete Entries). It finally discovers new strings and inserts them in
the resulting PO file as untranslated entries (see section Untranslated Entries). See section Invoking the msgmerge Program, for more information about what
msgmerge really does.
Whatever route or means taken, the goal is to obtain an updated `lang.pox' file offering translations for all strings. When this is properly achieved, this file `lang.pox' may take the place of the previous official `lang.po' file.
The temporal mobility, or fluidity of PO files, is an integral part of the translation game, and should be well understood, and accepted. People resisting it will have a hard time participating in the Translation Project, or will give a hard time to other participants! In particular, maintainers should relax and include all available official PO files in their distributions, even if these have not recently been updated, without banging or otherwise trying to exert pressure on the translator teams to get the job done. The pressure should rather come from the community of users speaking a particular language, and maintainers should consider themselves fairly relieved of any concern about the adequacy of translation files. On the other hand, translators should reasonably try updating the PO files they are responsible for, while the package is undergoing pretest, prior to an official distribution.
Once the PO file is complete and dependable, the msgfmt program
is used for turning the PO file into a machine-oriented format, which
may yield efficient retrieval of translations by the programs of the
package, whenever needed at runtime (see section The Format of GNU MO Files). See section Invoking the msgfmt Program, for more information about all modalities of execution
for the msgfmt program.
Finally, the modified and marked C sources are compiled and linked
with the GNU gettext library, usually through the operation of
make, given a suitable `Makefile' exists for the project,
and the resulting executable is installed somewhere users will find it.
The MO files themselves should also be properly installed. Given the
appropriate environment variables are set (see section Magic for End Users), the
program should localize itself automatically, whenever it executes.
The remainder of this manual has the purpose of explaining in depth the various steps outlined above.
The GNU gettext toolset helps programmers and translators
at producing, updating and using translation files, mainly those
PO files which are textual, editable files. This chapter stresses
the format of PO files, and contains a PO mode starter. PO mode
description is spread throughout this manual instead of being concentrated
in one place. Here we present only the basics of PO mode.
gettext Installation
Once you have received, unpacked, configured and compiled the GNU
gettext distribution, the `make install' command puts in
place the programs xgettext, msgfmt, gettext, and
msgmerge, as well as their available message catalogs. To
top off a comfortable installation, you might also want to make the
PO mode available to your GNU Emacs users.
During the installation of the PO mode, you might want modify your file `.emacs', once and for all, so it contains a few lines looking like:
(setq auto-mode-alist
(cons '("\\.po[tx]?\\'\\|\\.po\\." . po-mode) auto-mode-alist))
(autoload 'po-mode "po-mode")
Later, whenever you edit some `.po', `.pot' or `.pox' file, or any file having the string `.po.' within its name, Emacs loads `po-mode.elc' (or `po-mode.el') as needed, and automatically activates PO mode commands for the associated buffer. The string PO appears in the mode line for any buffer for which PO mode is active. Many PO files may be active at once in a single Emacs session.
If you are using Emacs version 20 or better, and have already installed the appropriate international fonts on your system, you may also manage for the these fonts to be automatically loaded and used for displaying the translations on your Emacs screen, whenever necessary. For this to happen, you might want to add the lines:
(autoload 'po-find-file-coding-system "po-mode")
(modify-coding-system-alist 'file "\\.po[tx]?\\'\\|\\.po\\."
'po-find-file-coding-system)
to your `.emacs' file.
A PO file is made up of many entries, each entry holding the relation between an original untranslated string and its corresponding translation. All entries in a given PO file usually pertain to a single project, and all translations are expressed in a single target language. One PO file entry has the following schematic structure:
white-space # translator-comments #. automatic-comments #: reference... #, flag... msgid untranslated-string msgstr translated-string
The general structure of a PO file should be well understood by the translator. When using PO mode, very little has to be known about the format details, as PO mode takes care of them for her.
Entries begin with some optional white space. Usually, when generated
through GNU gettext tools, there is exactly one blank line
between entries. Then comments follow, on lines all starting with the
character #. There are two kinds of comments: those which have
some white space immediately following the #, which comments are
created and maintained exclusively by the translator, and those which
have some non-white character just after the #, which comments
are created and maintained automatically by GNU gettext tools.
All comments, of either kind, are optional.
After white space and comments, entries show two strings, giving
first the untranslated string as it appears in the original program
sources, and then, the translation of this string. The original
string is introduced by the keyword msgid, and the translation,
by msgstr. The two strings, untranslated and translated,
are quoted in various ways in the PO file, using "
delimiters and \ escapes, but the translator does not really
have to pay attention to the precise quoting format, as PO mode fully
intend to take care of quoting for her.
The msgid strings, as well as automatic comments, are produced
and managed by other GNU gettext tools, and PO mode does not
provide means for the translator to alter these. The most she can
do is merely deleting them, and only by deleting the whole entry.
On the other hand, the msgstr string, as well as translator
comments, are really meant for the translator, and PO mode gives her
the full control she needs.
The comment lines beginning with #, are special because they are
not completely ignored by the programs as comments generally are. The
comma separated list of flags is used by the msgfmt
program to give the user some better disgnostic messages. Currently
there are two forms of flags defined:
msgmerge program or it can be
inserted by the translator herself. It shows that the msgstr
string might not be a correct translation (anymore). Only the translator
can judge if the translation requires further modification, or is
acceptable as is. Once satisfied with the translation, she then removes
this fuzzy attribute. The msgmerge programs inserts this
when it combined the msgid and msgstr entries after fuzzy
search only. See section Fuzzy Entries.
xgettext program adds them. In an automatized PO file processing
system as proposed here the user changes would be thrown away again as
soon as the xgettext program generates a new template file.
In case the c-format flag is given for a string the msgfmt
does some more tests to check to validity of the translation.
See section Invoking the msgfmt Program.
It happens that some lines, usually whitespace or comments, follow the very last entry of a PO file. Such lines are not part of any entry, and PO mode is unable to take action on those lines. By using the PO mode function M-x po-normalize, the translator may get rid of those spurious lines. See section Normalizing Strings in Entries.
The remainder of this section may be safely skipped by those using PO mode, yet it may be interesting for everybody to have a better idea of the precise format of a PO file. On the other hand, those not having GNU Emacs handy should carefully continue reading on.
Each of untranslated-string and translated-string respects the C syntax for a character string, including the surrounding quotes and imbedded backslashed escape sequences. When the time comes to write multi-line strings, one should not use escaped newlines. Instead, a closing quote should follow the last character on the line to be continued, and an opening quote should resume the string at the beginning of the following PO file line. For example:
msgid "" "Here is an example of how one might continue a very long string\n" "for the common case the string represents multi-line output.\n"
In this example, the empty string is used on the first line, to
allow better alignment of the H from the word `Here'
over the f from the word `for'. In this example, the
msgid keyword is followed by three strings, which are meant
to be concatenated. Concatenating the empty string does not change
the resulting overall string, but it is a way for us to comply with
the necessity of msgid to be followed by a string on the same
line, while keeping the multi-line presentation left-justified, as
we find this to be a cleaner disposition. The empty string could have
been omitted, but only if the string starting with `Here' was
promoted on the first line, right after msgid.(1) It was not really necessary
either to switch between the two last quoted strings immediately after
the newline `\n', the switch could have occurred after any
other character, we just did it this way because it is neater.
One should carefully distinguish between end of lines marked as `\n' inside quotes, which are part of the represented string, and end of lines in the PO file itself, outside string quotes, which have no incidence on the represented string.
Outside strings, white lines and comments may be used freely.
Comments start at the beginning of a line with `#' and extend
until the end of the PO file line. Comments written by translators
should have the initial `#' immediately followed by some white
space. If the `#' is not immediately followed by white space,
this comment is most likely generated and managed by specialized GNU
tools, and might disappear or be replaced unexpectedly when the PO
file is given to msgmerge.
After setting up Emacs with something similar to the lines in
section Completing GNU gettext Installation, PO mode is activated for a window when Emacs finds a
PO file in that window. This puts the window read-only and establishes a
po-mode-map, which is a genuine Emacs mode, in a way that is not derived
from text mode in any way. Functions found on po-mode-hook,
if any, will be executed.
When PO mode is active in a window, the letters `PO' appear in the mode line for that window. The mode line also displays how many entries of each kind are held in the PO file. For example, the string `132t+3f+10u+2o' would tell the translator that the PO mode contains 132 translated entries (see section Translated Entries, 3 fuzzy entries (see section Fuzzy Entries), 10 untranslated entries (see section Untranslated Entries) and 2 obsolete entries (see section Obsolete Entries). Zero-coefficients items are not shown. So, in this example, if the fuzzy entries were unfuzzied, the untranslated entries were translated and the obsolete entries were deleted, the mode line would merely display `145t' for the counters.
The main PO commands are those which do not fit into the other categories of subsequent sections. These allow for quitting PO mode or for managing windows in special ways.
The command U (po-undo) interfaces to the GNU Emacs
undo facility. See section `Undoing Changes' in The Emacs Editor. Each time U is typed, modifications which the translator
did to the PO file are undone a little more. For the purpose of
undoing, each PO mode command is atomic. This is especially true for
the RET command: the whole edition made by using a single
use of this command is undone at once, even if the edition itself
implied several actions. However, while in the editing window, one
can undo the edition work quite parsimoniously.
The commands Q (po-quit) and q
(po-confirm-and-quit) are used when the translator is done with the
PO file. The former is a bit less verbose than the latter. If the file
has been modified, it is saved to disk first. In both cases, and prior to
all this, the commands check if some untranslated message remains in the
PO file and, if yes, the translator is asked if she really wants to leave
off working with this PO file. This is the preferred way of getting rid
of an Emacs PO file buffer. Merely killing it through the usual command
C-x k (kill-buffer) is not the tidiest way to proceed.
The command O (po-other-window) is another, softer way,
to leave PO mode, temporarily. It just moves the cursor to some other
Emacs window, and pops one if necessary. For example, if the translator
just got PO mode to show some source context in some other, she might
discover some apparent bug in the program source that needs correction.
This command allows the translator to change sex, become a programmer,
and have the cursor right into the window containing the program she
(or rather he) wants to modify. By later getting the cursor back
in the PO file window, or by asking Emacs to edit this file once again,
PO mode is then recovered.
The command h (po-help) displays a summary of all available PO
mode commands. The translator should then type any character to resume
normal PO mode operations. The command ? has the same effect
as h.
The command = (po-statistics) computes the total number of
entries in the PO file, the ordinal of the current entry (counted from
1), the number of untranslated entries, the number of obsolete entries,
and displays all these numbers.
The command V (po-validate) launches msgfmt in verbose
mode over the current PO file. This command first offers to save the
current PO file on disk. The msgfmt tool, from GNU gettext,
has the purpose of creating a MO file out of a PO file, and PO mode uses
the features of this program for checking the overall format of a PO file,
as well as all individual entries.
The program msgfmt runs asynchronously with Emacs, so the
translator regains control immediately while her PO file is being studied.
Error output is collected in the GNU Emacs `*compilation*' buffer,
displayed in another window. The regular GNU Emacs command C-x`
(next-error), as well as other usual compile commands, allow the
translator to reposition quickly to the offending parts of the PO file.
Once the cursor is on the line in error, the translator may decide on
any PO mode action which would help correcting the error.
The cursor in a PO file window is almost always part of an entry. The only exceptions are the special case when the cursor is after the last entry in the file, or when the PO file is empty. The entry where the cursor is found to be is said to be the current entry. Many PO mode commands operate on the current entry, so moving the cursor does more than allowing the translator to browse the PO file, this also selects on which entry commands operate.
Some PO mode commands alter the position of the cursor in a specialized way. A few of those special purpose positioning are described here, the others are described in following sections.
Any GNU Emacs command able to reposition the cursor may be used
to select the current entry in PO mode, including commands which
move by characters, lines, paragraphs, screens or pages, and search
commands. However, there is a kind of standard way to display the
current entry in PO mode, which usual GNU Emacs commands moving
the cursor do not especially try to enforce. The command .
(po-current-entry) has the sole purpose of redisplaying the
current entry properly, after the current entry has been changed by
means external to PO mode, or the Emacs screen otherwise altered.
It is yet to be decided if PO mode helps the translator, or otherwise irritates her, by forcing a rigid window disposition while she is doing her work. We originally had quite precise ideas about how windows should behave, but on the other hand, anyone used to GNU Emacs is often happy to keep full control. Maybe a fixed window disposition might be offered as a PO mode option that the translator might activate or deactivate at will, so it could be offered on an experimental basis. If nobody feels a real need for using it, or a compulsion for writing it, we should drop this whole idea. The incentive for doing it should come from translators rather than programmers, as opinions from an experienced translator are surely more worth to me than opinions from programmers thinking about how others should do translation.
The commands n (po-next-entry) and p
(po-previous-entry) move the cursor the entry following,
or preceding, the current one. If n is given while the
cursor is on the last entry of the PO file, or if p
is given while the cursor is on the first entry, no move is done.
The commands < (po-first-entry) and >
(po-last-entry) move the cursor to the first entry, or last
entry, of the PO file. When the cursor is located past the last
entry in a PO file, most PO mode commands will return an error saying
`After last entry'. Moreover, the commands < and >
have the special property of being able to work even when the cursor
is not into some PO file entry, and one may use them for nicely
correcting this situation. But even these commands will fail on a
truly empty PO file. There are development plans for the PO mode for it
to interactively fill an empty PO file from sources. See section Marking Translatable Strings.
The translator may decide, before working at the translation of a particular entry, that she needs to browse the remainder of the PO file, maybe for finding the terminology or phraseology used in related entries. She can of course use the standard Emacs idioms for saving the current cursor location in some register, and use that register for getting back, or else, use the location ring.
PO mode offers another approach, by which cursor locations may be saved
onto a special stack. The command m (po-push-location)
merely adds the location of current entry to the stack, pushing
the already saved locations under the new one. The command
r (po-pop-location) consumes the top stack element and
reposition the cursor to the entry associated with that top element.
This position is then lost, for the next r will move the cursor
to the previously saved location, and so on until no locations remain
on the stack.
If the translator wants the position to be kept on the location stack, maybe for taking a look at the entry associated with the top element, then go elsewhere with the intent of getting back later, she ought to use m immediately after r.
The command x (po-exchange-location) simultaneously
reposition the cursor to the entry associated with the top element of
the stack of saved locations, and replace that top element with the
location of the current entry before the move. Consequently, repeating
the x command toggles alternatively between two entries.
For achieving this, the translator will position the cursor on the
first entry, use m, then position to the second entry, and
merely use x for making the switch.
There are many different ways for encoding a particular string into a
PO file entry, because there are so many different ways to split and
quote multi-line strings, and even, to represent special characters
by backslahsed escaped sequences. Some features of PO mode rely on
the ability for PO mode to scan an already existing PO file for a
particular string encoded into the msgid field of some entry.
Even if PO mode has internally all the built-in machinery for
implementing this recognition easily, doing it fast is technically
difficult. To facilitate a solution to this efficiency problem,
we decided on a canonical representation for strings.
A conventional representation of strings in a PO file is currently
under discussion, and PO mode experiments with a canonical representation.
Having both xgettext and PO mode converging towards a uniform
way of representing equivalent strings would be useful, as the internal
normalization needed by PO mode could be automatically satisfied
when using xgettext from GNU gettext. An explicit
PO mode normalization should then be only necessary for PO files
imported from elsewhere, or for when the convention itself evolves.
So, for achieving normalization of at least the strings of a given PO file needing a canonical representation, the following PO mode command is available:
The special command M-x po-normalize, which has no associate
keys, revises all entries, ensuring that strings of both original
and translated entries use uniform internal quoting in the PO file.
It also removes any crumb after the last entry. This command may be
useful for PO files freshly imported from elsewhere, or if we ever
improve on the canonical quoting format we use. This canonical format
is not only meant for getting cleaner PO files, but also for greatly
speeding up msgid string lookup for some other PO mode commands.
M-x po-normalize presently makes three passes over the entries.
The first implements heuristics for converting PO files for GNU
gettext 0.6 and earlier, in which msgid and msgstr
fields were using K&R style C string syntax for multi-line strings.
These heuristics may fail for comments not related to obsolete
entries and ending with a backslash; they also depend on subsequent
passes for finalizing the proper commenting of continued lines for
obsolete entries. This first pass might disappear once all oldish PO
files would have been adjusted. The second and third pass normalize
all msgid and msgstr strings respectively. They also
clean out those trailing backslashes used by XView's msgfmt
for continued lines.
Having such an explicit normalizing command allows for importing PO
files from other sources, but also eases the evolution of the current
convention, evolution driven mostly by aesthetic concerns, as of now.
It is easy to make suggested adjustments at a later time, as the
normalizing command and eventually, other GNU gettext tools
should greatly automate conformance. A description of the canonical
string format is given below, for the particular benefit of those not
having GNU Emacs handy, and who would nevertheless want to handcraft
their PO files in nice ways.
Right now, in PO mode, strings are single line or multi-line. A string goes multi-line if and only if it has embedded newlines, that is, if it matches `[^\n]\n+[^\n]'. So, we would have:
msgstr "\n\nHello, world!\n\n\n"
but, replacing the space by a newline, this becomes:
msgstr "" "\n" "\n" "Hello,\n" "world!\n" "\n" "\n"
We are deliberately using a caricatural example, here, to make the point clearer. Usually, multi-lines are not that bad looking. It is probable that we will implement the following suggestion. We might lump together all initial newlines into the empty string, and also all newlines introducing empty lines (that is, for n > 1, the n-1'th last newlines would go together on a separate string), so making the previous example appear:
msgstr "\n\n" "Hello,\n" "world!\n" "\n\n"
There are a few yet undecided little points about string normalization, to be documented in this manual, once these questions settle.
For the programmer, changes to the C source code fall into three
categories. First, you have to make the localization functions
known to all modules needing message translation. Second, you should
properly trigger the operation of GNU gettext when the program
initializes, usually from the main function. Last, you should
identify and especially mark all constant strings in your program
needing translation.
Presuming that your set of programs, or package, has been adjusted
so all needed GNU gettext files are available, and your
`Makefile' files are adjusted (see section The Maintainer's View), each C module
having translated C strings should contain the line:
#include <libintl.h>
The remaining changes to your C sources are discussed in the further sections of this chapter.
gettext OperationsThe initialization of locale data should be done with more or less the same code in every program, as demonstrated below:
int
main (argc, argv)
int argc;
char argv;
{
...
setlocale (LC_ALL, "");
bindtextdomain (PACKAGE, LOCALEDIR);
textdomain (PACKAGE);
...
}
PACKAGE and LOCALEDIR should be provided either by
`config.h' or by the Makefile. For now consult the gettext
sources for more information.
The use of LC_ALL might not be appropriate for you.
LC_ALL includes all locale categories and especially
LC_CTYPE. This later category is responsible for determining
character classes with the isalnum etc. functions from
`ctype.h' which could especially for programs, which process some
kind of input language, be wrong. For example this would mean that a
source code using the @,{c} (c-cedilla character) is runnable in
France but not in the U.S.
Some systems also have problems with parsing number using the
scanf functions if an other but the LC_ALL locale is used.
The standards say that additional formats but the one known in the
"C" locale might be recognized. But some systems seem to reject
numbers in the "C" locale format. In some situation, it might
also be a problem with the notation itself which makes it impossible to
recognize whether the number is in the "C" locale or the local
format. This can happen if thousands separator characters are used.
Some locales define this character accordfing to the national
conventions to '.' which is the same character used in the
"C" locale to denote the decimal point.
So it is sometimes necessary to replace the LC_ALL line in the
code above by a sequence of setlocale lines
{
...
setlocale (LC_TIME, "");
setlocale (LC_MESSAGES, "");
...
}
or to switch for and back to the character class in question. On all
POSIX conformant systems the locale categories LC_CTYPE,
LC_COLLATE, LC_MONETARY, LC_NUMERIC, and
LC_TIME are available. On some modern systems there is also a
locale LC_MESSAGES which is called on some old, XPG2 compliant
systems LC_RESPONSES.
All strings requiring translation should be marked in the C sources. Marking
is done in such a way that each translatable string appears to be
the sole argument of some function or preprocessor macro. There are
only a few such possible functions or macros meant for translation,
and their names are said to be marking keywords. The marking is
attached to strings themselves, rather than to what we do with them.
This approach has more uses. A blatant example is an error message
produced by formatting. The format string needs translation, as
well as some strings inserted through some `%s' specification
in the format, while the result from sprintf may have so many
different instances that it is impractical to list them all in some
`error_string_out()' routine, say.
This marking operation has two goals. The first goal of marking is for triggering the retrieval of the translation, at run time. The keyword are possibly resolved into a routine able to dynamically return the proper translation, as far as possible or wanted, for the argument string. Most localizable strings are found in executable positions, that is, attached to variables or given as parameters to functions. But this is not universal usage, and some translatable strings appear in structured initializations. See section Special Cases of Translatable Strings.
The second goal of the marking operation is to help xgettext
at properly extracting all translatable strings when it scans a set
of program sources and produces PO file templates.
The canonical keyword for marking translatable strings is
`gettext', it gave its name to the whole GNU gettext
package. For packages making only light use of the `gettext'
keyword, macro or function, it is easily used as is. However,
for packages using the gettext interface more heavily, it
is usually more convenient to give the main keyword a shorter, less
obtrusive name. Indeed, the keyword might appear on a lot of strings
all over the package, and programmers usually do not want nor need
their program sources to remind them forcefully, all the time, that they
are internationalized. Further, a long keyword has the disadvantage
of using more horizontal space, forcing more indentation work on
sources for those trying to keep them within 79 or 80 columns.
Many packages use `_' (a simple underline) as a keyword,
and write `_("Translatable string")' instead of `gettext
("Translatable string")'. Further, the coding rule, from GNU standards,
wanting that there is a space between the keyword and the opening
parenthesis is relaxed, in practice, for this particular usage.
So, the textual overhead per translatable string is reduced to
only three characters: the underline and the two parentheses.
However, even if GNU gettext uses this convention internally,
it does not offer it officially. The real, genuine keyword is truly
`gettext' indeed. It is fairly easy for those wanting to use
`_' instead of `gettext' to declare:
#include <libintl.h> #define _(String) gettext (String)
instead of merely using `#include <libintl.h>'.
Later on, the maintenance is relatively easy. If, as a programmer, you add or modify a string, you will have to ask yourself if the new or altered string requires translation, and include it within `_()' if you think it should be translated. `"%s: %d"' is an example of string not requiring translation!
In PO mode, one set of features is meant more for the programmer than for the translator, and allows him to interactively mark which strings, in a set of program sources, are translatable, and which are not. Even if it is a fairly easy job for a programmer to find and mark such strings by other means, using any editor of his choice, PO mode makes this work more comfortable. Further, this gives translators who feel a little like programmers, or programmers who feel a little like translators, a tool letting them work at marking translatable strings in the program sources, while simultaneously producing a set of translation in some language, for the package being internationalized.
The set of program sources, targetted by the PO mode commands describe here, should have an Emacs tags table constructed for your project, prior to using these PO file commands. This is easy to do. In any shell window, change the directory to the root of your project, then execute a command resembling:
etags src/*.[hc] lib/*.[hc]
presuming here you want to process all `.h' and `.c' files from the `src/' and `lib/' directories. This command will explore all said files and create a `TAGS' file in your root directory, somewhat summarizing the contents using a special file format Emacs can understand.
For packages following the GNU coding standards, there is
a make goal tags or TAGS which construct the tag files in
all directories and for all files containing source code.
Once your `TAGS' file is ready, the following commands assist the programmer at marking translatable strings in his set of sources. But these commands are necessarily driven from within a PO file window, and it is likely that you do not even have such a PO file yet. This is not a problem at all, as you may safely open a new, empty PO file, mainly for using these commands. This empty PO file will slowly fill in while you mark strings as translatable in your program sources.
The , (po-tags-search) command search for the next
occurrence of a string which looks like a possible candidate for
translation, and displays the program source in another Emacs window,
positioned in such a way that the string is near the top of this other
window. If the string is too big to fit whole in this window, it is
positioned so only its end is shown. In any case, the cursor
is left in the PO file window. If the shown string would be better
presented differently in different native languages, you may mark it
using M-, or M-.. Otherwise, you might rather ignore it
and skip to the next string by merely repeating the , command.
A string is a good candidate for translation if it contains a sequence of three or more letters. A string containing at most two letters in a row will be considered as a candidate if it has more letters than non-letters. The command disregards strings containing no letters, or isolated letters only. It also disregards strings within comments, or strings already marked with some keyword PO mode knows (see below).
If you have never told Emacs about some `TAGS' file to use, the command will request that you specify one from the minibuffer, the first time you use the command. You may later change your `TAGS' file by using the regular Emacs command M-x visit-tags-table, which will ask you to name the precise `TAGS' file you want to use. See section `Tag Tables' in The Emacs Editor.
Each time you use the , command, the search resumes from where it was left by the previous search, and goes through all program sources, obeying the `TAGS' file, until all sources have been processed. However, by giving a prefix argument to the command (C-u ,), you may request that the search be restarted all over again from the first program source; but in this case, strings that you recently marked as translatable will be automatically skipped.
Using this , command does not prevent using of other regular
Emacs tags commands. For example, regular tags-search or
tags-query-replace commands may be used without disrupting the
independent , search sequence. However, as implemented, the
initial , command (or the , command is used with a
prefix) might also reinitialize the regular Emacs tags searching to the
first tags file, this reinitialization might be considered spurious.
The M-, (po-mark-translatable) command will mark the
recently found string with the `_' keyword. The M-.
(po-select-mark-and-mark) command will request that you type
one keyword from the minibuffer and use that keyword for marking
the string. Both commands will automatically create a new PO file
untranslated entry for the string being marked, and make it the
current entry (making it easy for you to immediately proceed to its
translation, if you feel like doing it right away). It is possible
that the modifications made to the program source by M-, or
M-. render some source line longer than 80 columns, forcing you
to break and re-indent this line differently. You may use the O
command from PO mode, or any other window changing command from
GNU Emacs, to break out into the program source window, and do any
needed adjustments. You will have to use some regular Emacs command
to return the cursor to the PO file window, if you want command
, for the next string, say.
The M-. command has a few built-in speedups, so you do not have to explicitly type all keywords all the time. The first such speedup is that you are presented with a preferred keyword, which you may accept by merely typing RET at the prompt. The second speedup is that you may type any non-ambiguous prefix of the keyword you really mean, and the command will complete it automatically for you. This also means that PO mode has to know all your possible keywords, and that it will not accept mistyped keywords.
If you reply ? to the keyword request, the command gives a list of all known keywords, from which you may choose. When the command is prefixed by an argument (C-u M-.), it inhibits updating any program source or PO file buffer, and does some simple keyword management instead. In this case, the command asks for a keyword, written in full, which becomes a new allowed keyword for later M-. commands. Moreover, this new keyword automatically becomes the preferred keyword for later commands. By typing an already known keyword in response to C-u M-., one merely changes the preferred keyword and does nothing more.
All keywords known for M-. are recognized by the , command when scanning for strings, and strings already marked by any of those known keywords are automatically skipped. If many PO files are opened simultaneously, each one has its own independent set of known keywords. There is no provision in PO mode, currently, for deleting a known keyword, you have to quit the file (maybe using q) and reopen it afresh. When a PO file is newly brought up in an Emacs window, only `gettext' and `_' are known as keywords, and `gettext' is preferred for the M-. command. In fact, this is not useful to prefer `_', as this one is already built in the M-, command.
In C programs strings are often used within calls of functions from the
printf family. The special thing about these format strings is
that they can contain format specifiers introduced with %. Assume
we have the code
printf (gettext ("String `%s' has %d characters\n"), s, strlen (s));
A possible German translation for the above string might be:
"%d Zeichen lang ist die Zeichenkette `%s'"
A C programmer, even if he cannot speak German, will recognize that
there is something wrong here. The order of the two format specifiers
is changed but of course the arguments in the printf don't have.
This will most probably lead to problems because now the length of the
string is regarded as the address.
To prevent errors at runtime caused by translations the msgfmt
tool can check statically whether the arguments in the original and the
translation string match in type and number. If this is not the case a
warning will be given and the error cannot causes problems at runtime.
If the word order in the above German translation would be correct one would have to write
"%2$d Zeichen lang ist die Zeichenkette `%1$s'"
The routines in msgfmt know about this special notation.
Because not all strings in a program must be format strings it is not
useful for msgfmt to test all the strings in the `.po' file.
This might cause problems because the string might contain what looks
like a format specifier, but the string is not used in printf.
Therefore the xgettext adds a special tag to those messages it
thinks might be a format string. There is no absolute rule for this,
only a heuristic. In the `.po' file the entry is marked using the
c-format flag in the #, comment line (see section The Format of PO Files).
The careful reader now might say that this again can cause problems.
The heuristic might guess it wrong. This is true and therefore
xgettext knows about special kind of comment which lets
the programmer take over the decision. If in the same line or
the immediately preceding line of the gettext keyword
the xgettext program find a comment containing the words
xgettext:c-format it will mark the string in any case with
the c-format flag. This kind of comment should be used when
xgettext does not recognize the string as a format string but
is really is one and it should be tested. Please note that when the
comment is in the same line of the gettext keyword, it must be
before the string to be translated.
This situation happens quite often. The printf function is often
called with strings which do not contain a format specifier. Of course
one would normally use fputs but it does happen. In this case
xgettext does not recognize this as a format string but what
happens if the translation introduces a valid format specifier? The
printf function will try to access one of the parameter but none
exists because the original code does not refer to any parameter.
xgettext of course could make a wrong decision the other way
round. A string marked as a format string is not really a format
string. In this case the msgfmt might give too many warnings and
would prevent translating the `.po' file. The method to prevent
this wrong decision is similar to the one used above, only the comment
to use must contain the string xgettext:no-c-format.
If a string is marked with c-format and this is not correct the
user can find out who is responsible for the decision. See section Invoking the xgettext Program to see how the --debug option can be used for solving
this problem.
The attentive reader might now point out that it is not always possible
to mark translatable string with gettext or something like this.
Consider the following case:
{
static const char *messages[] = {
"some very meaningful message",
"and another one"
};
const char *string;
...
string
= index > 1 ? "a default message" : messages[index];
fputs (string);
...
}
While it is no problem to mark the string "a default message" it
is not possible to mark the string initializers for messages.
What is to be done? We have to fulfill two tasks. First we have to mark the
strings so that the xgettext program (see section Invoking the xgettext Program)
can find them, and second we have to translate the string at runtime
before printing them.
The first task can be fulfilled by creating a new keyword, which names a no-op. For the second we have to mark all access points to a string from the array. So one solution can look like this:
#define gettext_noop(String) (String)
{
static const char *messages[] = {
gettext_noop ("some very meaningful message"),
gettext_noop ("and another one")
};
const char *string;
...
string
= index > 1 ? gettext ("a default message") : gettext (messages[index]);
fputs (string);
...
}
Please convince yourself that the string which is written by
fputs is translated in any case. How to get xgettext know
the additional keyword gettext_noop is explained in section Invoking the xgettext Program.
The above is of course not the only solution. You could also come along with the following one:
#define gettext_noop(String) (String)
{
static const char *messages[] = {
gettext_noop ("some very meaningful message",
gettext_noop ("and another one")
};
const char *string;
...
string
= index > 1 ? gettext_noop ("a default message") : messages[index];
fputs (gettext (string));
...
}
But this has some drawbacks. First the programmer has to take care that
he uses gettext_noop for the string "a default message".
A use of gettext could have in rare cases unpredictable results.
The second reason is found in the internals of the GNU gettext
Library which will make this solution less efficient.
One advantage is that you need not make control flow analysis to make sure the output is really translated in any case. But this analysis is generally not very difficult. If it should be in any situation you can use this second method in this situation.
xgettext Programxgettext [option] inputfile ...
xgettext program decided, the format form is used if
the programmer prescribed it.
By default only the c-format form is used. The translator should
not have to care about these details.
gettext, dgettext, dcgettext and
gettext_noop.
.gmo files. We can ship some of
these files in the GNU gettext package, and the result of
regenerating them through msgfmt should yield the same values.
Search path for supplementary PO files is: `/usr/local/share/nls/src/'.
If inputfile is `-', standard input is read.
This implementation of xgettext is able to process a few awkward
cases, like strings in preprocessor macros, ANSI concatenation of
adjacent strings, and escaped end of lines for continued strings.
PO mode is particularily powerful when used with PO files
created through GNU gettext utilities, as those utilities
insert special comments in the PO files they generate.
Some of these special comments relate the PO file entry to
exactly where the untranslated string appears in the program sources.
When the translator gets to an untranslated entry, she is fairly often faced with an original string which is not as informative as it normally should be, being succinct, cryptic, or otherwise ambiguous. Before chosing how to translate the string, she needs to understand better what the string really means and how tight the translation has to be. Most of times, when problems arise, the only way left to make her judgment is looking at the true program sources from where this string originated, searching for surrounding comments the programmer might have put in there, and looking around for helping clues of any kind.
Surely, when looking at program sources, the translator will receive more help if she is a fluent programmer. However, even if she is not versed in programming and feels a little lost in C code, the translator should not be shy at taking a look, once in a while. It is most probable that she will still be able to find some of the hints she needs. She will learn quickly to not feel uncomfortable in program code, paying more attention to programmer's comments, variable and function names (if he dared chosing them well), and overall organization, than to programmation itself.
The following commands are meant to help the translator at getting program source context for a PO file entry.
The commands s (po-cycle-reference) and M-s
(po-select-source-reference) both open another window displaying
some source program file, and already positioned in such a way that
it shows an actual use of the string to be translated. By doing
so, the command gives source program context for the string. But if
the entry has no source context references, or if all references
are unresolved along the search path for program sources, then the
command diagnoses this as an error.
Even if s (or M-s) opens a new window, the cursor stays in the PO file window. If the translator really wants to get into the program source window, she ought to do it explicitly, maybe by using command O.
When s is typed for the first time, or for a PO file entry which is different of the last one used for getting source context, then the command reacts by giving the first context available for this entry, if any. If some context has already been recently displayed for the current PO file entry, and the translator wandered off to do other things, typing s again will merely resume, in another window, the context last displayed. In particular, if the translator moved the cursor away from the context in the source file, the command will bring the cursor back to the context. By using s many times in a row, with no other commands intervening, PO mode will cycle to the next available contexts for this particular entry, getting back to the first context once the last has been shown.
The command M-s behaves differently. Instead of cycling through references, it lets the translator choose of particular reference among many, and displays that reference. It is best used with completion, if the translator types TAB immediately after M-s, in response to the question, she will be offered a menu of all possible references, as a reminder of which are the acceptable answers. This command is useful only where there are really many contexts available for a single string to translate.
Program source files are usually found relative to where the PO
file stands. As a special provision, when this fails, the file is
also looked for, but relative to the directory immediately above it.
Those two cases take proper care of most PO files. However, it might
happen that a PO file has been moved, or is edited in a different
place than its normal location. When this happens, the translator
should tell PO mode in which directory normally sits the genuine PO
file. Many such directories may be specified, and all together, they
constitute what is called the search path for program sources.
The command S (po-consider-source-path) is used to interactively
enter a new directory at the front of the search path, and the command
M-S (po-ignore-source-path) is used to select, with completion,
one of the directories she does not want anymore on the search path.
Compendiums are yet to be implemented.
An incoming PO mode feature will let the translator maintain a compendium of already achieved translations. A compendium is a special PO file containing a set of translations recurring in many different packages. The translator will be given commands for adding entries to her compendium, and later initializing untranslated entries, or updating already translated entries, from translations kept in the compendium. For this to work, however, the compendium would have to be normalized. See section Normalizing Strings in Entries.
msgmerge Program
Each PO file entry for which the msgstr field has been filled with
a translation, and which is not marked as fuzzy (see section Fuzzy Entries),
is a said to be a translated entry. Only translated entries will
later be compiled by GNU msgfmt and become usable in programs.
Other entry types will be excluded; translation will not occur for them.
Some commands are more specifically related to translated entry processing.
The commands t (po-next-translated-entry) and M-t
(po-previous-transted-entry) move forwards or backwards, chasing
for an translated entry. If none is found, the search is extended and
wraps around in the PO file buffer.
Translated entries usually result from the translator having edited in
a translation for them, section Modifying Translations. However, if the
variable po-auto-fuzzy-on-edit is not nil, the entry having
received a new translation first becomes a fuzzy entry, which ought to
be later unfuzzied before becoming an official, genuine translated entry.
See section Fuzzy Entries.
Each PO file entry may have a set of attributes, which are
qualities given an name and explicitely associated with the entry
translation, using a special system comment. One of these attributes
has the name fuzzy, and entries having this attribute are said
to have a fuzzy translation. They are called fuzzy entries, for short.
Fuzzy entries, even if they account for translated entries for
most other purposes, usually call for revision by the translator.
Those may be produced by applying the program msgmerge to
update an older translated PO files according to a new PO template
file, when this tool hypothesises that some new msgid has
been modified only slightly out of an older one, and chooses to pair
what it thinks to be the old translation for the new modified entry.
The slight alteration in the original string (the msgid string)
should often be reflected in the translated string, and this requires
the intervention of the translator. For this reason, msgmerge
might mark some entries as being fuzzy.
Also, the translator may decide herself to mark an entry as fuzzy for her own convenience, when she wants to remember that the entry has to be later revisited. So, some commands are more specifically related to fuzzy entry processing.
The commands f (po-next-fuzzy) and M-f
(po-previous-fuzzy) move forwards or backwards, chasing for
a fuzzy entry. If none is found, the search is extended and wraps
around in the PO file buffer.
The command TAB (po-unfuzzy) removes the fuzzy
attribute associated with an entry, usually leaving it translated.
Further, if the variable po-auto-select-on-unfuzzy has not
the nil value, the TAB command will automatically chase
for another interesting entry to work on. The initial value of
po-auto-select-on-unfuzzy is nil.
The initial value of po-auto-fuzzy-on-edit is nil. However,
if the variable po-auto-fuzzy-on-edit is set to t, any entry
edited through the RET command is marked fuzzy, as a way to ensure
some kind of double check, later. In this case, the usual paradigm is
that an entry becomes fuzzy (if not already) whenever the translator
modifies it. If she is satisfied with the translation, she then uses
TAB to pick another entry to work on, clearing the fuzzy attribute
on the same blow. If she is not satisfied yet, she merely uses SPC
to chase another entry, leaving the entry fuzzy.
The translator may also use the DEL command
(po-fade-out-entry) over any translated entry to mark it as being
fuzzy, when she wants to easily leave a trace she wants to later return
working at this entry.
Also, when time comes to quit working on a PO file buffer with the q command, the translator is asked for confirmation, if fuzzy string still exists.
When xgettext originally creates a PO file, unless told
otherwise, it initializes the msgid field with the untranslated
string, and leaves the msgstr string to be empty. Such entries,
having an empty translation, are said to be untranslated entries.
Later, when the programmer slightly modifies some string right in
the program, this change is later reflected in the PO file
by the appearance of a new untranslated entry for the modified string.
The usual commands moving from entry to entry consider untranslated entries on the same level as active entries. Untranslated entries are easily recognizable by the fact they end with `msgstr ""'.
The work of the translator might be (quite naively) seen as the process of seeking after an untranslated entry, editing a translation for it, and repeating these actions until no untranslated entries remain. Some commands are more specifically related to untranslated entry processing.
The commands u (po-next-untranslated-entry) and M-u
(po-previous-untransted-entry) move forwards or backwards,
chasing for an untranslated entry. If none is found, the search is
extended and wraps around in the PO file buffer.
An entry can be turned back into an untranslated entry by
merely emptying its translation, using the command k
(po-kill-msgstr). See section Modifying Translations.
Also, when time comes to quit working on a PO file buffer with the q command, the translator is asked for confirmation, if some untranslated string still exists.
By obsolete PO file entries, we mean those entries which are
commented out, usually by msgmerge when it found that the
translation is not needed anymore by the package being localized.
The usual commands moving from entry to entry consider obsolete
entries on the same level as active entries. Obsolete entries are
easily recognizable by the fact that all their lines start with
#, even those lines containing msgid or msgstr.
Commands exist for emptying the translation or reinitializing it to the original untranslated string. Commands interfacing with the kill ring may force some previously saved text into the translation. The user may interactively edit the translation. All these commands may apply to obsolete entries, carefully leaving the entry obsolete after the fact.
Moreover, some commands are more specifically related to obsolete entry processing.
The commands o (po-next-obsolete-entry) and M-o
(po-previous-obsolete-entry) move forwards or backwards,
chasing for an obsolete entry. If none is found, the search is
extended and wraps around in the PO file buffer.
PO mode does not provide ways for un-commenting an obsolete entry
and making it active, because this would reintroduce an original
untranslated string which does not correspond to any marked string
in the program sources. This goes with the philosophy of never
introducing useless msgid values.
However, it is possible to comment out an active entry, so making
it obsolete. GNU gettext utilities will later react to the
disappearance of a translation by using the untranslated string.
The command DEL (po-fade-out-entry) pushes the current entry
a little further towards annihilation. If the entry is active (it is a
translated entry), then it is first made fuzzy. If it is already fuzzy,
then the entry is merely commented out, with confirmation. If the entry
is already obsolete, then it is completely deleted from the PO file.
It is easy to recycle the translation so deleted into some other PO file
entry, usually one which is untranslated. See section Modifying Translations.
Here is a quite interesting problem to solve for later development of PO mode, for those nights you are not sleepy. The idea would be that PO mode might become bright enough, one of these days, to make good guesses at retrieving the most probable candidate, among all obsolete entries, for initializing the translation of a newly appeared string. I think it might be a quite hard problem to do this algorithmically, as we have to develop good and efficient measures of string similarity. Right now, PO mode completely lets the decision to the translator, when the time comes to find the adequate obsolete translation, it merely tries to provide handy tools for helping her to do so.
PO mode prevents direct edition of the PO file, by the usual means Emacs give for altering a buffer's contents. By doing so, it pretends helping the translator to avoid little clerical errors about the overall file format, or the proper quoting of strings, as those errors would be easily made. Other kinds of errors are still possible, but some may be caught and diagnosed by the batch validation process, which the translator may always trigger by the V command. For all other errors, the translator has to rely on her own judgment, and also on the linguistic reports submitted to her by the users of the translated package, having the same mother tongue.
When the time comes to create a translation, correct an error diagnosed mechanically or reported by a user, the translators have to resort to using the following commands for modifying the translations.
The command RET (po-edit-msgstr) opens a new Emacs window
containing a copy of the translation taken from the current PO file entry,
all ready for edition, fully modifiable and with the complete extent of
GNU Emacs modifying commands. The string is presented to the translator
expunged of all quoting marks, and she will modify the unquoted
string in this window to heart's content. Once done, the regular Emacs
command M-C-c (exit-recursive-edit) may be used to return the
edited translation into the PO file, replacing the original translation.
The keys C-c C-c are bound so they have the same effect as
M-C-c.
If the translator becomes unsatisfied with her translation to the extent
she prefers keeping the translation which was existent prior to the
RET command, she may use the standard Emacs command C-]
(abort-recursive-edit) to merely get rid of edition, while
preserving the original translation. The keys C-c C-k are
bound so they have the same effect as C-]. Another way would
be for her to exit normally with C-c C-c, then type U
once for undoing the whole effect of last edition.
Functions found on po-subedit-mode-hook, if any, are executed after
the string has been inserted in the edit buffer and before recursive edit
is entered.
While editing her translation, the translator should pay attention to not inserting unwanted RET (carriage returns) characters at the end of the translated string if those are not meant to be there, or to removing such characters when they are required. Since these characters are not visible in the editing buffer, they are easily introduced by mistake. To help her, RET automatically puts the character < at the end of the string being edited, but this < is not really part of the string. On exiting the editing window with C-c C-c, PO mode automatically removes such < and all whitespace added after it. If the translator adds characters after the terminating <, it looses its delimiting property and integrally becomes part of the string. If she removes the delimiting <, then the edited string is taken as is, with all trailing newlines, even if invisible. Also, if the translated string ought to end itself with a genuine <, then the delimiting < may not be removed; so the string should appear, in the editing window, as ending with two < in a row.
When a translation (or a comment) is being edited, the translator may move the cursor back into the PO file buffer and freely move to other entries, browsing at will. The edited entry will be recovered as soon as the edit ceases, because it is this entry only which is being modified. If, with an edition still opened, the translator wanders in the PO file buffer, she cannot modify any other entry. If she tries to, PO mode will react by suggesting that she abort the current edit, or else, by inviting her to finish the current edit prior to any other modification.
The command LFD (po-msgid-to-msgstr) initializes, or
reinitializes the translation with the original string. This command
is normally used when the translator wants to redo a fresh translation
of the original string, disregarding any previous work.
It is possible to arrange so, whenever editing an untranslated
entry, the LFD command be automatically executed. If you set
po-auto-edit-with-msgid to t, the translation gets
initialised with the original string, in case none exist already.
The default value for po-auto-edit-with-msgid is nil.
In fact, whether it is best to start a translation with an empty string, or rather with a copy of the original string, is a matter of taste or habit. Sometimes, the source language and the target language are so different that is simply best to start writing on an empty page. At other times, the source and target languages are so close that it would be a waste to retype a number of words already being written in the original string. A translator may also like having the original string right under her eyes, as she will progressively overwrite the original text with the translation, even if this requires some extra editing work to get rid of the original.
The command k (po-kill-msgstr) merely empties the
translation string, so turning the entry into an untranslated
one. But while doing so, its previous contents is put apart in
a special place, known as the kill ring. The command w
(po-kill-ring-save-msgstr) has also the effect of taking a
copy of the translation onto the kill ring, but it otherwise leaves
the entry alone, and does not remove the translation from the
entry. Both commands use exactly the Emacs kill ring, which is shared
between buffers, and which is well known already to GNU Emacs lovers.
The translator may use k or w many times in the course of her work, as the kill ring may hold several saved translations. From the kill ring, strings may later be reinserted in various Emacs buffers. In particular, the kill ring may be used for moving translation strings between different entries of a single PO file buffer, or if the translator is handling many such buffers at once, even between PO files.
To facilitate exchanges with buffers which are not in PO mode, the translation string put on the kill ring by the k command is fully unquoted before being saved: external quotes are removed, multi-lines strings are concatenated, and backslashed escaped sequences are turned into their corresponding characters. In the special case of obsolete entries, the translation is also uncommented prior to saving.
The command y (po-yank-msgstr) completely replaces the
translation of the current entry by a string taken from the kill ring.
Following GNU Emacs terminology, we then say that the replacement
string is yanked into the PO file buffer.
See section `Yanking' in The Emacs Editor.
The first time y is used, the translation receives the value of
the most recent addition to the kill ring. If y is typed once
again, immediately, without intervening keystrokes, the translation
just inserted is taken away and replaced by the second most recent
addition to the kill ring. By repeating y many times in a row,
the translator may travel along the kill ring for saved strings,
until she finds the string she really wanted.
When a string is yanked into a PO file entry, it is fully and automatically requoted for complying with the format PO files should have. Further, if the entry is obsolete, PO mode then appropriately push the inserted string inside comments. Once again, translators should not burden themselves with quoting considerations besides, of course, the necessity of the translated string itself respective to the program using it.
Note that k or w are not the only commands pushing strings on the kill ring, as almost any PO mode command replacing translation strings (or the translator comments) automatically save the old string on the kill ring. The main exceptions to this general rule are the yanking commands themselves.
To better illustrate the operation of killing and yanking, let's
use an actual example, taken from a common situation. When the
programmer slightly modifies some string right in the program, his
change is later reflected in the PO file by the appearance
of a new untranslated entry for the modified string, and the fact
that the entry translating the original or unmodified string becomes
obsolete. In many cases, the translator might spare herself some work
by retrieving the unmodified translation from the obsolete entry,
then initializing the untranslated entry msgstr field with
this retrieved translation. Once this done, the obsolete entry is
not wanted anymore, and may be safely deleted.
When the translator finds an untranslated entry and suspects that a
slight variant of the translation exists, she immediately uses m
to mark the current entry location, then starts chasing obsolete
entries with o, hoping to find some translation corresponding
to the unmodified string. Once found, she uses the DEL command
for deleting the obsolete entry, knowing that DEL also kills
the translation, that is, pushes the translation on the kill ring.
Then, r returns to the initial untranslated entry, y
then yanks the saved translation right into the msgstr
field. The translator is then free to use RET for fine
tuning the translation contents, and maybe to later use u,
then m again, for going on with the next untranslated string.
When some sequence of keys has to be typed over and over again, the translator may find it useful to become better acquainted with the GNU Emacs capability of learning these sequences and playing them back under request. See section `Keyboard Macros' in The Emacs Editor.
Any translation work done seriously will raise many linguistic difficulties, for which decisions have to be made, and the choices further documented. These documents may be saved within the PO file in form of translator comments, which the translator is free to create, delete, or modify at will. These comments may be useful to herself when she returns to this PO file after a while.
Comments not having whitespace after the initial `#', for example,
those beginning with `#.' or `#:', are not translator
comments, they are exclusively created by other gettext tools.
So, the commands below will never alter such system added comments,
they are not meant for the translator to modify. See section The Format of PO Files.
The following commands are somewhat similar to those modifying translations, so the general indications given for those apply here. See section Modifying Translations.
These commands parallel PO mode commands for modifying the translation strings, and behave much the same way as they do, except that they handle this part of PO file comments meant for translator usage, rather than the translation strings. So, if the descriptions given below are slightly succinct, it is because the full details have already been given. See section Modifying Translations.
The command # (po-edit-comment) opens a new Emacs
window containing a copy of the translator comments on the current
PO file entry. If there are no such comments, PO mode
understands that the translator wants to add a comment to the entry,
and she is presented with an empty screen. Comment marks (#) and
the space following them are automatically removed before edition,
and reinstated after. For translator comments pertaining to obsolete
entries, the uncommenting and recommenting operations are done twice.
Once in the editing window, the keys C-c C-c allow the
translator to tell she is finished with editing the comment.
Functions found on po-subedit-mode-hook, if any, are executed after
the string has been inserted in the edit buffer and before recursive edit
is entered.
The command K (po-kill-comment) get rid of all
translator comments, while saving those comments on the kill ring.
The command W (po-kill-ring-save-comment) takes
a copy of the translator comments on the kill ring, but leaves
them undisturbed in the current entry. The command Y
(po-yank-comment) completely replaces the translator comments
by a string taken at the front of the kill ring. When this command
is immediately repeated, the comments just inserted are withdrawn,
and replaced by other strings taken along the kill ring.
On the kill ring, all strings have the same nature. There is no distinction between translation strings and translator comments strings. So, for example, let's presume the translator has just finished editing a translation, and wants to create a new translator comment to document why the previous translation was not good, just to remember what was the problem. Foreseeing that she will do that in her documentation, the translator may want to quote the previous translation in her translator comments. To do so, she may initialize the translator comments with the previous translation, still at the head of the kill ring. Because editing already pushed the previous translation on the kill ring, she merely has to type M-w prior to #, and the previous translation will be right there, all ready for being introduced by some explanatory text.
On the other hand, presume there are some translator comments already
and that the translator wants to add to those comments, instead
of wholly replacing them. Then, she should edit the comment right
away with #. Once inside the editing window, she can use the
regular GNU Emacs commands C-y (yank) and M-y
(yank-pop) to get the previous translation where she likes.
PO mode is able to help the knowledgeable translator, being fluent in many languages, at taking advantage of translations already achieved in other languages she just happens to know. It provides these other language translations as additional context for her own work. Moreover, it has features to ease the production of translations for many languages at once, for translators preferring to work in this way.
An auxiliary PO file is an existing PO file meant for the same package the translator is working on, but targeted to a different mother tongue language. Commands exist for declaring and handling auxiliary PO files, and also for showing contexts for the entry under work.
Here are the auxiliary file commands available in PO mode.
Command A (po-consider-as-auxiliary) adds the current
PO file to the list of auxiliary files, while command M-A
(po-ignore-as-auxiliary just removes it.
The command a (po-cycle-auxiliary) seeks all auxiliary PO
files, round-robin, searching for a translated entry in some other language
having an msgid field identical as the one for the current entry.
The found PO file, if any, takes the place of the current PO file in
the display (its window gets on top). Before doing so, the current PO
file is also made into an auxiliary file, if not already. So, a
in this newly displayed PO file will seek another PO file, and so on,
so repeating a will eventually yield back the original PO file.
The command M-a (po-select-auxiliary) asks the translator
for her choice of a particular auxiliary file, with completion, and
then switches to that selected PO file. The command also checks if
the selected file has an msgid field identical as the one for
the current entry, and if yes, this entry becomes current. Otherwise,
the cursor of the selected file is left undisturbed.
For all this to work fully, auxiliary PO files will have to be normalized,
in that way that msgid fields should be written exactly
the same way. It is possible to write msgid fields in various
ways for representing the same string, different writing would break the
proper behaviour of the auxiliary file commands of PO mode. This is not
expected to be much a problem in practice, as most existing PO files have
their msgid entries written by the same GNU gettext tools.
However, PO files initially created by PO mode itself, while marking
strings in source files, are normalised differently. So are PO
files resulting of the the `M-x normalize' command. Until these
discrepancies between PO mode and other GNU gettext tools get
fully resolved, the translator should stay aware of normalisation issues.
msgfmt ProgramUsage: msgfmt [option] filename.po ...
msgid and msgstr strings are
studied and compared. It is considered abnormal that one string
starts or ends with a newline while the other does not.
Also, if the string represents a format sring used in a
printf-like function both strings should have the same number of
`%' format specifiers, with matching types. If the flag
c-format or possible-c-format appears in the special
comment #, for this entry a check is performed. For example, the
check will diagnose using `%.*s' against `%s', or `%d'
against `%s', or `%d' against `%x'. It can even handle
positional parameters.
Normally the xgettext program automatically decides whether a
string is a format string or not. This algorithm is not perfect,
though. It might regard a string as a format string though it is not
used in a printf-like function and so msgfmt might report
errors where there are none. Or the other way round: a string is not
regarded as a format string but it is used in a printf-like
function.
So solve this problem the programmer can dictate the decision to the
xgettext program (see section Special Comments preceding Keywords). The translator should not
consider removing the flag from the #, line. This "fix" would be
reversed again as soon as msgmerge is called the next time.
If input file is `-', standard input is read. If output file is `-', output is written to standard output.
The format of the generated MO files is best described by a picture, which appears below.
The first two words serve the identification of the file. The magic
number will always signal GNU MO files. The number is stored in the
byte order of the generating machine, so the magic number really is
two numbers: 0x950412de and 0xde120495. The second
word describes the current revision of the file format. For now the
revision is 0. This might change in future versions, and ensures
that the readers of MO files can distinguish new formats from old
ones, so that both can be handled correctly. The version is kept
separate from the magic number, instead of using different magic
numbers for different formats, mainly because `/etc/magic' is
not updated often. It might be better to have magic separated from
internal format version identification.
Follow a number of pointers to later tables in the file, allowing for the extension of the prefix part of MO files without having to recompile programs reading them. This might become useful for later inserting a few flag bits, indication about the charset used, new tables, or other things.
Then, at offset O and offset T in the picture, two tables of string descriptors can be found. In both tables, each string descriptor uses two 32 bits integers, one for the string length, another for the offset of the string in the MO file, counting in bytes from the start of the file. The first table contains descriptors for the original strings, and is sorted so the original strings are in increasing lexicographical order. The second table contains descriptors for the translated strings, and is parallel to the first table: to find the corresponding translation one has to access the array slot in the second array with the same index.
Having the original strings sorted enables the use of simple binary
search, for when the MO file does not contain an hashing table, or
for when it is not practical to use the hashing table provided in
the MO file. This also has another advantage, as the empty string
in a PO file GNU gettext is usually translated into
some system information attached to that particular MO file, and the
empty string necessarily becomes the first in both the original and
translated tables, making the system information very easy to find.
The size S of the hash table can be zero. In this case, the
hash table itself is not contained in the MO file. Some people might
prefer this because a precomputed hashing table takes disk space, and
does not win that much speed. The hash table contains indices
to the sorted array of strings in the MO file. Conflict resolution is
done by double hashing. The precise hashing algorithm used is fairly
dependent of GNU gettext code, and is not documented here.
As for the strings themselves, they follow the hash file, and each
is terminated with a NUL, and this NUL is not counted in
the length which appears in the string descriptor. The msgfmt
program has an option selecting the alignment for MO file strings.
With this option, each string is separately aligned so it starts at
an offset which is a multiple of the alignment value. On some RISC
machines, a correct alignment will speed things up.
Nothing prevents a MO file from having embedded NULs in strings. However, the program interface currently used already presumes that strings are NUL terminated, so embedded NULs are somewhat useless. But MO file format is general enough so other interfaces would be later possible, if for example, we ever want to implement wide characters right in MO files, where NUL bytes may accidently appear.
This particular issue has been strongly debated in the GNU
gettext development forum, and it is expectable that MO file
format will evolve or change over time. It is even possible that many
formats may later be supported concurrently. But surely, we have to
start somewhere, and the MO file format described here is a good start.
Nothing is cast in concrete, and the format may later evolve fairly
easily, so we should feel comfortable with the current approach.
byte
+------------------------------------------+
0 | magic number = 0x950412de |
| |
4 | file format revision = 0 |
| |
8 | number of strings | == N
| |
12 | offset of table with original strings | == O
| |
16 | offset of table with translation strings | == T
| |
20 | size of hashing table | == S
| |
24 | offset of hashing table | == H
| |
. .
. (possibly more entries later) .
. .
| |
O | length & offset 0th string ----------------.
O + 8 | length & offset 1st string ------------------.
... ... | |
O + ((N-1)*8)| length & offset (N-1)th string | | |
| | | |
T | length & offset 0th translation ---------------.
T + 8 | length & offset 1st translation -----------------.
... ... | | | |
T + ((N-1)*8)| length & offset (N-1)th translation | | | | |
| | | | | |
H | start hash table | | | | |
... ... | | | |
H + S * 4 | end hash table | | | | |
| | | | | |
| NUL terminated 0th string <----------------' | | |
| | | | |
| NUL terminated 1st string <------------------' | |
| | | |
... ... | |
| | | |
| NUL terminated 0th translation <---------------' |
| | |
| NUL terminated 1st translation <-----------------'
| |
... ...
| |
+------------------------------------------+
When GNU gettext will truly have reached is goal, average users
should feel some kind of astonished pleasure, seeing the effect of
that strange kind of magic that just makes their own native language
appear everywhere on their screens. As for naive users, they would
ideally have no special pleasure about it, merely taking their own
language for granted, and becoming rather unhappy otherwise.
So, let's try to describe here how we would like the magic to operate,
as we want the users' view to be the simplest, among all ways one
could look at GNU gettext. All other software engineers:
programmers, translators, maintainers, should work together in such a
way that the magic becomes possible. This is a long and progressive
undertaking, and information is available about the progress of the
Translation Project.
When a package is distributed, there are two kind of users:
installers who fetch the distribution, unpack it, configure
it, compile it and install it for themselves or others to use; and
end users that call programs of the package, once these have
been installed at their site. GNU gettext is offering magic
for both installers and end users.
Languages are not equally supported in all packages using GNU
gettext. To know if some package uses GNU gettext, one
may check the distribution for the `ABOUT-NLS' information file, for
some `ll.po' files, often kept together into some `po/'
directory, or for an `intl/' directory. Internationalized packages
have usually many `ll.po' files, where ll represents
the language. section Magic for End Users for a complete description of the format
for ll.
More generally, a matrix is available for showing the current state
of the Translation Project, listing which packages are prepared for
multi-lingual messages, and which languages is supported by each.
Because this information changes often, this matrix is not kept within
this GNU gettext manual. This information is often found in
file `ABOUT-NLS' from various distributions, but is also as old as
the distribution itself. A recent copy of this `ABOUT-NLS' file,
containing up-to-date information, should generally be found on the
Translation Project sites, and also on most GNU archive sites.
By default, packages fully using GNU gettext, internally,
are installed in such a way that they to allow translation of
messages. At configuration time, those packages should
automatically detect whether the underlying host system provides usable
catgets or gettext functions. If neither is present,
the GNU gettext library should be automatically prepared
and used. Installers may use special options at configuration
time for changing this behavior. The command `./configure
--with-included-gettext' bypasses system catgets or gettext to
use GNU gettext instead, while `./configure --disable-nls'
produces program totally unable to translate messages.
Internationalized packages have usually many `ll.po'
files. Unless
translations are disabled, all those available are installed together
with the package. However, the environment variable LINGUAS
may be set, prior to configuration, to limit the installed set.
LINGUAS should then contain a space separated list of two-letter
codes, stating which languages are allowed.
We consider here those packages using GNU gettext internally,
and for which the installers did not disable translation at
configure time. Then, users only have to set the LANG
environment variable to the appropriate `ll' prior to
using the programs in the package. See section The Current `ABOUT-NLS' Matrix. For example,
let's presume a German site. At the shell prompt, users merely have to
execute `setenv LANG de' (in csh) or `export
LANG; LANG=de' (in sh). They could even do this from their
`.login' or `.profile' file.
One aim of the current message catalog implementation provided by
GNU gettext was to use the systems message catalog handling, if the
installer wishes to do so. So we perhaps should first take a look at
the solutions we know about. The people in the POSIX committee does not
manage to agree on one of the semi-official standards which we'll
describe below. In fact they couldn't agree on anything, so nothing
decide only to include an example of an interface. The major Unix vendors
are split in the usage of the two most important specifications: X/Opens
catgets vs. Uniforums gettext interface. We'll describe them both and
later explain our solution of this dilemma.
catgets
The catgets implementation is defined in the X/Open Portability
Guide, Volume 3, XSI Supplementary Definitions, Chapter 5. But the
process of creating this standard seemed to be too slow for some of
the Unix vendors so they created their implementations on preliminary
versions of the standard. Of course this leads again to problems while
writing platform independent programs: even the usage of catgets
does not guarantee a unique interface.
Another, personal comment on this that only a bunch of committee members could have made this interface. They never really tried to program using this interface. It is a fast, memory-saving implementation, an user can happily live with it. But programmers hate it (at least me and some others do...)
But we must not forget one point: after all the trouble with transfering the rights on Unix(tm) they at last came to X/Open, the very same who published this specifications. This leads me to making the prediction that this interface will be in future Unix standards (e.g. Spec1170) and therefore part of all Unix implementation (implementations, which are allowed to wear this name).
The interface to the catgets implementation consists of three
functions which correspond to those used in file access: catopen
to open the catalog for using, catgets for accessing the message
tables, and catclose for closing after work is done. Prototypes
for the functions and the needed definitions are in the
<nl_types.h> header file.
catopen is used like in this:
nl_catd catd = catopen ("catalog_name", 0);
The function takes as the argument the name of the catalog. This usual
refers to the name of the program or the package. The second parameter
is not further specified in the standard. I don't even know whether it
is implemented consistently among various systems. So the common advice
is to use 0 as the value. The return value is a handle to the
message catalog, equivalent to handles to file returned by open.
This handle is of course used in the catgets function which can
be used like this:
char *translation = catgets (catd, set_no, msg_id, "original string");
The first parameter is this catalog descriptor. The second parameter
specifies the set of messages in this catalog, in which the message
described by msg_id is obtained. catgets therefore uses a
three-stage addressing:
catalog name => set number => message ID => translation
The fourth argument is not used to address the translation. It is given
as a default value in case when one of the addressing stages fail. One
important thing to remember is that although the return type of catgets
is char * the resulting string must not be changed. It
should better const char *, but the standard is published in
1988, one year before ANSI C.
The last of these function functions is used and behaves as expected:
catclose (catd);
After this no catgets call using the descriptor is legal anymore.
catgets Interface?!
Now that this descriptions seemed to be really easy where are the
problem we speak of. In fact the interface could be used in a
reasonable way, but constructing the message catalogs is a pain. The
reason for this lies in the third argument of catgets: the unique
message ID. This has to be a numeric value for all messages in a single
set. Perhaps you could imagine the problems keeping such list while
changing the source code. Add a new message here, remove one there. Of
course there have been developed a lot of tools helping to organize this
chaos but one as the other fails in one aspect or the other. We don't
want to say that the other approach has no problems but they are far
more easily to manage.
gettext
The definition of the gettext interface comes from a Uniforum
proposal and it is followed by at least one major Unix vendor
(Sun) in its last developments. It is not specified in any official
standard, though.
The main points about this solution is that it does not follow the method of normal file handling (open-use-close) and that it does not burden the programmer so many task, especially the unique key handling. Of course here is also a unique key needed, but this key is the message itself (how long or short it is). See section Comparing the Two Interfaces for a more detailed comparison of the two methods.
The following section contains a rather detailed description of the
interface. We make it that detailed because this is the interface
we chose for the GNU gettext Library. Programmers interested
in using this library will be interested in this description.
The minimal functionality an interface must have is a) to select a domain the strings are coming from (a single domain for all programs is not reasonable because its construction and maintenance is difficult, perhaps impossible) and b) to access a string in a selected domain.
This is principally the description of the gettext interface. It
has an global domain which unqualified usages reference. Of course this
domain is selectable by the user.
char *textdomain (const char *domain_name);
This provides the possibility to change or query the current status of
the current global domain of the LC_MESSAGE category. The
argument is a null-terminated string, whose characters must be legal in
the use in filenames. If the domain_name argument is NULL,
the function return the current value. If no value has been set
before, the name of the default domain is returned: messages.
Please note that although the return value of textdomain is of
type char * no changing is allowed. It is also important to know
that no checks of the availability are made. If the name is not
available you will see this by the fact that no translations are provided.
To use a domain set by textdomain the function
char *gettext (const char *msgid);
is to be used. This is the simplest reasonable form one can imagine.
The translation of the string msgid is returned if it is available
in the current domain. If not available the argument itself is
returned. If the argument is NULL the result is undefined.
One things which should come into mind is that no explicit dependency to
the used domain is given. The current value of the domain for the
LC_MESSAGES locale is used. If this changes between two
executions of the same gettext call in the program, both calls
reference a different message catalog.
For the easiest case, which is normally used in internationalized
packages, once at the beginning of execution a call to textdomain
is issued, setting the domain to a unique name, normally the package
name. In the following code all strings which have to be translated are
filtered through the gettext function. That's all, the package speaks
your language.
While this single name domain work good for most applications there
might be the need to get translations from more than one domain. Of
course one could switch between different domains with calls to
textdomain, but this is really not convenient nor is it fast. A
possible situation could be one case discussing while this writing: all
error messages of functions in the set of common used functions should
go into a separate domain error. By this mean we would only need
to translate them once.
For this reasons there are two more functions to retrieve strings:
char *dgettext (const char *domain_name, const char *msgid);
char *dcgettext (const char *domain_name, const char *msgid,
int category);
Both take an additional argument at the first place, which corresponds
to the argument of textdomain. The third argument of
dcgettext allows to use another locale but LC_MESSAGES.
But I really don't know where this can be useful. If the
domain_name is NULL or category has an value beside
the known ones, the result is undefined. It should also be noted that
this function is not part of the second known implementation of this
function family, the one found in Solaris.
A second ambiguity can arise by the fact, that perhaps more than one domain has the same name. This can be solved by specifying where the needed message catalog files can be found.
char *bindtextdomain (const char *domain_name,
const char *dir_name);
Calling this function binds the given domain to a file in the specified
directory (how this file is determined follows below). Especially a
file in the systems default place is not favored against the specified
file anymore (as it would be by solely using textdomain). A
NULL pointer for the dir_name parameter returns the binding
associated with domain_name. If domain_name itself is
NULL nothing happens and a NULL pointer is returned. Here
again as for all the other functions is true that none of the return
value must be changed!
It is important to remember that relative path names for the
dir_name parameter can be trouble. Since the path is always
computed relative to the current directory different results will be
achieved when the program executes a chdir command. Relative
paths should always be avoided to avoid dependencies and
unreliabilities.
Because many different languages for many different packages have to be
stored we need some way to add these information to file message catalog
files. The way usually used in Unix environments is have this encoding
in the file name. This is also done here. The directory name given in
bindtextdomains second argument (or the default directory),
followed by the value and name of the locale and the domain name are
concatenated:
dir_name/locale/LC_category/domain_name.mo
The default value for dir_name is system specific. For the GNU library, and for packages adhering to its conventions, it's:
/usr/local/share/locale
locale is the value of the locale whose name is this
LC_category. For gettext and dgettext this
locale is always LC_MESSAGES. dcgettext specifies the
locale by the third argument.(2) (3)
At this point of the discussion we should talk about an advantage of the
GNU gettext implementation. Some readers might have pointed out
that an internationalized program might have a poor performance if some
string has to be translated in an inner loop. While this is unavoidable
when the string varies from one run of the loop to the other it is
simply a waste of time when the string is always the same. Take the
following example:
{
while (...)
{
puts (gettext ("Hello world"));
}
}
When the locale selection does not change between two runs the resulting string is always the same. One way to use this is:
{
str = gettext ("Hello world");
while (...)
{
puts (str);
}
}
But this solution is not usable in all situation (e.g. when the locale selection changes) nor is it good readable.
The GNU C compiler, version 2.7 and above, provide another solution for this. To describe this we show here some lines of the `intl/libgettext.h' file. For an explanation of the expression command block see section `Statements and Declarations in Expressions' in The GNU CC Manual.
# if defined __GNUC__ && __GNUC__ == 2 && __GNUC_MINOR__ >= 7
extern int _nl_msg_cat_cntr;
# define dcgettext(domainname, msgid, category) \
(__extension__ \
({ \
char *result; \
if (__builtin_constant_p (msgid)) \
{ \
static char *__translation__; \
static int __catalog_counter__; \
if (! __translation__ \
|| __catalog_counter__ != _nl_msg_cat_cntr) \
{ \
__translation__ = \
dcgettext__ ((domainname), (msgid), (category)); \
__catalog_counter__ = _nl_msg_cat_cntr; \
} \
result = __translation__; \
} \
else \
result = dcgettext__ ((domainname), (msgid), (category)); \
result; \
}))
# endif
The interesting thing here is the __builtin_constant_p predicate.
This is evaluated at compile time and so optimization can take place
immediately. Here two cases are distinguished: the argument to
gettext is not a constant value in which case simply the function
dcgettext__ is called, the real implementation of the
dcgettext function.
If the string argument is constant we can reuse the once gained
translation when the locale selection has not changed. This is exactly
what is done here. The _nl_msg_cat_cntr variable is defined in
the `loadmsgcat.c' which is available in `libintl.a' and is
changed whenever a new message catalog is loaded.
The following discussion is perhaps a little bit colored. As said
above we implemented GNU gettext following the Uniforum
proposal and this surely has its reasons. But it should show how we
came to this decision.
First we take a look at the developing process. When we write an
application using NLS provided by gettext we proceed as always.
Only when we come to a string which might be seen by the users and thus
has to be translated we use gettext("...") instead of
"...". At the beginning of each source file (or in a central
header file) we define
#define gettext(String) (String)
Even this definition can be avoided when the system supports the
gettext function in its C library. When we compile this code the
result is the same as if no NLS code is used. When you take a look at
the GNU gettext code you will see that we use _("...")
instead of gettext("..."). This reduces the number of
additional characters per translatable string to 3 (in words:
three).
When now a production version of the program is needed we simply replace the definition
#define _(String) (String)
by
#include <libintl.h> #define _(String) gettext (String)
Additionally we run the program `xgettext' on all source code file which contain translatable strings and that's it: we have a running program which does not depend on translations to be available, but which can use any that becomes available.
The same procedure can be done for the gettext_noop invocations
(see section Special Cases of Translatable Strings). First you can define gettext_noop to a
no-op macro and later use the definition from `libintl.h'. Because
this name is not used in Suns implementation of `libintl.h',
you should consider the following code for your project:
#ifdef gettext_noop # define N_(String) gettext_noop (String) #else # define N_(String) (String) #endif
N_ is a short form similar to _. The `Makefile' in
the `po/' directory of GNU gettext knows by default both of the
mentioned short forms so you are invited to follow this proposal for
your own ease.
Now to catgets. The main problem is the work for the
programmer. Every time he comes to a translatable string he has to
define a number (or a symbolic constant) which has also be defined in
the message catalog file. He also has to take care for duplicate
entries, duplicate message IDs etc. If he wants to have the same
quality in the message catalog as the GNU gettext program
provides he also has to put the descriptive comments for the strings and
the location in all source code files in the message catalog. This is
nearly a Mission: Impossible.
But there are also some points people might call advantages speaking for
catgets. If you have a single word in a string and this string
is used in different contexts it is likely that in one or the other
language the word has different translations. Example:
printf ("%s: %d", gettext ("number"), number_of_errors)
printf ("you should see %d %s", number_count,
number_count == 1 ? gettext ("number") : gettext ("numbers"))
Here we have to translate two times the string "number". Even
if you do not speak a language beside English it might be possible to
recognize that the two words have a different meaning. In German the
first appearance has to be translated to "Anzahl" and the second
to "Zahl".
Now you can say that this example is really esoteric. And you are right! This is exactly how we felt about this problem and decide that it does not weight that much. The solution for the above problem could be very easy:
printf ("%s %d", gettext ("number:"), number_of_errors)
printf (number_count == 1 ? gettext ("you should see %d number")
: gettext ("you should see %d numbers"),
number_count)
We believe that we can solve all conflicts with this method. If it is difficult one can also consider changing one of the conflicting string a little bit. But it is not impossible to overcome.
Translator note: It is perhaps appropriate here to tell those English speaking programmers that the plural form of a noun cannot be formed by appending a single `s'. Most other languages use different methods. Even the above form is not general enough to cope with all languages. Rafal Maszkowski <[email protected]> reports:
In Polish we use e.g. plik (file) this way:
1 plik 2,3,4 pliki 5-21 pliko'w 22-24 pliki 25-31 pliko'wand so on (o' means 8859-2 oacute which should be rather okreska, similar to aogonek).
A workable approach might be to consider methods like the one used for
LC_TIME in the POSIX.2 standard. The value of the
alt_digits field can be up to 100 strings which represent the
numbers 1 to 100. Using this in a situation of an internationalized
program means that an array of translatable strings should be indexed by
the number which should represent. A small example:
void
print_month_info (int month)
{
const char *month_pos[12] =
{ N_("first"), N_("second"), N_("third"), N_("fourth"),
N_("fifth"), N_("sixth"), N_("seventh"), N_("eighth"),
N_("ninth"), N_("tenth"), N_("eleventh"), N_("twelfth") };
printf (_("%s is the %s month\n"), nl_langinfo (MON_1 + month),
_(month_pos[month]));
}
It should be obvious that this method is only reasonable for small ranges of numbers.
Starting with version 0.9.4 the library libintl.h should be
self-contained. I.e., you can use it in your own programs without
providing additional functions. The `Makefile' will put the header
and the library in directories selected using the $(prefix).
One exception of the above is found on HP-UX systems. Here the C library
does not contain the alloca function (and the HP compiler does
not generate it inlined). But it is not intended to rewrite the whole
library just because of this dumb system. Instead include the
alloca function in all package you use the libintl.a in.
gettext grok
To fully exploit the functionality of the GNU gettext library it
is surely helpful to read the source code. But for those who don't want
to spend that much time in reading the (sometimes complicated) code here
is a list comments:
gettext
function. The method which is presented here only works correctly
with the GNU implementation of the gettext functions. It is not
possible with underlying catgets functions or gettext
functions from the systems C library. The exception is of course the
GNU C Library which uses the GNU gettext Library for message handling.
In the function dcgettext at every call the current setting of
the highest priority environment variable is determined and used.
Highest priority means here the following list with decreasing
priority:
LANGUAGE
LC_ALL
LC_xxx, according to selected locale
LANG
LANGUAGE changes. According
to the process explained above the new value of this variable is found
as soon as the dcgettext function is called. But this also means
the (perhaps) different message catalog file is loaded. In other
words: the used language is changed.
But there is one little hook. The code for gcc-2.7.0 and up provides
some optimization. This optimization normally prevents the calling of
the dcgettext function as long as no new catalog is loaded. But
if dcgettext is not called the program also cannot find the
LANGUAGE variable be changed (see section Optimization of the *gettext functions). A
solution for this is very easy. Include the following code in the
language switching function.
/* Change language. */
setenv ("LANGUAGE", "fr", 1);
/* Make change known. */
{
extern int _nl_msg_cat_cntr;
++_nl_msg_cat_cntr;
}
The variable _nl_msg_cat_cntr is defined in `loadmsgcat.c'.
The programmer will find himself in need for a construct like this only
when developing programs which do run longer and provide the user to
select the language at runtime. Non-interactive programs (like all
these little Unix tools) should never need this.
There are two competing methods for language independent messages:
the X/Open catgets method, and the Uniforum gettext
method. The catgets method indexes messages by integers; the
gettext method indexes them by their English translations.
The catgets method has been around longer and is supported
by more vendors. The gettext method is supported by Sun,
and it has been heard that the COSE multi-vendor initiative is
supporting it. Neither method is a POSIX standard; the POSIX.1
committee had a lot of disagreement in this area.
Neither one is in the POSIX standard. There was much disagreement
in the POSIX.1 committee about using the gettext routines
vs. catgets (XPG). In the end the committee couldn't
agree on anything, so no messaging system was included as part
of the standard. I believe the informative annex of the standard
includes the XPG3 messaging interfaces, "...as an example of
a messaging system that has been implemented..."
They were very careful not to say anywhere that you should use one set of interfaces over the other. For more on this topic please see the Programming for Internationalization FAQ.
catgets
There have been a few discussions of late on the use of
catgets as a base. I think it important to present both
sides of the argument and hence am opting to play devil's advocate
for a little bit.
I'll not deny the fact that catgets could have been designed
a lot better. It currently has quite a number of limitations and
these have already been pointed out.
However there is a great deal to be said for consistency and standardization. A common recurring problem when writing Unix software is the myriad portability problems across Unix platforms. It seems as if every Unix vendor had a look at the operating system and found parts they could improve upon. Undoubtedly, these modifications are probably innovative and solve real problems. However, software developers have a hard time keeping up with all these changes across so many platforms.
And this has prompted the Unix vendors to begin to standardize their systems. Hence the impetus for Spec1170. Every major Unix vendor has committed to supporting this standard and every Unix software developer waits with glee the day they can write software to this standard and simply recompile (without having to use autoconf) across different platforms.
As I understand it, Spec1170 is roughly based upon version 4 of the
X/Open Portability Guidelines (XPG4). Because catgets and
friends are defined in XPG4, I'm led to believe that catgets
is a part of Spec1170 and hence will become a standardized component
of all Unix systems.
Now it seems kind of wasteful to me to have two different systems
installed for accessing message catalogs. If we do want to remedy
catgets deficiencies why don't we try to expand catgets
(in a compatible manner) rather than implement an entirely new system.
Otherwise, we'll end up with two message catalog access systems installed
with an operating system - one set of routines for packages using GNU
gettext for their internationalization, and another set of routines
(catgets) for all other software. Bloated?
Supposing another catalog access system is implemented. Which do
we recommend? At least for Linux, we need to attract as many
software developers as possible. Hence we need to make it as easy
for them to port their software as possible. Which means supporting
catgets. We will be implementing the glocale code
within our libc, but does this mean we also have to incorporate
another message catalog access scheme within our libc as well?
And what about people who are going to be using the glocale
+ non-catgets routines. When they port their software to
other platforms, they're now going to have to include the front-end
(glocale) code plus the back-end code (the non-catgets
access routines) with their software instead of just including the
glocale code with their software.
Message catalog support is however only the tip of the iceberg.
What about the data for the other locale categories. They also have
a number of deficiencies. Are we going to abandon them as well and
develop another duplicate set of routines (should glocale
expand beyond message catalog support)?
Like many parts of Unix that can be improved upon, we're stuck with balancing compatibility with the past with useful improvements and innovations for the future.
X/Open agreed very late on the standard form so that many implementations differ from the final form. Both of my system (old Linux catgets and Ultrix-4) have a strange variation.
OK. After incorporating the last changes I have to spend some time on
making the GNU/Linux libc gettext functions. So in future
Solaris is not the only system having gettext.
GNU is going international! The Translation Project is a way to get maintainers, translators and users all together, so GNU will gradually become able to speak many native languages.
The GNU gettext tool set contains everything maintainers
need for internationalizing their packages for messages. It also
contains quite useful tools for helping translators at localizing
messages to their native language, once a package has already been
internationalized.
To achieve the Translation Project, we need many interested people who like their own language and write it well, and who are also able to synergize with other translators speaking the same language. If you'd like to volunteer to work at translating messages, please send mail to your translating team.
Each team has its own mailing list, courtesy of Linux International. You may reach your translating team at the address `ll@li.org', replacing ll by the two-letter ISO 639 code for your language. Language codes are not the same as country codes given in ISO 3166. The following translating teams exist:
Chinese
zh, Czechcs, Danishda, Dutchnl, Esperantoeo, Finnishfi, Frenchfr, Irishga, Germande, Greekel, Italianit, Japaneseja, Indonesianin, Norwegianno, Polishpl, Portuguesept, Russianru, Spanishes, Swedishsvand Turkishtr.
For example, you may reach the Chinese translating team by writing to `[email protected]'. When you become a member of the translating team for your own language, you may subscribe to its list. For example, Swedish people can send a message to `[email protected]', having this message body:
subscribe
Keep in mind that team members should be interested in working at translations, or at solving translational difficulties, rather than merely lurking around. If your team does not exist yet and you want to start one, please write to `[email protected]'; you will then reach the GNU coordinator for all translator teams.
A handful of GNU packages have already been adapted and provided with message translations for several languages. Translation teams have begun to organize, using these packages as a starting point. But there are many more packages and many languages for which we have no volunteer translators. If you would like to volunteer to work at translating messages, please send mail to `[email protected]' indicating what language(s) you can work on.
This is now official, GNU is going international! Here is the announcement submitted for the January 1995 GNU Bulletin:
A handful of GNU packages have already been adapted and provided with message translations for several languages. Translation teams have begun to organize, using these packages as a starting point. But there are many more packages and many languages for which we have no volunteer translators. If you'd like to volunteer to work at translating messages, please send mail to `[email protected]' indicating what language(s) you can work on.
This document should answer many questions for those who are curious about the process or would like to contribute. Please at least skim over it, hoping to cut down a little of the high volume of e-mail generated by this collective effort towards GNU internationalization.
Most free programming which is widely shared is done in English, and currently, English is used as the main communicating language between national communities collaborating to the GNU project. This very document is written in English. This will not change in the foreseeable future.
However, there is a strong appetite from national communities for having more software able to write using national language and habits, and there is an on-going effort to modify GNU software in such a way that it becomes able to do so. The experiments driven so far raised an enthusiastic response from pretesters, so we believe that GNU internationalization is dedicated to succeed.
For suggestion clarifications, additions or corrections to this document, please e-mail to `[email protected]'.
Facing this internationalization effort, a few users expressed their concerns. Some of these doubts are presented and discussed, here.
gettext necessarily brings their package
under the protective wing of the GNU General Public License, when they
do not want to make their program free, or want other kinds of freedom.
The simplest answer is yes.
The mere marking of localizable strings in a package, or conditional
inclusion of a few lines for initialization, is not really including
GPL'ed code. However, the localization routines themselves are under
the GPL and would bring the remainder of the package under the GPL
if they were distributed with it. So, I presume that, for those
for which this is a problem, it could be circumvented by letting to
the end installers the burden of assembling a package prepared for
localization, but not providing the localization routines themselves.
On a larger scale, the true solution would be to organize some kind of fairly precise set up in which volunteers could participate. I gave some thought to this idea lately, and realize there will be some touchy points. I thought of writing to Richard Stallman to launch such a project, but feel it might be good to shake out the ideas between ourselves first. Most probably that Linux International has some experience in the field already, or would like to orchestrate the volunteer work, maybe. Food for thought, in any case!
I guess we have to setup something early, somehow, that will help many possible contributors of the same language to interlock and avoid work duplication, and further be put in contact for solving together problems particular to their tongue (in most languages, there are many difficulties peculiar to translating technical English). My Swedish contributor acknowledged these difficulties, and I'm well aware of them for French.
This is surely not a technical issue, but we should manage so the effort of locale contributors be maximally useful, despite the national team layer interface between contributors and maintainers.
The Translation Project needs some setup for coordinating language
coordinators. Localizing evolving programs will surely
become a permanent and continuous activity in the free software community,
once well started.
The setup should be minimally completed and tested before GNU
gettext becomes an official reality. The e-mail address
`[email protected]' has been setup for receiving
offers from volunteers and general e-mail on these topics. This address
reaches the Translation Project coordinator.
I also think GNU will need sooner than it thinks, that someone setup a way to organize and coordinate these groups. Some kind of group of groups. My opinion is that it would be good that GNU delegates this task to a small group of collaborating volunteers, shortly. Perhaps in `gnu.announce' a list of this national committee's can be published.
My role as coordinator would simply be to refer to Ulrich any German speaking volunteer interested to localization of free software packages, and maybe helping national groups to initially organize, while maintaining national registries for until national groups are ready to take over. In fact, the coordinator should ease volunteers to get in contact with one another for creating national teams, which should then select one coordinator per language, or country (regionalized language). If well done, the coordination should be useful without being an overwhelming task, the time to put delegations in place.
I suggest we look for volunteer coordinators/editors for individual languages. These people will scan contributions of translation files for various programs, for their own languages, and will ensure high and uniform standards of diction.
From my current experience with other people in these days, those who provide localizations are very enthusiastic about the process, and are more interested in the localization process than in the program they localize, and want to do many programs, not just one. This seems to confirm that having a coordinator/editor for each language is a good idea.
We need to choose someone who is good at writing clear and concise prose in the language in question. That is hard--we can't check it ourselves. So we need to ask a few people to judge each others' writing and select the one who is best.
I announce my prerelease to a few dozen people, and you would not believe all the discussions it generated already. I shudder to think what will happen when this will be launched, for true, officially, world wide. Who am I to arbitrate between two Czekolsovak users contradicting each other, for example?
I assume that your German is not much better than my French so that I would not be able to judge about these formulations. What I would suggest is that for each language there is a group for people who maintain the PO files and judge about changes. I suspect there will be cultural differences between how such groups of people will behave. Some will have relaxed ways, reach consensus easily, and have anyone of the group relate to the maintainers, while others will fight to death, organize heavy administrations up to national standards, and use strict channels.
The German team is putting out a good example. Right now, they are maybe half a dozen people revising translations of each other and discussing the linguistic issues. I do not even have all the names. Ulrich Drepper is taking care of coordinating the German team. He subscribed to all my pretest lists, so I do not even have to warn him specifically of incoming releases.
I'm sure, that is a good idea to get teams for each language working on translations. That will make the translations better and more consistent.
Taking French for example, there are a few sub-cultures around computers which developed diverging vocabularies. Picking volunteers here and there without addressing this problem in an organized way, soon in the project, might produce a distasteful mix of internationalized programs, and possibly trigger endless quarrels among those who really care.
Keeping some kind of unity in the way French localization of
internationalized programs is achieved is a difficult (and delicate) job.
Knowing the latin character of French people (:-), if we take this
the wrong way, we could end up nowhere, or spoil a lot of energies.
Maybe we should begin to address this problem seriously before
GNU gettext become officially published. And I suspect that this
means soon!
I expect the next big changes after the official release. Please note that I use the German translation of the short GPL message. We need to set a few good examples before the localization goes out for true in the free software community. Here are a few points to discuss:
If we get any inquiries about GNU gettext, send them on to:
`[email protected]'
The `*-pretest' lists are quite useful to me, maybe the idea could be generalized to many GNU, and non-GNU packages. But each maintainer his/her way!
Fran@,{c}ois, we have a mechanism in place here at `gnu.ai.mit.edu' to track teams, support mailing lists for them and log members. We have a slight preference that you use it. If this is OK with you, I can get you clued in.
Things are changing! A few years ago, when Daniel Fekete and I
asked for a mailing list for GNU localization, nested at the FSF, we
were politely invited to organize it anywhere else, and so did we.
For communicating with my pretesters, I later made a handful of
mailing lists located at iro.umontreal.ca and administrated by
majordomo. These lists have been very dependable
so far...
I suspect that the German team will organize itself a mailing list located in Germany, and so forth for other countries. But before they organize for true, it could surely be useful to offer mailing lists located at the FSF to each national team. So yes, please explain me how I should proceed to create and handle them.
We should create temporary mailing lists, one per country, to help people organize. Temporary, because once regrouped and structured, it would be fair the volunteers from country bring back their list in there and manage it as they want. My feeling is that, in the long run, each team should run its own list, from within their country. There also should be some central list to which all teams could subscribe as they see fit, as long as each team is represented in it.
There will surely be some discussion about this messages after the packages are finally released. If people now send you some proposals for better messages, how do you proceed? Jim, please note that right now, as I put forward nearly a dozen of localizable programs, I receive both the translations and the coordination concerns about them.
If I put one of my things to pretest, Ulrich receives the announcement and passes it on to the German team, who make last minute revisions. Then he submits the translation files to me as the maintainer. For free packages I do not maintain, I would not even hear about it. This scheme could be made to work for the whole Translation Project, I think. For security reasons, maybe Ulrich (national coordinators, in fact) should update central registry kept at the Translation Project (Jim, me, or Len's recruits) once in a while.
In December/January, I was aggressively ready to internationalize all of GNU, giving myself the duty of one small GNU package per week or so, taking many weeks or months for bigger packages. But it does not work this way. I first did all the things I'm responsible for. I've nothing against some missionary work on other maintainers, but I'm also loosing a lot of energy over it--same debates over again.
And when the first localized packages are released we'll get a lot of responses about ugly translations :-). Surely, and we need to have beforehand a fairly good idea about how to handle the information flow between the national teams and the package maintainers.
Please start saving somewhere a quick history of each PO file. I know for sure that the file format will change, allowing for comments. It would be nice that each file has a kind of log, and references for those who want to submit comments or gripes, or otherwise contribute. I sent a proposal for a fast and flexible format, but it is not receiving acceptance yet by the GNU deciders. I'll tell you when I have more information about this.
The maintainer of a package has many responsibilities. One of them is ensuring that the package will install easily on many platforms, and that the magic we described earlier (see section The User's View) will work for installers and end users.
Of course, there are many possible ways by which GNU gettext
might be integrated in a distribution, and this chapter does not cover
them in all generality. Instead, it details one possible approach which
is especially adequate for many free software distributions following GNU
standards, or even better, Gnits standards, because GNU gettext
is purposely for helping the internationalization of the whole GNU
project, and as many other good free packages as possible. So, the
maintainer's view presented here presumes that the package already has
a `configure.in' file and uses GNU Autoconf.
Nevertheless, GNU gettext may surely be useful for free packages
not following GNU standards and conventions, but the maintainers of such
packages might have to show imagination and initiative in organizing
their distributions so gettext work for them in all situations.
There are surely many, out there.
Even if gettext methods are now stabilizing, slight adjustments
might be needed between successive gettext versions, so you
should ideally revise this chapter in subsequent releases, looking
for changes.
Some free software packages are distributed as tar files which unpack
in a single directory, these are said to be flat distributions.
Other free software packages have a one level hierarchy of subdirectories, using
for example a subdirectory named `doc/' for the Texinfo manual and
man pages, another called `lib/' for holding functions meant to
replace or complement C libraries, and a subdirectory `src/' for
holding the proper sources for the package. These other distributions
are said to be non-flat.
For now, we cannot say much about flat distributions. A flat
directory structure has the disadvantage of increasing the difficulty
of updating to a new version of GNU gettext. Also, if you have
many PO files, this could somewhat pollute your single directory.
In the GNU gettext distribution, the `misc/' directory
contains a shell script named `combine-sh'. That script may
be used for combining all the C files of the `intl/' directory
into a pair of C files (one `.c' and one `.h'). Those two
generated files would fit more easily in a flat directory structure,
and you will then have to add these two files to your project.
Maybe because GNU gettext itself has a non-flat structure,
we have more experience with this approach, and this is what will be
described in the remaining of this chapter. Some maintainers might
use this as an opportunity to unflatten their package structure.
Only later, once gained more experience adapting GNU gettext
to flat distributions, we might add some notes about how to proceed
in flat situations.
There are some works which are required for using GNU gettext
in one of your package. These works have some kind of generality
that escape the point by point descriptions used in the remainder
of this chapter. So, we describe them here.
m4, GNU Autoconf and GNU
gettext are already installed at your site, and if not, proceed
to do this first. If you got to install these things, beware that
GNU m4 must be fully installed before GNU Autoconf is even
configured.
To further ease the task of a package maintainer the automake
package was designed and implemented. GNU gettext now uses this
tool and the `Makefile's in the `intl/' and `po/'
therefore know about all the goals necessary for using automake
and `libintl' in one project.
Those four packages are only needed to you, as a maintainer; the
installers of your own package and end users do not really need any of
GNU m4, GNU Autoconf, GNU gettext, or GNU automake
for successfully installing and running your package, with messages
properly translated. But this is not completely true if you provide
internationalized shell scripts within your own package: GNU
gettext shall then be installed at the user site if the end users
want to see the translation of shell script messages.
It is worth adding here a few words about how the maintainer should ideally behave with PO files submissions. As a maintainer, your role is to authentify the origin of the submission as being the representative of the appropriate translating teams of the Translation Project (forward the submission to `[email protected]' in case of doubt), to ensure that the PO file format is not severely broken and does not prevent successful installation, and for the rest, to merely to put these PO files in `po/' for distribution.
As a maintainer, you do not have to take on your shoulders the responsibility of checking if the translations are adequate or complete, and should avoid diving into linguistic matters. Translation teams drive themselves and are fully responsible of their linguistic choices for the Translation Project. Keep in mind that translator teams are not driven by maintainers. You can help by carefully redirecting all communications and reports from users about linguistic matters to the appropriate translation team, or explain users how to reach or join their team. The simplest might be to send them the `ABOUT-NLS' file.
Maintainers should never ever apply PO file bug reports themselves, short-cutting translation teams. If some translator has difficulty to get some of her points through her team, it should not be an issue for her to directly negotiate translations with maintainers. Teams ought to settle their problems themselves, if any. If you, as a maintainer, ever think there is a real problem with a team, please never try to solve a team's problem on your own.
gettextize Program
Some files are consistently and identically needed in every package
internationalized through GNU gettext. As a matter of
convenience, the gettextize program puts all these files right
in your package. This program has the following synopsis:
gettextize [ option... ] [ directory ]
and accepts the following options:
gettext code
available on the system, but it might disturb some mechanism the
maintainer is used to apply to the sources. Because running
gettextize is easy there shouldn't be problems with using copies.
If directory is given, this is the top level directory of a
package to prepare for using GNU gettext. If not given, it
is assumed that the current directory is the top level directory of
such a package.
The program gettextize provides the following files. However,
no existing file will be replaced unless the option --force
(-f) is specified.
gettextize,
if you have one handy. You may also fetch a more recent copy of file
`ABOUT-NLS' from Translation Project sites, and from most GNU
archive sites.
gettext distribution.
(beware the double `.in' in the file name). If the `po/'
directory already exists, it will be preserved along with the files
it contains, and only `Makefile.in.in' will be overwritten.
gettext
distribution. Also, if option --force (-f) is given,
the `intl/' directory is emptied first.
If your site support symbolic links, gettextize will not
actually copy the files into your package, but establish symbolic
links instead. This avoids duplicating the disk space needed in
all packages. Merely using the `-h' option while creating the
tar archive of your distribution will resolve each link by an
actual copy in the distribution archive. So, to insist, you really
should use `-h' option with tar within your dist
goal of your main `Makefile.in'.
It is interesting to understand that most new files for supporting
GNU gettext facilities in one package go in `intl/'
and `po/' subdirectories. One distinction between these two
directories is that `intl/' is meant to be completely identical
in all packages using GNU gettext, while all newly created
files, which have to be different, go into `po/'. There is a
common `Makefile.in.in' in `po/', because the `po/'
directory needs its own `Makefile', and it has been designed so
it can be identical in all packages.
Besides files which are automatically added through gettextize,
there are many files needing revision for properly interacting with
GNU gettext. If you are closely following GNU standards for
Makefile engineering and auto-configuration, the adaptations should
be easier to achieve. Here is a point by point description of the
changes needed in each.
So, here comes a list of files, each one followed by a description of
all alterations it needs. Many examples are taken out from the GNU
gettext 0.10.35 distribution itself. You may indeed
refer to the source code of the GNU gettext package, as it
is intended to be a good example and master implementation for using
its own functionality.
The `po/' directory should receive a file named `POTFILES.in'. This file tells which files, among all program sources, have marked strings needing translation. Here is an example of such a file:
# List of source files containing translatable strings. # Copyright (C) 1995 Free Software Foundation, Inc. # Common library files lib/error.c lib/getopt.c lib/xmalloc.c # Package source files src/gettextp.c src/msgfmt.c src/xgettext.c
Dashed comments and white lines are ignored. All other lines list those source files containing strings marked for translation (see section How Marks Appears in Sources), in a notation relative to the top level of your whole distribution, rather than the location of the `POTFILES.in' file itself.
PACKAGE=gettext VERSION=0.10.35 AC_DEFINE_UNQUOTED(PACKAGE, "$PACKAGE") AC_DEFINE_UNQUOTED(VERSION, "$VERSION") AC_SUBST(PACKAGE) AC_SUBST(VERSION)Of course, you replace `gettext' with the name of your package, and `0.10.35' by its version numbers, exactly as they should appear in the packaged
tar file name of your distribution
(`gettext-0.10.35.tar.gz', here).
ALL_LINGUAS to the white separated,
quoted list of available languages, in a single line, like this:
ALL_LINGUAS="de fr"This example means that German and French PO files are available, so that these languages are currently supported by your package. If you want to further restrict, at installation time, the set of installed languages, this should not be done by modifying
ALL_LINGUAS in
`configure.in', but rather by using the LINGUAS environment
variable (see section Magic for Installers).
m4 macro for triggering internationalization
support. Just add this line to `configure.in':
AM_GNU_GETTEXTThis call is purposely simple, even if it generates a lot of configure time checking and actions.
AC_OUTPUT directive, at the end of your `configure.in'
file, needs to be modified in two ways:
AC_OUTPUT([existing configuration files intl/Makefile po/Makefile.in], existing additional actions])The modification to the first argument to
AC_OUTPUT asks
for substitution in the `intl/' and `po/' directories.
Note the `.in' suffix used for `po/' only. This is because
the distributed file is really `po/Makefile.in.in'.
If you do not have an `aclocal.m4' file in your distribution,
the simplest is taking a copy of `aclocal.m4' from
GNU gettext. But to be precise, you only need macros
AM_LC_MESSAGES, AM_WITH_NLS and AM_GNU_GETTEXT,
and AM_PATH_PROG_WITH_TEST, which is called by AM_WITH_NLS,
so you may use an editor and remove macros you do not need.
If you already have an `aclocal.m4' file, then you will have
to merge the said macros into your `aclocal.m4'. Note that if
you are upgrading from a previous release of GNU gettext, you
should most probably replace the said macros, as they usually
change a little from one release of GNU gettext to the next.
Their contents may vary as we get more experience with strange systems
out there.
These macros check for the internationalization support functions
and related informations. Hopefully, once stabilized, these macros
might be integrated in the standard Autoconf set, because this
piece of m4 code will be the same for all projects using GNU
gettext.
If you do not have an `acconfig.h' file in your distribution, the
simplest is use take a copy of `acconfig.h' from GNU
gettext. But to be precise, you only need the lines and comments
for ENABLE_NLS, HAVE_CATGETS, HAVE_GETTEXT and
HAVE_LC_MESSAGES, HAVE_STPCPY, PACKAGE and
VERSION, so you may use an editor and remove everything else. If
you already have an `acconfig.h' file, then you should merge the
said definitions into your `acconfig.h'.
Here are a few modifications you need to make to your main, top-level `Makefile.in' file.
PACKAGE = @PACKAGE@ VERSION = @VERSION@
DISTFILES definition, so the file gets
distributed.
SUBDIRS in Makefile.in for it
to be further used in the `dist:' goal.
SUBDIRS = doc lib @INTLSUB@ src @POSUB@that you will have to adapt to your own package.
distdir = $(PACKAGE)-$(VERSION) dist: Makefile rm -fr $(distdir) mkdir $(distdir) chmod 777 $(distdir) for file in $(DISTFILES); do \ ln $$file $(distdir) 2>/dev/null || cp -p $$file $(distdir); \ done for subdir in $(SUBDIRS); do \ mkdir $(distdir)/$$subdir || exit 1; \ chmod 777 $(distdir)/$$subdir; \ (cd $$subdir && $(MAKE) $@) || exit 1; \ done tar chozf $(distdir).tar.gz $(distdir) rm -fr $(distdir)
Some of the modifications made in the main `Makefile.in' will also be needed in the `Makefile.in' from your package sources, which we assume here to be in the `src/' subdirectory. Here are all the modifications needed in `src/Makefile.in':
PACKAGE = @PACKAGE@ VERSION = @VERSION@
top_srcdir
gets defined. This will serve for cpp include files. Just add
the line:
top_srcdir = @top_srcdir@
subdir as `src', later
allowing for almost uniform `dist:' goals in all your
`Makefile.in'. At list, the `dist:' goal below assume that
you used:
subdir = src
@INTLLIBS@ as
a library. An easy way to achieve this is to manage that it gets into
LIBS, like this:
LIBS = @INTLLIBS@ @LIBS@In most packages internationalized with GNU
gettext, one will
find a directory `lib/' in which a library containing some helper
functions will be build. (You need at least the few functions which the
GNU gettext Library itself needs.) However some of the functions
in the `lib/' also give messages to the user which of course should be
translated, too. Taking care of this it is not enough to place the support
library (say `libsupport.a') just between the @INTLLIBS@
and @LIBS@ in the above example. Instead one has to write this:
LIBS = ../lib/libsupport.a @INTLLIBS@ ../lib/libsupport.a @LIBS@
distdir = ../$(PACKAGE)-$(VERSION)/$(subdir) dist: Makefile $(DISTFILES) for file in $(DISTFILES); do \ ln $$file $(distdir) 2>/dev/null || cp -p $$file $(distdir); \ done
We would like to conclude this GNU gettext manual by presenting
an history of the Translation Project so far. We finally give
a few pointers for those who want to do further research or readings
about Native Language Support matters.
gettext
Internationalization concerns and algorithms have been informally
and casually discussed for years in GNU, sometimes around GNU
libc, maybe around the incoming Hurd, or otherwise
(nobody clearly remembers). And even then, when the work started for
real, this was somewhat independently of these previous discussions.
This all began in July 1994, when Patrick D'Cruze had the idea and
initiative of internationalizing version 3.9.2 of GNU fileutils.
He then asked Jim Meyering, the maintainer, how to get those changes
folded into an official release. That first draft was full of
#ifdefs and somewhat disconcerting, and Jim wanted to find
nicer ways. Patrick and Jim shared some tries and experimentations
in this area. Then, feeling that this might eventually have a deeper
impact on GNU, Jim wanted to know what standards were, and contacted
Richard Stallman, who very quickly and verbally described an overall
design for what was meant to become glocale, at that time.
Jim implemented glocale and got a lot of exhausting feedback
from Patrick and Richard, of course, but also from Mitchum DSouza
(who wrote a catgets-like package), Roland McGrath, maybe David
MacKenzie, Fran@,{c}ois Pinard, and Paul Eggert, all pushing and
pulling in various directions, not always compatible, to the extent
that after a couple of test releases, glocale was torn apart.
While Jim took some distance and time and became dad for a second
time, Roland wanted to get GNU libc internationalized, and
got Ulrich Drepper involved in that project. Instead of starting
from glocale, Ulrich rewrote something from scratch, but
more conformant to the set of guidelines who emerged out of the
glocale effort. Then, Ulrich got people from the previous
forum to involve themselves into this new project, and the switch
from glocale to what was first named msgutils, renamed
nlsutils, and later gettext, became officially accepted
by Richard in May 1995 or so.
Let's summarize by saying that Ulrich Drepper wrote GNU gettext
in April 1995. The first official release of the package, including
PO mode, occurred in July 1995, and was numbered 0.7. Other people
contributed to the effort by providing a discussion forum around
Ulrich, writing little pieces of code, or testing. These are quoted
in the THANKS file which comes with the GNU gettext
distribution.
While this was being done, Fran@,{c}ois adapted half a dozen of
GNU packages to glocale first, then later to gettext,
putting them in pretest, so providing along the way an effective
user environment for fine tuning the evolving tools. He also took
the responsibility of organizing and coordinating the Translation
Project. After nearly a year of informal exchanges between people from
many countries, translator teams started to exist in May 1995, through
the creation and support by Patrick D'Cruze of twenty unmoderated
mailing lists for that many native languages, and two moderated
lists: one for reaching all teams at once, the other for reaching
all willing maintainers of internationalized free software packages.
Fran@,{c}ois also wrote PO mode in June 1995 with the collaboration
of Greg McGary, as a kind of contribution to Ulrich's package.
He also gave a hand with the GNU gettext Texinfo manual.
Eugene H. Dorr (`[email protected]') maintains an interesting bibliography on internationalization matters, called Internationalization Reference List, which is available as:
ftp://ftp.ora.com/pub/examples/nutshell/ujip/doc/i18n-books.txt
Michael Gschwind (`[email protected]') maintains a Frequently Asked Questions (FAQ) list, entitled Programming for Internationalisation. This FAQ discusses writing programs which can handle different language conventions, character sets, etc.; and is applicable to all character set encodings, with particular emphasis on ISO 8859-1. It is regularly published in Usenet groups `comp.unix.questions', `comp.std.internat', `comp.software.international', `comp.lang.c', `comp.windows.x', `comp.std.c', `comp.answers' and `news.answers'. The home location of this document is:
ftp://ftp.vlsivie.tuwien.ac.at/pub/8bit/ISO-programming
Patrick D'Cruze (`[email protected]') wrote a tutorial about NLS matters, and Jochen Hein (`[email protected]') took over the responsibility of maintaining it. It may be found as:
ftp://sunsite.unc.edu/pub/Linux/utils/nls/catalogs/Incoming/...
...locale-tutorial-0.8.txt.gz
This site is mirrored in:
ftp://ftp.ibp.fr/pub/linux/sunsite/
A French version of the same tutorial should be findable at:
ftp://ftp.ibp.fr/pub/linux/french/docs/
together with French translations of many Linux-related documents.
The ISO 639 standard defines two character codes for many countries. All abreviations for countries or languages used in the Translation Project should come from this standard.
This
limitation is not imposed by GNU gettext, but comes from the
msgfmt implementation on Solaris.
Some
system, eg Ultrix, don't have LC_MESSAGES. Here we use a more or
less arbitrary value for it.
When the system does not support
setlocale its behavior in setting the locale values is simulated
by looking at the environment variables.
This document was generated on 21 January 2000 using the texi2html translator version 1.54.