8. Producing Binary MO Files

8.1 Invoking the msgfmt Program
8.2 Invoking the msgunfmt Program
8.3 The Format of GNU MO Files

8.1 Invoking the msgfmt Program

msgfmt [option] filename.po ...

The msgfmt programs generates a binary message catalog from a textual translation description.

8.1.1 Input file location

`filename.po ...'

`-D directory'

`--directory=directory'

Add directory to the list of directories. Source files are searched relative to this list of directories. The resulting `.po' file will be written relative to the current directory, though.

If an input file is `-', standard input is read.

8.1.2 Operation mode

`-j'

`--java'

Java mode: generate a Java ResourceBundle class.

`--java2'

Like --java, and assume Java2 (JDK 1.2 or higher).

`--tcl'

Tcl mode: generate a tcl/msgcat `.msg' file.

8.1.3 Output file location

`-o file'

`--output-file=file'

Write output to specified file.

`--strict'

Direct the program to work strictly following the Uniforum/Sun implementation. Currently this only affects the naming of the output file. If this option is not given the name of the output file is the same as the domain name. If the strict Uniforum mode is enabled the suffix `.mo' is added to the file name if it is not already present.

We find this behaviour of Sun's implementation rather silly and so by default this mode is not selected.

If the output file is `-', output is written to standard output.

8.1.4 Output file location in Java mode

`-r resource'

`--resource=resource'

Specify the resource name.

`-l locale'

`--locale=locale'

Specify the locale name, either a language specification of the form ll or a combined language and country specification of the form ll_CC.

`-d directory'

Specify the base directory of classes directory hierarchy.

The class name is determined by appending the locale name to the resource name, separated with an underscore. The `-d' option is mandatory. The class is written under the specified directory.

8.1.5 Output file location in Tcl mode

`-l locale'

`--locale=locale'

Specify the locale name, either a language specification of the form ll or a combined language and country specification of the form ll_CC.

`-d directory'

Specify the base directory of `.msg' message catalogs.

The `-l' and `-d' options are mandatory. The `.msg' file is written in the specified directory.

8.1.6 Input file interpretation

`-c'

`--check'

Perform all the checks implied by --check-format, --check-header, --check-domain.

`--check-format'

Check language dependent format strings.

If the string represents a format string 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.

To solve this problem the programmer can dictate the decision to the xgettext program (see section 3.5 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.

`--check-header'

Verify presence and contents of the header entry. See section 5.2 Filling in the Header Entry, for a description of the various fields in the header entry.

`--check-domain'

Check for conflicts between domain directives and the --output-file option

`-C'

`--check-compatibility'

Check that GNU msgfmt behaves like X/Open msgfmt. This will give an error when attempting to use the GNU extensions.

`--check-accelerators[=char]'

Check presence of keyboard accelerators for menu items. This is based on the convention used in some GUIs that a keyboard accelerator in a menu item string is designated by an immediately preceding `&' character. Sometimes a keyboard accelerator is also called "keyboard mnemonic". This check verifies that if the untranslated string has exactly one `&' character, the translated string has exactly one `&' as well. If this option is given with a char argument, this char should be a non-alphanumeric character and is used as keyboard acceleator mark instead of `&'.

`-f'

`--use-fuzzy'

Use fuzzy entries in output. Note that using this option is usually wrong, because fuzzy messages are exactly those which have not been validated by a human translator.

8.1.7 Output details

`-a number'

`--alignment=number'

Align strings to number bytes (default: 1).

`--no-hash'

Don't include a hash table in the binary file. Lookup will be more expensive at run time (binary search instead of hash table lookup).

8.1.8 Informative output

`-h'

`--help'

Display this help and exit.

`-V'

`--version'

Output version information and exit.

`--statistics'

Print statistics about translations.

`-v'

`--verbose'

Increase verbosity level.

8.2 Invoking the msgunfmt Program

msgunfmt [option] [file]...

The msgunfmt program converts a binary message catalog to a Uniforum style .po file.

8.2.1 Operation mode

`-j'

`--java'

Java mode: input is a Java ResourceBundle class.

`--tcl'

Tcl mode: input is a tcl/msgcat `.msg' file.

8.2.2 Input file location

`file ...'

Input .mo files.

If no input file is given or if it is `-', standard input is read.

8.2.3 Input file location in Java mode

`-r resource'

`--resource=resource'

Specify the resource name.

`-l locale'

`--locale=locale'

Specify the locale name, either a language specification of the form ll or a combined language and country specification of the form ll_CC.

The class name is determined by appending the locale name to the resource name, separated with an underscore. The class is located using the CLASSPATH.

8.2.4 Input file location in Tcl mode

`-l locale'

`--locale=locale'

Specify the locale name, either a language specification of the form ll or a combined language and country specification of the form ll_CC.

`-d directory'

Specify the base directory of `.msg' message catalogs.

The `-l' and `-d' options are mandatory. The `.msg' file is located in the specified directory.

8.2.5 Output file location

`-o file'

`--output-file=file'

Write output to specified file.

The results are written to standard output if no output file is specified or if it is `-'.

8.2.6 Output details

`--force-po'

Always write an output file even if it contains no message.

`-i'

`--indent'

Write the .po file using indented style.

`--strict'

Write out a strict Uniforum conforming PO file. Note that this Uniforum format should be avoided because it doesn't support the GNU extensions.

`-w number'

`--width=number'

Set the output page width. Long strings in the output files will be split across multiple lines in order to ensure that each line's width (= number of screen columns) is less or equal to the given number.

`--no-wrap'

Do not break long message lines. Message lines whose width exceeds the output page width will not be split into several lines. Only file reference lines which are wider than the output page width will be split.

`-s'

`--sort-output'

Generate sorted output. Note that using this option makes it much harder for the translator to understand each message's context.

8.2.7 Informative output

`-h'

`--help'

Display this help and exit.

`-V'

`--version'

Output version information and exit.

`-v'

`--verbose'

Increase verbosity level.

8.3 The Format of GNU MO Files

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 on 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.

Plural forms are stored by letting the plural of the original string follow the singular of the original string, separated through a NUL byte. The length which appears in the string descriptor includes both. However, only the singular of the original string takes part in the hash table lookup. The plural variants of the translation are all stored consecutively, separated through a NUL byte. Here also, the length in the string descriptor includes all of them.

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 the 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. (No, we don't want to have wide characters in MO files. They would make the file unnecessarily large, and the `wchar_t' type being platform dependent, MO files would be platform dependent as well.)

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 <-----------------' | | ... ... | | +------------------------------------------+