Skip to main content

Translation toolset

Project description

What is lingua?

Lingua is a package with tools to extract translatable texts from your code, and to check existing translations. It replaces the use of the xgettext command from gettext, or pybabel from Babel.

Message extraction

The simplest way to extract all translatable messages is to point the pot-create tool at the root of your source tree.

$ pot-create src

This will create a messages.pot file containing all found messages.

Specifying input files

There are three ways to tell lingua which files you want it to scan:

  1. Specify filenames directly on the command line. For example:

    $ pot-create
  2. Specify a directory on the command line. Lingua will recursively scan that directory for all files it knows how to handle.

    $ pot-create src
  3. Use the --files-from parameter to point to a file with a list of files to scan. Lines starting with # and empty lines will be ignored.

    $ pot-create

You can also use the --directory=PATH parameter to add the given path to the list of directories to check for files. This may sound confusing, but can be useful. For example this command will look for and in the current directory, and if they are not found there in the ../src directory:

$ pot-create --directory=../src


In its default configuration lingua will use its python extractor for .py files, its XML extractor for .pt and .zpt files and its ZCML extractor for .zcml files. If you use different extensions you setup a configuration file which tells lingua how to process files. This file uses a simple ini-style format.

There are two types of configuration that can be set in the configuration file: which extractor to use for a file extension, and the configuration for a single extractor.

File extensions are configured in the extensions section. Each entry in this section maps a file extension to an extractor name. For example to tell lingua to use its XML extractor for files with a .html extension you can use this configuration:

.html = xml

To find out which extractors are available use the -list-extractors option.

$ bin/pot-create --list-extractors
chameleon         Chameleon templates (defaults to Python expressions)
python            Python sources
xml               Chameleon templates (defaults to Python expressions)
zcml              Zope Configuration Markup Language (ZCML)
zope              Zope templates (defaults to TALES expressions)

A section named extractor:<name> can be used to configure a specific extractor. For example to tell the XML extractor that the default language used for expressions is TALES instead of Python:

default-engine = tales

Either place a global configuration file named .config/lingua to your home folder or use the --config option to point lingua to your configuration file.

$ pot-create -c lingua.cfg src

Domain filtering

When working with large systems you may use multiple translation domains in a single source tree. Lingua can support that by filtering messages by domain when scanning sources. To enable domain filtering use the -d option:

$ pot-create -d mydomain src

Lingua will always include messages for which it can not determine the domain. For example, take this Python code:

print(gettext(u'Hello, World'))
print(dgettext('mydomain', u'Bye bye'))

The first hello-message does not specify its domain and will always be included. The second line uses dgettext to explicitly specify the domain. Lingua will use this information when filtering domains.

Including comments

You can add comments to messages to help translators, for example to explain how a text is used, or provide hints on how it should be translated. For chameleon templates this can be done using the i18n:comment attribute:

<label i18n:comment="This is a form label" i18n:translate="">Password</label>

Comments are inherited, so you can put them on a parent element as well.

<form i18n:comment="This is used in the password reset form">
  <label i18n:translate="">Password</label>
  <button i18n:translate="">Change</button>

For Python code you can tell lingua to include comments by using the --add-comments option. This will make Linua include all comments on the line(s) immediately preceeding (there may be no empty line in between) a translation call.

# This text should address the user directly.
return _('Thank you for using our service.')

Alternatively you can also put a comment at the end of the line starting your translation function call.

return _('Thank you for using our service.')  # Address the user directly

If you do not want all comments to be included but only specific ones you can add a keyword to the --add-comments option, for example --add-comments=I18N.

# I18N This text should address the user directly, and use formal addressing.
return _('Thank you for using our service')

Setting message flags in comments

Messages can have flags. These are to indicate what format a message has, and are typically used by validation tools to check if a translation does not break variable references or template syntax. Lingua does a reasonable job to detect strings using C and Python formatting, but sometimes you may need to set flags yourself. This can be done with a [flag, flag] marker in a comment.

# I18N [markdown,c-format]
header =  _(u'# Hello *%s*')

Specifying keywords

When looking for messages a lingua parser uses a default list of keywords to identify translation calls. You can add extra keywords via the --keyword option. If you have your own mygettext function which takes a string to translate as its first parameter you can use this:

$ pot-create --keyword=mygettext

If your function takes more parameters you will need to tell lingua about them. This can be done in several ways:

  • If the translatable text is not the first parameter you can specify the parameter number with <keyword>:<parameter number>. For example if you use i18n_log(level, msg) the keyword specifier would be i18n_log:2
  • If you support plurals you can specify the parameter used for the plural message by specifying the parameter number for both the singular and plural text. For example if your function signature is show_result(single, plural) the keyword specifier is show_result:1,2
  • If you use message contexts you can specify the parameter used for the context by adding a c to the parameter number. For example the keyword specifier for pgettext is pgettext:1c,2.
  • If your function takes the domain as a parameter you can specify which parameter is used for the domain by adding a d to the parameter number. For example the keyword specifier for dgettext is dgettext:1d,2. This is a lingua-specified extension.
  • You can specify the exact number of parameters a function call must have using the t postfix. For example if a function must have four parameters to be a valid call, the specifier could be myfunc:1,4t.


Lingua includes a number of extractors:

  • python: handles Python source code.
  • chameleon: handles Chameleon files, using the Zope i18n syntax
  • zcml: handles Zope Configuration Markup Language (ZCML) files.
  • zope: a variant of the chameleon extractor, which assumes the default
    expression language is TALES instead of Python.
  • xml: old name for the chameleon extractor. This name should not be used anymore and is only supported for backwards compatibility.

Babel extractors

There are several packages with plugins for Babel’s message extraction tool. Lingua can use those plugins as well. The plugin names will be prefixed with babel- to distinguish them from lingua extractors.

For example, if you have the PyBabel-json package installed you can instruct lingua to use it for .json files by adding this to your configuration file:

.json = babel-json

Some Babel plugins require you to specify comment tags. This can be set with the comment-tags option.

comment-tags = TRANSLATOR:

Comparison to other tools

Differences compared to GNU gettext:

  • Support for file formats such as Zope Page Templates (popular in Pyramid, Chameleon, Plone and Zope).
  • Better support for detecting format strings used in Python.
  • No direct support for C, C++, Perl, and many other languages. Lingua focuses on languages commonly used in Python projects, although support for other languages can be added via plugins.

Differences compared to Babel:

  • More reliable detection of Python format strings.
  • Lingua includes plural support.
  • Support for only extracting texts for a given translation domain. This is often useful for extensible software where you use multiple translation domains in a single application.

Validating translations

Lingua includes a simple polint tool which performs a few basic checks on PO files. Currently implemented tests are:

  • duplicated message ids (can also be checked with GNU gettext’s msgfmt). These should never happen and are usually a result of a bug in the message extraction logic.
  • identical translations used for multiple canonical texts. This can happen for valid reasons, for example when the original text is not spelled consistently.

To check a po file simply run polint with the po file as argument:

$ polint nl.po

    ${val} ist keine Zeichenkette
Used for 2 canonical texts:
1       ${val} is not a string
2       "${val}" is not a string

Writing custom extractors

First we need to create the custom extractor:

from lingua.extractors import Extractor
from lingua.extractors import Message

class MyExtractor(Extractor):
    '''One-line description for --list-extractors'''
    extensions = ['.txt']

    def __call__(self, filename, options):
        return [Message(None, 'msgid', None, [], u'', u'', (filename, 1))]

Hooking up extractors to lingua is done by lingua.extractors entry points in

      my_extractor = mypackage.extractor:MyExtractor

Note - the registered extractor must be a class derived from the Extractor base class.

After installing mypackage lingua will automatically detect the new custom extractor.

Helper Script

There exists a helper shell script for managing translations of packages in docs/examples named Copy it to package root where you want to work on translations, edit the configuration params inside the script and use:

./ lang

for initial catalog creation and:


for updating translation and compiling the catalog.


4.14 - November 10, 2019

  • Allow comments to appear inside function arguments, instead of only before the callable invocation.
  • Correct the combining of flags in i18n comments when multiple comments are present for a single string.
  • Use Click for command-line parsing. This fixes issue 76.
  • Fix handling of non-keyword values from a Babel extractor. This fixes issue 86.
  • Allow running without Chameleon installed. Patch #97 from sinoroc.
  • Skip function definitions in Python extractor. Patch #93 from Alexandre Brault.
  • Skip extractor entry points if dependencies are missing. Patch #97 from sinoroc.
  • Update TravisCI config to current Python versions. Patch #96 from sinoroc.
  • Fix Pytest4.x compatibility error. Patch #92 from Stanislav Levin.
  • Update the XML extractor to support for custom expressions engines containing numbers. Patch #88 from Thierry Florac.
  • Update to show language during the compilation process. Patch #83 from Hans-Peter Jansen.

4.13 - January 24, 2016

  • Fix --sort-by-file to use a numeric ordering for line numbers.

4.12 - December 28, 2016

  • Improve handling of Chameleon parse errors.
  • Correctly set the context for i18n:attributes in TAL templates.

4.11 - October 11, 2016

  • Correct a Python 3 compatibility error. This fixes issue 82.

4.10 - October 10, 2016

  • Fix error saving POT files on Windows. This fixes issue 81.
  • Fix incorrect handling of stripped line numbers. Patch #79 from sinoroc.
  • Improve handling of expressions (${..}) in HTML/PT files: - decode HTML entities before trying to parse Python code - gracefully handle syntax errors.

4.9 - April 5, 2016

  • Handle calls to the Python extractor with bytes input. This fixes problems with Lingua plugins build for Lingua < 4.
  • Atomically replace the POT file to guarantee no broken POT file will be created.
  • Do not replace a POT file if there are no changes (ignoring metadata changes). This fixes issue 67

4.8.1 - February 18, 2016

  • Fix breaking error introduced in 4.8 release.

4.8 - February 18, 2016

  • The XML extractor correctly handle i18n comments for attributes now.

4.7 - January 23, 2016

  • Make i18n comments in HTML file inheritable. This was already documented, but did not work.

4.6 - January 9, 2016

  • Use unittest.mock on Python 3.3 and later.
  • The Python extractor correctly handles the --keyword parameter now. Patch #75 from Omar Kohl, fixing issue 74

4.5.1 - November 11, 2015

  • Fix Python 3 compatibility.

4.5 - November 11, 2015

  • Allow setting message flags in Python comments.

4.4 - November 10, 2015

  • Add pgettext() to the default keyword list.
  • Do full keyword processing for results from Babel extraction plugins. This fixes issue 71
  • Fix Python 3 errors when processing Python files. This fixes issue 74.

4.3.1 - November 5, 2015

  • Re-uplad with different version number to work around an earlir release error.

4.3 - November 5, 2015

  • Fix parsing of Python code using the dot operator in a keyword argument. Patch #70 from Patrick Valsecchi.

4.2 - November 5, 2015

  • Allow message comments on the same line as gettext calls in Python.

4.1 - November 4, 2015

  • Correctly handle messages that have both a translation comment, and use Zope-style message ids.
  • Catch and properly report errors during tokenisation of Python code.

4.0 - November 4, 2015

  • Rewrite the Python extractor again. Lingua will now parse Python code itself instead of using the Python ast module.
  • Support message comments in Python code. These can be put in lines immediately before a line containing a message, and must be prefixed with a special marker.
  • Use domain filtering for Python expressions where possible (i.e. when using translaction functions that take a domain parameter, such as dgettext.
  • Fix –no-location option to actually work.
  • Add –no-linenumbers option.

3.12 - October 10, 2015

  • Fix HTML extractor to apply domain filter when processing attributes and ${..} expressions. This fixes issue 66
  • Fix line number reporting when processing instructions are used. Patch #64 from Florian Schulze.
  • Fix handling of extractor configuration. Patch #63 from Stéphane Brunner.

3.11 - August 6, 2015

  • Correctly handle Python expressions using braces in template text. This fixes issue 62.

3.10 - May 1, 2015

  • Update example to show statistics when compiling catalogs. This reveals catalogs with fuzzy messages. This fixes issue 59.
  • Fix handling of line number parameter in the Python extractor. This fixes invalid line numbers generated for Python code embedded in other files, for example in Mako templates. This fixes issue 58 based on a fix from Laurent Daverio.
  • Warn when using a function call instead of a string as parameter in a gettext keyword in Python code. This fixes issue 57.

3.9 - February 19, 2015

  • Fix line number reporting for XML/zope/Chameleon extractors. Pull request 53 from Florian Schulze.

3.8 - January 20, 2015

  • Add options to sort messages by either location or message id when creating a POT file. Based on pull request 51 from Emanuele Gaifas.

3.7 - December 17, 2014

  • Include used lingua version in POT metadata.
  • Add support for message contexts in translationstring instances.
  • Add support for i18n:comment attributes in ZPT templates.

3.6.1 - November 11, 2014

  • Restore Python 2.6 compatibility.

3.6 - November 11, 2014

  • Extend automatic context-comments for ZPT templates to also show the canonical text for sub-elements. For example this markup:

    <p i18n:translate="">This is just
      <em i18n:name="wonderful" i18n:translate="">wonderful</em>!</p>

    The resulting PO entry for the outer element will look like this:

    #. Canonical text for ${wonderful} is: "wonderful"
    msgid "This is just ${wonderful}!"
    msgstr ""

    This is only do for sub-elements that have a name (i.e. the have an i18n:name attribute).

3.5.1 - November 11, 2014

  • Fix a bug in the generated context-comment for substrings which caused part of the outer sentence text to be skipped.

3.5 - November 11, 2014

  • When encountering nested translations in ZPT templates automatically add a comment to the sub-message with the full text for the outer sentence. For example this markup:

    <p i18n:translate="">This is just
      <em i18n:name="wonderful" i18n:translate="">wonderful</em>!</p>

    The resulting PO entry will look like this:

    #. Used in sentence: "This is just ${wonderful}!"
    msgid "wonderful"
    msgstr ""

    This extra context information can be very important for translators.

3.4 - November 3, 2014

  • Add support for the i18n:context attribute in ZPT templates. This is supported by Chameleon 2.17 and later to set the translation context.

3.3 - September 14, 2014

  • Modify the message format-checker to not consider a space ofter a percent- character as a format flag. Space is a valid flag but is almost never used, and this was creating a lot of false positives (for example a sentence like “take a sample of 5% of all candidates”).
  • Do not try to extract a message from N_() calls: these are explicitly intended to be used for situations where you pass in a variable instead of a string.

3.2 - August 26, 2014

  • Refactor the extractor API a little bit to make it easier for extractors to call each other. This is particularly useful when an extractor needs to call the Python extractor to handle local Python expressions.
  • Correctly extract messages from empty elements that used HTML notation. This fixes issue 48.

3.1 - August 18, 2014

  • Make sure two message strings are added when a plural message is encountered. This fixes an error from msginit when it sees msgid_plural but only a single msgstr

3.0 - August 15, 2014

Zope users: this release changes the extraction for .pt files. You will need to configure lingua to use the new zope extractor for .pt files.

  • Make it possible to configure Babel plugins.
  • Include a brief description in --list-extractors output.
  • Rename the xml extractor to chameleon to better fit the syntax it handles.
  • Add a new zope extractor. This is identical to the chameleon extractor, but changes the default expression engine used to TALES. This engine is now the default for .cpt and .zpt files.
  • Allow per-extractor configuration.
  • Fix handling of comments returned by Babel extractor plugins. This fixes issue 44.

2.5 - August 14, 2014

  • Correct handling of alternatives in TALES expressions when the pipe-symbol was surrounded by whitespace. This fixes issue 43.
  • Extractors can now be hooked up from outside via entry points.
  • Introduce lingua.extractors.Extractor object and use it as base for all extractors.
  • Check for global .config/lingua config file in user’s home directory if no explicit config file defined.
  • Fix error message when detecting a syntax error in scanned Python code. Patch from tisdall (pull request 42).
  • Use a single wheel distrbution for Python 2 and Python 3.

2.4 - July 23, 2014

  • Correctly handle multiple alternatives in ZPT expressions. This fixes issue 39.
  • Do not add c-format flag for messages without an percent-character.
  • Fix timezone format at pot creation (remove colon).

2.3 - June 13, 2014

  • Fix incorrect invocation of legacy Babel extraction plugins. This fixes issue 28.
  • TAL template handling fixes:
    • Correctly handle structure: prefixes in TAL expressions. Patch from Ingmar Steen (pull request 32).
    • Fix handling of multi-line tal:content, tal:define and tal:replace statements. Patch from Ingmar Steen (pull requests 35 and 36).
    • Fix handling of tal:repeat statements with multiple assignments. Patch from Ingmar Steen (pull request 37).

2.2 - June 10, 2014

  • Remove seconds from POT timestamps. No other tool includes seconds, and this appearently breaks Babel.
  • Fix Python 2.6 compatibility. Patch from Hugo Branquinho (pull request 25).
  • Fix installation problems on Python 3. Patch from William Wu (pull request 27).
  • Handle TALES expression engine selection. This fixes issue 30.
  • Handle Python expressions using curly braces in HTML templates. This fixes issue 29.

2.1 - April 8, 2014

  • Do not break when encountering HTML entities in Python expressions in XML templates.
  • Show the correct linenumber in error messages for syntax errors in Python expressions occurring in XML templates.
  • Fix bug in parsing of tal:repeat and tal:define attributes in the XML parser.
  • Tweak ReST-usage in changelog so the package documentation renders correctly on PyPI.

2.0 - April 8, 2014

  • Lingua is now fully Python 3 compatible.
  • Add a new pot-create command to extract translateable texts. This is (almost) a drop-in replacement for GNU gettext’s xgettext command and replaces the use of Babel’s extraction tools. For backwards compatibility this tool can use existing Babel extraction plugins.
  • Define a new extraction plugin API which enables several improvements to be made:
    • You can now select which domain to extract from files. This is currently only supported by the XML and ZCML extractors.
    • Format strings checks are now handled by the extraction plugin instead of applied globally. This prevents false positives.
    • Message contexts are fully supported.
  • Format string detection has been improved: both C and Python format strings are now handled correctly.
  • The XML/HTML extractor has been rewritten to use HTML parser from Chameleon. This allows lingua to handle HTML files that are not valid XML.
  • Whitespace handling in XML extractor has been improved..
  • The po-xls conversion tools have been moved to a new po-xls package.

1.6 - December 9, 2013

  • Add support for ngettext and pluralize() for correctly generating plurals in pot files.

1.5 - April 1, 2013

  • Do not silently ignore XML parsing errors. Instead print an error message and abort.

1.4 - February 11, 2013

  • Po->XLS convertor accidentily included obsolete messages.

1.3 - January 28, 2012

  • XLS->Po conversion failed for the first language if no comment or reference columns were generated. Reported by Rocky Feng.
  • Properly support Windows in the xls-po convertors: Windows does not support atomic file renames, so revert to shutils.rename on that platform. Reported by Rocky Feng.

1.2 - January 13, 2012

  • Extend XML extractor to check python expressions in templates. This fixes issue 7. Thanks to Nuno Teixeira for the patch.

1.1 - November 16, 2011

  • Set ‘i18n’ attribute as default prefix where there was no prefix found. This fixes issues 5 and 6. Thanks to Mathieu Le Marec - Pasquet for the patch.

1.0 - September 8, 2011

  • Update XML extractor to ignore elements which only contain a Chameleon expression (${....}). These can happen to give the template engine a hint that it should try to translate the result of an expression. This fixes issue 2.
  • Update XML extractor to not abort when encountering undeclared namespaces. This fixes issue 3.
  • Fix Python extractor to handle strings split over multiple lines correctly.

1.0b4 - July 20, 2011

  • Fix po-to-xls when including multiple languages in a single xls file.

1.0b3 - July 18, 2011

  • Paper brown bag: remove debug leftover which broke po-to-xls.

1.0b2 - July 18, 2011

  • Update PO-XLS convertors to allow selection of comments to include in the xls files.
  • Correct XML extractor to strip leading and trailing white. This fixes issue 1.
  • Add a very minimal polint tool to perform sanity checks in PO files.
  • Update trove data: Python 2.4 is not supported due to lack of absolute import ability.

1.0b1 - May 13, 2011

  • First release.

Project details

Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Files for lingua, version 4.14
Filename, size File type Python version Upload date Hashes
Filename, size lingua-4.14-py2.py3-none-any.whl (28.8 kB) File type Wheel Python version py2.py3 Upload date Hashes View
Filename, size lingua-4.14.tar.gz (48.8 kB) File type Source Python version None Upload date Hashes View

Supported by

AWS AWS Cloud computing Datadog Datadog Monitoring Facebook / Instagram Facebook / Instagram PSF Sponsor Fastly Fastly CDN Google Google Object Storage and Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Salesforce Salesforce PSF Sponsor Sentry Sentry Error logging StatusPage StatusPage Status page