Fix KeyError for translations

[drew2a]

This PR fixes #7363 by fixing the root cause of the issue — incorrect translation files.

Also, it modifies the tr function in a way that it is safe from this type of error:

class TranslatedString(str):
    """ This class is used to wrap translated strings to be able to log untranslated strings in case of errors.
        Thanks to this class no `KeyError` exceptions are raised when a translation is missing.
    """

    def __new__(cls, translation, original_string):  # pylint: disable=unused-argument
        return super().__new__(cls, translation)

    def __init__(self, translation: str, original_string: str):  # pylint: disable=unused-argument
        super().__init__()
        self.original_string = original_string

    def __mod__(self, other):
        try:
            return str.__mod__(self, other)
        except KeyError as e:
            msg = f'No value provided for {e} in translation "{self}", original string: "{self.original_string}"'
            logger.warning(f'{type(e).__name__}: {msg}')
            return self.original_string % other


def tr(key):
    return f"{QCoreApplication.translate('@default', key)}"
    translated_string = QCoreApplication.translate('@default', key)
    return TranslatedString(translated_string, original_string=key)

[kozlovsky]

A nice update of the tr function!
It is also necessary to convert .ts files to .qm files by executing the update_translations.sh script and commit the updated .qm files.

[drew2a]

@kozlovsky it gives me the following error:

➜ ./update_translations.sh
./update_translations.sh: line 3: lrelease: command not found

ref:

• https://doc.qt.io/qtcreator/creator-editor-external.html#using-qt-linguist
• https://github.com/lelegard/qtlinguist-installers

[kozlovsky]

I tested Tribler with ES_ES locale, and it still crashes on

    <message>
        <location filename="../utilities.py" line="166"/>
        <source>%(seconds)is</source>
        <translation>%(segundos)is</translation>
    </message>

[sentry-for-tribler]

Sentry issue: TMP-5P

[kozlovsky]

It looks like intercepting KeyError inside the tr function doesn't work because the actual error happens outside, for example:

return tr("%(seconds)is") % data

Here, the tr function returns an incorrect translation with a wrong key, and then the % operation is applied to this returned translation string.

The QCoreApplication.translate call inside the tr function doesn't raise any KeyError, so it is better to remove try/except here.

If we want to prevent this type of KeyError exception, we can return a special subclass of a string with overridden % method that catches KeyError and returns the result of % applied to the original string.

[kozlovsky]

For example, we can use the following TranslatedString class to catch KeyError exceptions and properly log warnings:

class TranslatedString(str):
    def __new__(cls, translation, original_string):
        return super().__new__(cls, translation)

    def __init__(self, translation: str, original_string: str):  # pylint: disable=unused-argument
        super().__init__()
        self.original_string = original_string

    def __mod__(self, other):
        try:
            return str.__mod__(self, other)
        except KeyError as e:
            msg = f'No value provided for {e} in translation "{self}", original string: "{self.original_string}"'
            logger.warning(f'{type(e).__name__}: {msg}')
            return self.original_string % other


def tr(key):
    translated_string = QCoreApplication.translate('@default', key)
    return TranslatedString(translated_string, original_string=key)

Tests:

def test_correct_translation():
    original_string = 'original %(key1)s'
    translated_string = 'translated %(key1)s'
    s = TranslatedString(translated_string, original_string)
    assert s % {'key1': '123'} == 'translated 123'


@patch('tribler.gui.utilities.logger.warning')
def test_missed_key_in_translated_string(warning: Mock):
    original_string = 'original %(key1)s'
    translated_string = 'translated %(key2)s'
    s = TranslatedString(translated_string, original_string)

    assert s % {'key1': '123'} == 'original 123'

    warning.assert_called_once_with('KeyError: No value provided for \'key2\' in translation "translated %(key2)s", '
                                    'original string: "original %(key1)s"')


@patch('tribler.gui.utilities.logger.warning')
def test_missed_key_in_both_translated_and_original_strings(warning: Mock):
    original_string = 'original %(key1)s'
    translated_string = 'translated %(key2)s'
    s = TranslatedString(translated_string, original_string)

    with pytest.raises(KeyError, match=r"^'key1'$"):
        s % {'key3': '123'}

    warning.assert_called_once_with('KeyError: No value provided for \'key2\' in translation "translated %(key2)s", '
                                    'original string: "original %(key1)s"')

[kozlovsky]

This part of the .ts file still needs to be fixed (with recompilation of the .qm file):

    <message>
        <location filename="../utilities.py" line="166"/>
        <source>%(seconds)is</source>
        <translation>%(segundos)is</translation>
    </message>

[kozlovsky]

nitpick: Currently, both weeks (semanas) and seconds (segundos) are abbreviated as s. I suspect it can cause some confusion. According to the Wikipedia, in Spain, "the letters "m" and "s" are used to denote minutes and seconds". With this, I'd prefer to leave "weeks" without an abbreviation:

Suggested change

	<translation>%(weeks)is %(days)id</translation>
	<translation>%(weeks)i semanas %(days)i días</translation>

But we can leave it as is as well.