i18n babel can not extract lazy text strings correct

[Robotvasya]

Checklist

• I am sure the error is coming from aiogram code
• I have searched in the issue tracker for similar bug reports, including closed ones

Operating system

Any

Python version

3.10

aiogram version

3.3.0

Expected behavior

I make en example according this document: https://docs.aiogram.dev/en/dev-3.x/utils/i18n.html

...
from aiogram.utils.i18n import gettext as _
from aiogram.utils.i18n import lazy_gettext as __
from aiogram.utils.i18n import I18n, ConstI18nMiddleware
...
@dp.message(F.text == __("Start"))  # lazy text
async def handler_1(message: Message) -> None:
    await message.answer(_("Welcome, {name}!").format(name=html.quote(message.from_user.full_name)))
    await message.answer(_("How many messages do you have? Input number, please:"))

@dp.message(F.text)
async def handler_2(message: Message) -> None:
    # this is plural and wrapped by double underscore
    await message.answer(__("You have {} message!", "You have {} messages!", 2).format(message.text)) 
...
def main() -> None:
...
    i18n = I18n(path="locales", default_locale="en", domain="my-super-bot")
    dp.message.outer_middleware(ConstI18nMiddleware(locale='en', i18n=i18n))
...

I create lines lazy text and gettext wrapped strings.
Also I add plural forms.
Extracting strings using Babel should return this strings in .pot template file:

"Content-Transfer-Encoding: 8bit\n"
"Generated-By: Babel 2.13.1\n"

#: lesson1.py:12
msgid "Start"
msgstr ""

#: lesson1.py:14
msgid "Welcome, {name}!"
msgstr ""

#: lesson1.py:15
msgid "How many messages do you have? Input number, please:"
msgstr ""


#: lesson1.py:19
msgid "You have {} message!"
msgid_plural "You have {} messages!"
msgstr[0] ""
msgstr[1] ""

Current behavior

After extraction we loose some strings.
Babel command

pybabel extract -o locales/messages.pot --keywords="__:1,2" --input-dirs=.

extracts either this:

"Content-Transfer-Encoding: 8bit\n"
"Generated-By: Babel 2.13.1\n"

<---- So here we loose lazy text string "Start", see expected behavior.

#: lesson1.py:14 
msgid "Welcome, {name}!"   # regular gettext
msgstr ""

#: lesson1.py:15
msgid "How many messages do you have? Input number, please:"  # regular gettext
msgstr ""

#: lesson1.py:19
msgid "You have {} message!"  # alias ngettext as gettext in \aiogram\utils\i18n\context.py
msgid_plural "You have {} messages!"
msgstr[0] ""
msgstr[1] ""

or

pybabel extract -o locales/messages.pot --keywords="__"  --input-dirs=.

extracts that:

"Content-Transfer-Encoding: 8bit\n"
"Generated-By: Babel 2.13.1\n"

#: lesson1.py:12
msgid "Start"
msgstr ""

#: lesson1.py:14
msgid "Welcome, {name}!"
msgstr ""

#: lesson1.py:15
msgid "How many messages do you have? Input number, please:"
msgstr ""

#: lesson1.py:19
msgid "You have {} message!"
msgstr ""
<---- Here we loose plural forms, see expected behavior.

Here we loose plural forms.

Ether this or that but not all together.
The use of these keys __:1,2 оr _:1,2 _ __ and their various combinations leads to an even more disastrous result.

Steps to reproduce

pybabel extract -o locales/messages.pot --input-dirs=.
pybabel extract -o locales/messages.pot --keywords="__" --input-dirs=.
pybabel extract -o locales/messages.pot --keywords="__:1,2" --input-dirs=.
pybabel extract -o locales/messages.pot --keywords="__ __:1,2"--input-dirs=.

Code example

from aiogram import Bot, Dispatcher, F, html
from aiogram.types import Message
from aiogram.utils.i18n import gettext as _

from aiogram.utils.i18n import lazy_gettext as __
from aiogram.utils.i18n import I18n, ConstI18nMiddleware

TOKEN = "token"
dp = Dispatcher()


@dp.message(F.text == __("Start"))
async def handler_1(message: Message) -> None:
    await message.answer(_("Welcome, {name}!").format(name=html.quote(message.from_user.full_name)))
    await message.answer(_("How many messages do you have? Input number, please:"))

@dp.message(F.text)
async def handler_2(message: Message) -> None:
    await message.answer(__("You have {} message!", "You have {} messages!", 2).format(message.text))


def main() -> None:
    bot = Bot(TOKEN, parse_mode="HTML")
    i18n = I18n(path="locales", default_locale="en", domain="my-super-bot")
    dp.message.outer_middleware(ConstI18nMiddleware(locale='en', i18n=i18n))
    dp.run_polling(bot)


if __name__ == "__main__":
    main()

Logs

No response

Additional information

No response

[Olegt0rr]

Is there any issue with the same behaviour without aiogram?

Looks like you can remove aiogram from this example and create an issue for babel

[Robotvasya]

Thank you, Oleg.

Really. I did some research on how functions work in i18n/utils.
Since the aiogram library combines together gettext and ngettext as _(), this effect occurs when retrieving strings using Babel. Although xgettext works a little better, I still had to suffer. This happens because gettext and ngettext functions have different numbers of arguments.

I propose to add the following clarification to the documentation https://docs.aiogram.dev/en/dev-3.x/utils/i18n.html in the Deal with Babel Step 1 Note:
To retrieve simultaneously the strings wrapped by gettext as _(), ngettext as _() and lazy_gettext as __() the following retrieve command
pybabel extract -o locales/messages.pot -k _:1,1t -k _:1,2 -k __
or
xgettext -L Python --keyword=_:1,2 --keyword=__ your_file.py

UPD
To success extraction plural forms should be wrapped by _() and not __().

[Olegt0rr]

Thanx for ur investigation!

Feel free to make a contribution via PR