When building for zh_TW, zh_CN is incorrectly used as fallback language

[alvinhochun]

Describe the bug

In sphinx/locale/__init__.py:26 it uses the language part of the locale code as a fallback, so using zh_TW means [ 'zh_TW', 'zh' ] will be passed to gettext.translation.

In CPython, gettext.translation uses gettext.find, which finds the .mo file for each language specified. For zh_TW it finds the .mo file for zh_TW normally. However, when it gets to zh it first normalizes the locale code to zh_CN, which means it ends up finding the .mo file for zh_CN and adding that as a fallback.

gettext.translation already includes the language-without-country-code fallback. The "fallback" that Sphinx adds is unneeded and incorrect.

How to Reproduce

$ https://invent.kde.org/documentation/docs-krita-org.git -b krita/5.0
$ cd docs-krita-org
$ pip install sphinx
$ sphinx-intl build
$ make html SPHINXOPTS="-D language=zh_TW"
$ # open _build/html/user_manual/getting_started.html and see Simplified Chinese mixed in the text

Expected behavior

Untranslated text shall use the original English text and not "fall back" to Simplified Chinese.

Your project

https://invent.kde.org/documentation/docs-krita-org

Screenshots

Expected:

Incorrect:

OS

Linux

Python version

Python 3.9

Sphinx version

4.2

Sphinx extensions

No response

Extra tools

No response

Additional context

No response