Manually set the encoding value after get the text value will not working

[821938089]

Since charset_normalizer has the possibility to give wrong encoding (like this), I wrote some code to check the encoding manually, but I found that this doesn't work.
After my research, I found that as long as I get the text value and then set the encoding, then the text value will not change.
I think this is a bug.

import httpx

headers = {'User-Agent': 'Mozilla/5.0 (Windows NT 10.0; Win64; x64) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/91.0.4472.124 Safari/537.36',
           'Content-Type': 'application/x-www-form-urlencoded'}

postData = 'searchtype=articlename&keyword=%CE%D2%B5%C4'

url = 'http://www.xntk.org/search.php'

r = httpx.post(url, headers=headers, data=postData)

print(r.encoding) # print 'mac_greek' or 'iso8859_5', charset_normalizer guessed a wrong encoding

if r.text.find('charset=gbk') != -1 or r.text.find('charset=gb2312') != -1 or r.text.find('charset="gbk"') != -1:
    r.encoding = "gbk"
else:
    r.encoding = "utf-8"

print(r.encoding) # gbk
print(r.text) # still print wrong encoding text

[tomchristie]

This is the behaviour I would expect to see.

I suppose we could do slightly better here if we raised an exception if a user attempts to set the encoding, after having already accessed r.text.

[821938089]

This behavior has the potential to cause confusion for the user.
I think it is necessary to throw an exception and instruct the user to use r.content to implement this requirement.
And this behavior is inconsistent with requests, which should also be detailed in the corresponding place in the documentation.

[xzmeng]

I found it counter-intuitive too, especially when you're playing around in an interactive environment, such like Pycharm Jupyter, it has a variable explorer which reads the text property when you expand the Response object, and then you overwrite the encoding, just find nothing changes, feeling confused and look into the source to learn what really happened under the hood.
I suggest it should at least give a warning, if raising an exception is too strict. Or why not just keep consistent with requests?