Steps to reproduce (Windows/Python 3.7.7):

1. Define replacement string that starts with an integer:

REPLACEMENT = '12345'

2. Use re.sub() as follows:

re.sub(r'([a-z]+)', fr"\1{REPLACEMENT}", 'something')

3. The outcome is not 'something12345' as expected, but 'J345'.

Note that I am using the group in the replacement argument, which is a raw f-string.

A quick investigation with other replacement strings renders similar unexpected behavior:

REPLACEMENT = '1': leads to re.error (invalid group reference 11 at position 1)
REPLACEMENT = '13': 'K'
etc.

So it seems like the f-string is evaluated first, yielding a string starting with an integer. Python then interprets the '\1' to indicate group 1 as '\1<first integer>', which leads to the behavior described above. Even if this is by design, it seems confusing and makes using groups with re.sub() cumbersome if the replacement f-string starts with an integer.

f-strings are indeed evaluated when the value of the string is needed. Your example is equivalent to:

>>> re.sub(r'([a-z]+)', fr"\112345", 'something')
'J345'

As always with regexes, you need to be careful when dynamically composing them.

Arguments are evaluated first and then the results are passed to the function. That's true throughout the language.

In this instance, you can use \g<1> in the replacement string to refer to group 1:

re.sub(r'([a-z]+)', fr"\g<1>{REPLACEMENT}", 'something')