Terry:

The eval is important. The bug was in evaluating an f-string that consisted of two bytes: a backslash followed by a newline.

And just as:
eval("'\\\n'") == ''  # len == 0
so should
eval("f'\\\n'") == ''  # len == 0

It's the second one that was throwing the assertion. The parser was seeing these bytes:
f  0x66
'  0x27
\  0x5c
nl 0xa
'  0x27
and behaving badly (to say the least) by asserting.

Without the eval, I can't think of a way to have a string consisting of those two bytes, but I assume someone who's trickier than I can come up with a way.