This seems a regression from 3.9.

>>> foo[x = 1
... ]
  File "<stdin>", line 1
    foo[x = 1
        ^^^^^
SyntaxError: invalid syntax. Maybe you meant '==' or ':=' instead of '='?
>>>

Note how the syntax error is in line 1 and yet the REPL asks for more input (the "..." prompt) and doesn't stop until I type "]".

In 3.9 I just get "SyntaxError: invalid syntax" but the REPL doesn't ask for the second line:

>>> foo[x = 1
  File "<stdin>", line 1
    foo[x = 1
          ^
SyntaxError: invalid syntax
>>>

Unfortunately this happens because to generate the error message we check for the equal and the right hand side of the equal (to also discard some false positives) and this triggers the parser to ask for additional tokens. Notice that you don't need to close the ']', just provide an additional token:

>>> x = [x=4
... f
  File "<stdin>", line 1
    x = [x=4
         ^^^
SyntaxError: invalid syntax. Maybe you meant '==' or ':=' instead of '='?

Unfortunately is not going to be easy to solve without looking at the right hand side :(

I believe that this is similar to, but not quite as severe as a similar bug in code.interact() https://bugs.python.org/issue43366 which affects IDLE; perhaps fixing this might fix the other issue?

No, this issue is in the parser while the other is caused by the code module. As I mentioned, this case doesn't ask you for tokens until you close the brace, it just ask you for one extra token because is doing a lookahead.

If you give any extra token to satisfy the lookahead:

>>> foo[x = 1 $
  File "<stdin>", line 1
    foo[x = 1 $
        ^^^^^
SyntaxError: invalid syntax. Maybe you meant '==' or ':=' instead of '='?
>>>

It fails immediately even if you didn't close the bracket

I will try to see if we can do something about this but the only thing I can think of is removing the lookaheads in the invalid comparison rules but then we need to solve the false positives (like keyword arguments) and that's going to be very hard to do without lookaheads

Could we implement something in the REPL where it refuses to read another
line when asked for a token in an error recovery rule?

Yes, I was exactly working in that :)

New changeset 1fb6b9e by Miss Islington (bot) in branch '3.10':
bpo-44201: Avoid side effects of "invalid_*" rules in the REPL (GH-26298) (GH-26313)
1fb6b9e