Seems that this is allowed since the PEG parser rewrite:

for x in *a, *b:
    print(x)

but I cannot find anywhere were we discussed this. I am not sure if we should keep it or treat it as a bug and fix it.

Since this was already allowed in 3.9 and 3.10 stable relases, and since it is consistent with the RHS of an assignment (something = *a, *b); I'd lean towards keeping it (and maybe fixing the old parser's grammar to reflect that) and documenting this.

I'd lean towards keeping this syntax:
- It's already been out for two releases, so there's user code out there relying on it. (In fact we found out about this because somebody complained that Black's parser couldn't handle this code.)
- The syntax isn't obviously confusing and the meaning is clear.
- It's not hard to parse; it was easy to adjust Black's parser to allow it.

+1 (what Jelle said)

Let's just document it for 3.11.

Will prepare a PR

Allowing no parentheses is also consistent with the following:
  for x in 1,2,3: print(x)

Adding jwilk back (a bpo interaction with browser form caching makes this happen frequently).

What exactly was happened? What rule was changed? Can it cause other changes which allow ambiguous code or change semantic?

What happened is that the new grammar using the PEG parser used the equivalent of starred_testlist instead of testlist for the iterable list of for statements.

The only extra thing allowed is starred elements, that are interpreted as if you are building a tuple without parentheses.

New changeset 7fb94fd7a88f14096a509dddd4d8a979a3912672 by Pablo Galindo Salgado in branch 'main':
bpo-46725: Document starred expressions in for statements (GH-31481)
https://github.com/python/cpython/commit/7fb94fd7a88f14096a509dddd4d8a979a3912672