As yield is an expression, it's legal in a lambda function, which then
means you have a generator function. But it's not quite the same as
the equivalent function made with def:

$ python3
Python 3.5.0a0 (default:1c51f1650c42+, Dec 29 2014, 02:29:06)
[GCC 4.7.2] on linux
Type "help", "copyright", "credits" or "license" for more information.
>>> f=lambda: (yield 5)
>>> x=f()
>>> next(x)
5
>>> x.send(123)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
StopIteration
>>> def f(): return (yield 5)
...
>>> x=f()
>>> next(x)
5
>>> x.send(123)
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
StopIteration: 123
>>> x = (lambda: print((yield 1)) or 2)()
>>> next(x)
1
>>> x.send(3)
3
Traceback (most recent call last):
  File "<stdin>", line 1, in <module>
StopIteration

The last example demonstrates that send() is working, but the return value is not getting propagated. Disassembly shows this:

>>> dis.dis(lambda: (yield 5))
  1           0 LOAD_CONST               1 (5)
              3 YIELD_VALUE
              4 POP_TOP
              5 LOAD_CONST               0 (None)
              8 RETURN_VALUE
>>> def f(): return (yield 5)
... 
>>> dis.dis(f)
  1           0 LOAD_CONST               1 (5)
              3 YIELD_VALUE
              4 RETURN_VALUE

I'm sure this is a bug that will affect very approximately zero people, but it's still a peculiar inconsistency!

Verified with 3.5 and 3.4.

Hm, looks like nobody bothered to update the lambda code generation to use the value from yield. I almost feel like there is some unnecessary check "if we are in a lambda" in the code generation for yield. Have you looked through the code generation yet?

I'm not sure what to look for in the code generation. In compile.c lines 3456 and following, there's a LOAD_CONST None coming through, in the else branch of "if (e->v.Yield.value)", but nothing talking about lambda functions. There are constants COMPILER_SCOPE_LAMBDA and COMPILER_SCOPE_FUNCTION, but the only place where they're used is compiler_set_qualname() and I can't see anything obvious there. Hopefully someone more familiar with the code internals will be able to figure this out!

Here are the operations being emitted (line, macro used and eventual argument):

>>> f = lambda: (yield 5)
3487: ADDOP_O LOAD_CONST e->v.Num.n
3472: ADDOP YIELD_VALUE
1907: ADDOP_IN_SCOPE POP_TOP
4349: ADDOP_O LOAD_CONST Py_None
4350: ADDOP RETURN_VALUE
1457: ADDOP_O LOAD_CONST (PyObject*)co
1458: ADDOP_O LOAD_CONST qualname
1459: ADDOP_I MAKE_FUNCTION args
4349: ADDOP_O LOAD_CONST Py_None
4350: ADDOP RETURN_VALUE
>>> def g(): return (yield 5)
... 
3487: ADDOP_O LOAD_CONST e->v.Num.n
3472: ADDOP YIELD_VALUE
2533: ADDOP RETURN_VALUE
1457: ADDOP_O LOAD_CONST (PyObject*)co
1458: ADDOP_O LOAD_CONST qualname
1459: ADDOP_I MAKE_FUNCTION args
4349: ADDOP_O LOAD_CONST Py_None
4350: ADDOP RETURN_VALUE

So there's an extra POP_TOP + LOAD_CONST Py_NONE for the lambda version that throws away the "123" (in the exemple).

The attached patch (0001-...) fixes it. However please note that I'm not knowledgable about that part of the code and devised the patch empirically.
Moreover a test should probably added but I did not know where (test_dis? tried and failed... see patch 0002-...).

Could you please add a test based on Chris's example? And it would be good to add a test for a lambda with "yield from".

Here is a working test, testing yield by lambda & function as well as lambda and function yielding from those.

New changeset 2b4a04c3681b by Serhiy Storchaka in branch '3.4':
Issue #23192: Fixed generator lambdas.  Patch by Bruno Cauet.
https://hg.python.org/cpython/rev/2b4a04c3681b

New changeset a3b889e9d3f3 by Serhiy Storchaka in branch 'default':
Issue #23192: Fixed generator lambdas.  Patch by Bruno Cauet.
https://hg.python.org/cpython/rev/a3b889e9d3f3

Committed with non-dis test. Thank you for your contribution Bruno.