Terry Reedy tjreedy at udel.edu
Tue Nov 17 20:19:59 CET 2009

Andreas Löscher wrote:
> Hi,
> I am not sure if this is the right newsgroup, so if not don't hesitate
> to tell me.

Since there is no CPython internals list, this is the right place to 
start, even though you might end up having to post a slightly off-topic 
query to python-devel just to get the attention of whoever has the 
answer. But there are a few deveoopers who read and post here also.

> I am developed a Python to C compiler, so that Byte Code files
> automatically can be translated into C Extension Modules. (And it works
> pretty well --> http://www.coremountains.com/products/bytecoat/)
> While developing, I found something strange concerning the YIELD_VALUE
> OpCode.
> Since Python Version 2.5 it behaves the following:
> 	1. pop yield value from stack and return it to
> 		a former gen_send_ex() call from Objects/genobject.c
> 	2. push the yield value on the stack in gen_send_ex() and return 		it
> 	3. when the generator is executed again, the yield value is
> 		'poped' from the stack again with the POP_TOP opcode
> Now I found that a little strange:
> 	1. why is the value removed from the stack, than pushed on the 
> 		stack to remove it finally again? the combination of 
> 		YIELD_VALUE and TOP_POP seems hard coded in compile.c
> 		which means that a TOP_POP follows every YIELD_VALUE 
> 	2. If the semantic of the YIELD_VALUE OpCode has changed, why
> 		is this reached by using another function? Such thing 
> 		should be done in the OpCode.
> 		(e.a.: instead of retval = POP() 
> 			--> retval = TOP(); Py_INCREF(retval); )
> I am a little confused about this.

I suspect that what you see reflects the 2.5 change of 'yield x' from a 
statement to an expression to enable gen.send(). It is possible that 
someone did the simplest thing that worked, rather than properly 
rewriting the code. Or maybe the trickery is needed. You can always try 
a simplyfing patch against the test suite and even submit it if it passes.

Terry Jan Reedy

More information about the Python-list mailing list