Please forgive me for hooking into this discussion so late. Below are my late comments to your original PEP, and below those some new stuff. I have been writing weightless/compose which does exactly what your PEP is trying to accomplish. I'll check my stuff against this PEP.
I really appreciate your initiative! It helps me a lot.
2009/4/15 Greg Ewing firstname.lastname@example.org:
Draft 11 of the PEP.
Changes in this version:
- GeneratorExit always calls close() and is always
- Special handling of thrown-in StopIterations
removed, since Guido doesn't think you should be doing that in the first place.
- Expansion uses next(_i) instead of _i.next() and
doesn't mention cacheing of methods.
PEP: XXX Title: Syntax for Delegating to a Subgenerator Version: $Revision$ Last-Modified: $Date$ Author: Gregory Ewing email@example.com Status: Draft Type: Standards Track Content-Type: text/x-rst Created: 13-Feb-2009 Python-Version: 3.x Post-History:
A syntax is proposed for a generator to delegate part of its operations to another generator. This allows a section of code containing 'yield' to be factored out and placed in another generator. Additionally, the subgenerator is allowed to return with a value, and the value is made available to the delegating generator.
The new syntax also opens up some opportunities for optimisation when one generator re-yields values produced by another.
A Python generator is a form of coroutine, but has the limitation that it can only yield to its immediate caller. This means that a piece of code containing a ``yield`` cannot be factored out and put into a separate function in the same way as other code. Performing such a factoring causes the called function to itself become a generator, and it is necessary to explicitly iterate over this second generator and re-yield any values that it produces.
If yielding of values is the only concern, this can be performed without much difficulty using a loop such as
for v in g: yield v
However, if the subgenerator is to interact properly with the caller in the case of calls to ``send()``, ``throw()`` and ``close()``, things become considerably more difficult. As will be seen later, the necessary code is very complicated, and it is tricky to handle all the corner cases correctly.
A new syntax will be proposed to address this issue. In the simplest use cases, it will be equivalent to the above for-loop, but it will also handle the full range of generator behaviour, and allow generator code to be refactored in a simple and straightforward way.
The following new expression syntax will be allowed in the body of a generator:
yield from <expr>
These are the exact problems I can't solve neatly in weightless/compose:
where <expr> is an expression evaluating to an iterable, from which an iterator is extracted. The iterator is run to exhaustion, during which time it yields and receives values directly to or from the caller of the generator containing the ``yield from`` expression (the "delegating generator").
this allows a programmer to express the intention of just returning a generator or wanting to delegate the work to a 'subgenerator'. Weightless/compose now just descends into every generator, while this is certainly not always wanted. Great I think, I like the syntax.
Furthermore, when the iterator is another generator, the subgenerator is allowed to execute a ``return`` statement with a value, and that value becomes the value of the ``yield from`` expression.
In Weightless/compose, after several different tries I settled for mimicking returning a value by using raise StopIteration(returnvalue). As return in a generator raises StopIteration(), I think it is very natural to use return like this in a generator (if fact I wished it would be possible sometimes, not being aware of python-ideas). So I like it too.
The full semantics of the ``yield from`` expression can be described in terms of the generator protocol as follows:
* Any values that the iterator yields are passed directly to the caller.
* Any values sent to the delegating generator using ``send()`` are passed directly to the iterator. If the sent value is None, the iterator's ``next()`` method is called. If the sent value is not None, the iterator's ``send()`` method is called. Any exception resulting from attempting to call ``next`` or ``send`` is raised in the delegating generator.
Clear. I have implemented this by just calling send(...) either with None or with a value. The VM dispatches that to next() when the value is None, I assume.
* Exceptions other than GeneratorExit passed to the ``throw()`` method of the delegating generator are forwarded to the ``throw()`` method of the iterator. Any exception resulting from attempting to call ``throw()`` are propagated to the delegating generator.
I let any Exception propagate using the throw() method. I believe this will not correctly handle GeneratorExit as outlined in the discussion before. I'll have to change this I think.
* If a GeneratorExit exception is thrown into the delegating generator, the ``close()`` method of the iterator is called if it has one. If this call results in an exception, it is propagated to the delegating generator. Otherwise, the GeneratorExit is reraised in the delegating generator.
I have a hard time understanding what this would mean in a pure python implementation. I added both bullets to my unittests to work it out later.
The implicit GeneratorExit resulting from closing the delegating generator is treated as though it were passed in using ``throw()``.
By "closing the delegating generator" you mean "from the outside, call close() on it"? It then will raise the GeneratorExit exception, and I understand it. I added a unittest as well.
* The value of the ``yield from`` expression is the first argument to the ``StopIteration`` exception raised by the iterator when it terminates.
* ``return expr`` in a generator causes ``StopIteration(expr)`` to be raised.
I assume that 'return 1 2 3' will have one return value being a tuple (1,2,3) which is one argument to StopIteration(), and which is unpacked when 'yield from' returns?
Enhancements to StopIteration
For convenience, the ``StopIteration`` exception will be given a ``value`` attribute that holds its first argument, or None if there are no arguments.
I am using StopIteration's 'args' atrribute? But after reading the motivation below, it could indeed confuse other generators, and a separate StopIteration would be better, I think.
Python 3 syntax is used in this section.
- The statement
RESULT = yield from EXPR
is semantically equivalent to
_i = iter(EXPR) try: _y = next(_i) except StopIteration as _e: _r = _e.value else: while 1: try: _s = yield _y except GeneratorExit: _m = getattr(_i, 'close', None) if _m is not None: _m() raise except: _m = getattr(_i, 'throw', None) if _m is not None: _y = _m(*sys.exc_info()) else: raise else: try: if _s is None: _y = next(_i) else: _y = _i.send(_s) except StopIteration as _e: _r = _e.value break RESULT = _r
I'll take this one with me, as I really need some time to compare it to my own code. I'll come back to it later.
- In a generator, the statement
is semantically equivalent to
except that, as currently, the exception cannot be caught by ``except`` clauses within the returning generator.
- The StopIteration exception behaves as though defined thusly:
def __init__(self, *args): if len(args) > 0: self.value = args else: self.value = None Exception.__init__(self, *args)
I probably miss the point, could you explain why this is needed?
The Refactoring Principle
The rationale behind most of the semantics presented above stems from the desire to be able to refactor generator code. It should be possible to take an section of code containing one or more ``yield`` expressions, move it into a separate function (using the usual techniques to deal with references to variables in the surrounding scope, etc.), and call the new function using a ``yield from`` expression.
The behaviour of the resulting compound generator should be, as far as possible, exactly the same as the original unfactored generator in all situations, including calls to ``next()``, ``send()``, ``throw()`` and ``close()``.
The semantics in cases of subiterators other than generators has been chosen as a reasonable generalization of the generator case.
Yes! Exactly. I just call this supporting 'program decomposition'. For clearity, you could probably add the name of the refactoring, it is called 'extract method' isn't it?
There was some debate as to whether explicitly finalizing the delegating generator by calling its ``close()`` method while it is suspended at a ``yield from`` should also finalize the subiterator. An argument against doing so is that it would result in premature finalization of the subiterator if references to it exist elsewhere.
Consideration of non-refcounting Python implementations led to the decision that this explicit finalization should be performed, so that explicitly closing a factored generator has the same effect as doing so to an unfactored one in all Python implementations.
The assumption made is that, in the majority of use cases, the subiterator will not be shared. The rare case of a shared subiterator can be accommodated by means of a wrapper that blocks ``throw()`` and ``close()`` calls, or by using a means other than ``yield from`` to call the subiterator.
I agree completely. I went through some lenght to get proper clean-up, and I solved it similarly.
Generators as Threads
A motivation for generators being able to return values concerns the use of generators to implement lightweight threads. When using generators in that way, it is reasonable to want to spread the computation performed by the lightweight thread over many functions. One would like to be able to call a subgenerator as though it were an ordinary function, passing it parameters and receiving a returned value.
Using the proposed syntax, a statement such as
y = f(x)
where f is an ordinary function, can be transformed into a delegation call
y = yield from g(x)
where g is a generator. One can reason about the behaviour of the resulting code by thinking of g as an ordinary function that can be suspended using a ``yield`` statement.
When using generators as threads in this way, typically one is not interested in the values being passed in or out of the yields. However, there are use cases for this as well, where the thread is seen as a producer or consumer of items. The ``yield from`` expression allows the logic of the thread to be spread over as many functions as desired, with the production or consumption of items occuring in any subfunction, and the items are automatically routed to or from their ultimate source or destination.
Concerning ``throw()`` and ``close()``, it is reasonable to expect that if an exception is thrown into the thread from outside, it should first be raised in the innermost generator where the thread is suspended, and propagate outwards from there; and that if the thread is terminated from outside by calling ``close()``, the chain of active generators should be finalised from the innermost outwards.
Yes, I believe you make sure that:
try: x = yield from y() except SomeError: return 'HELP'
actually does catch the SomeError exception when raised in y(), or one it its descendants?
The particular syntax proposed has been chosen as suggestive of its meaning, while not introducing any new keywords and clearly standing out as being different from a plain ``yield``.
Next section I skipped, I you don't mind.
Using a specialised syntax opens up possibilities for optimisation when there is a long chain of generators. Such chains can arise, for instance, when recursively traversing a tree structure. The overhead of passing ``next()`` calls and yielded values down and up the chain can cause what ought to be an O(n) operation to become, in the worst case, O(n**2).
A possible strategy is to add a slot to generator objects to hold a generator being delegated to. When a ``next()`` or ``send()`` call is made on the generator, this slot is checked first, and if it is nonempty, the generator that it references is resumed instead. If it raises StopIteration, the slot is cleared and the main generator is resumed.
This would reduce the delegation overhead to a chain of C function calls involving no Python code execution. A possible enhancement would be to traverse the whole chain of generators in a loop and directly resume the one at the end, although the handling of StopIteration is more complicated then.
Use of StopIteration to return values
There are a variety of ways that the return value from the generator could be passed back. Some alternatives include storing it as an attribute of the generator-iterator object, or returning it as the value of the ``close()`` call to the subgenerator. However, the proposed mechanism is attractive for a couple of reasons:
- Using a generalization of the StopIteration exception makes it easy
for other kinds of iterators to participate in the protocol without having to grow an extra attribute or a close() method.
- It simplifies the implementation, because the point at which the
return value from the subgenerator becomes available is the same point at which the exception is raised. Delaying until any later time would require storing the return value somewhere.
Originally it was proposed to simply extend StopIteration to accept a value. However, it was felt desirable by some to have a mechanism for detecting the erroneous use of a value-returning generator in a context that is not aware of generator return values. Using an exception that is a superclass of StopIteration means that code knowing about generator return values only has one exception to catch, and code that does not know about them will fail to catch the new exception.
I agree. And I begin to understand the need for that value attribute. Ok.
For now I have one more fundamental question left over.
Will the the delegating generator remain on the call-stack or not?
The current behaviour is that a stack-frame is created for a generator, which is not disposed when next()/send() returns, but kept somewere. When a new call to next()/send() happens, the same stack-frame is put back on the call-stack. This is crucial because it acts as the o-so-valuable closure.
I came across this problem because I was writing code that traverses the call-stack in order to find some place to put 'generator-local' variables (like thread-local). My implementation in weightless/compose does not physically keep the generators on the call-stack (I don't know how to do that in Python), but keep it's own stack. I would have to extend 'compose' to not only search on the real call-stack but also traverse it own semi-call-stack if/when it finds an instance of itself on the real call-stack.
I am writing code like:
def x(): somevar = 10 yield from y()
then it I write in y:
def y(): frame = currentframe().f_back while 'somevar' not in frame.f_locals: frame = frame.f_back return frame.f_locals['somevar']
would this find the variable in x?
Best regards, Erik