Re: [Python-Dev] Extended Buffer Interface/Protocol
Greg Ewing wrote: But since the NumPy object has to know about the provider, it can simply pass the release call on to it if appropriate. I don't see how this case necessitates making the release call on a different object. I'm -1 on involving any other objects or returning object references from the buffer interface, unless someone can come up with a use case which actually *requires* this (as opposed to it just being something which might be nice to have). The buffer interface should be Blazingly Fast(tm), and messing with PyObject*s is not the way to get that. The current proposal would be fast but would be more flexible for objects that don't have a memory representation that can be shared unless they create their own sharing object that perhaps copies the data into a contiguous chunk first. Objects which have memory which can be shared perfectly through the interface would simply pass themselves as the return value (after incrementing their extant buffers count by one). Seems to me the lock should apply to *everything* returned by getbuffer. If the requestor is going to iterate over the data, and there are multiple dimensions, surely it's going to want to refer to the shape and stride info at regular intervals while it's doing that. Requiring it to make its own copy would be a burden. There are two use cases that seem to be under discussion. 1) When you want to apply an algorithm to an arbitrary object that exposes the buffer interface 2) When you want to create an object that shares memory with another object exposing the buffer interface. These two use cases have slightly different needs. What I want to avoid is forcing the exporting object to be unable to change its shape and strides just because an object is using the memory for use case #2. I think the solution that states the shape and strides information are only guaranteed valid until the GIL is released is sufficent. Alternatively, one could release the shape and strides and format separately from the memory with a flag as a second argument to releasebuffer. -Travis -- Greg ___ Python-Dev mailing list [EMAIL PROTECTED] http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Carl Banks wrote: Tr ITSM that we are using the word view very differently. Consider this example: A = zeros((100,100)) B = A.transpose() You are thinking of NumPy's particular use case. I'm thinking of a generic use case. So, yes I'm using the word view in two different contexts. In this scenario, NumPy does not even use the buffer interface. It knows how to transpose it's own objects and does so by creating a new NumPy object (with it's own shape and strides space) with a data buffer pointed to by A. Yes, I use the word view for this NumPy usage, but only in the context of NumPy. In the PEP, I've been using the word view quite a bit more generically. So, I don't think this is a good example because A.transpose() will never call getbuffer of the A object (it will instead use the known structure of NumPy directly). So, let's talk about the generic situation instead of the NumPy specific one. I'd suggest the object returned by A.getbuffer should be called the buffer provider or something like that. I don't care what we call it. I've been using the word view because of the obvious analogy to my use of view in NumPy. When I had envisioned returning an actual object very similar to a NumPy array from the buffer interface it made a lot of sense to call it a view. Now, I'm fine to call it buffer provider For the sake of discussion, I'm going to avoid the word view altogether. I'll call A the exporter, as before. B I'll refer to as the requestor. The object returned by A.getbuffer is the provider. Fine. Let's use that terminology since it is new and not cluttered by other uses in other contexts. Having thought quite a bit about it, and having written several abortive replies, I now understand it and see the importance of it. getbuffer returns the object that you are to call releasebuffer on. It may or may not be the same object as exporter. Makes sense, is easy to explain. Yes, that's exactly all I had considered it to be. Only now, I'm wondering if we need to explicitly release a lock on the shape, strides, and format information as well as the buffer location information. It's easy to see possible use cases for returning a different object. A hypothetical future incarnation of NumPy might shift the responsibility of managing buffers from NumPy array object to a hidden raw buffer object. In this scenario, the NumPy object is the exporter, but the raw buffer object the provider. Considering this use case, it's clear that getbuffer should return the shape and stride data independently of the provider. The raw buffer object wouldn't have that information; all it does is store a pointer and keep a reference count. Shape and stride is defined by the exporter. So, who manages the memory to the shape and strides and isptr arrays? When a provider is created do these need to be created so that the shape and strides arrays are never deallocated when in use. The situation I'm considering is if you have a NumPy array of shape (2,3,3) which you then obtain a provider of (presumably from another package) and it retains a lock on the memory for a while. Should it also retain a lock on the shape and strides array? Can the NumPy array re-assign the shape and strides while the provider has still not been released? I would like to say yes, which means that the provider must supply it's own copy of shape and strides arrays. This could be the policy. Namely, that the provider must supply the memory for the shape, strides, and format arrays which is guaranteed for as long as a lock is held. In the case of NumPy, that provider could create it's own copy of the shape and strides arrays (or do it when the shape and strides arrays are re-assigned). Second question: what happens if a view wants to re-export the buffer? Do the views of the buffer ever change? Example, say you create a transposed view of a Numpy array. Now you want a slice of the transposed array. What does the transposed view's getbuffer export? Basically, you could not alter the internal representation of the object while views which depended on those values were around. In NumPy, a transposed array actually creates a new NumPy object that refers to the same data but has its own shape and strides arrays. With the new buffer protocol, the NumPy array would not be able to alter it's shape/strides/or reallocate its data areas while views were being held by other objects. But requestors could alter their own copies of the data, no? Back to the transpose example: B itself obviously can't use the same strides array as A uses. It would have to create its own strides, right? I don't like this example because B does have it's own strides because it is a complete NumPy array. I think we are talking about the same thing and that is who manages the memory for the shape and strides (and format). I think the
Re: [Python-Dev] Extended Buffer Interface/Protocol
(cc'ing back to Python-dev; the original reply was intended for it by I had an email malfunction.) Travis Oliphant wrote: Carl Banks wrote: 3. Allow getbuffer to return an array of derefence offsets, one for each dimension. For a given dimension i, if derefoff[i] is nonnegative, it's assumed that the current position (base pointer + indexing so far) is a pointer to a subarray, and derefoff[i] is the offest in that array where the current position goes for the next dimension. If derefoff[i] is negative, there is no dereferencing. Here is an example of how it'd work: This sounds interesting, but I'm not sure I totally see it. I probably need a picture to figure out what you are proposing. I'll get on it sometime. For now I hope an example will do. The derefoff sounds like some-kind of offset. Is that enough? Why not just make derefoff[i] == 0 instead of negative? I may have misunderstood something. I had thought the values exported by getbuffer could change as the view narrowed, but I'm not sure if it's the case now. I'll assume it isn't for now, because it simplifies things and demonstrates the concept better. Let's start from the beginning. First, change the prototype to this: typedef PyObject *(*getbufferproc)(PyObject *obj, void **buf, Py_ssize_t *len, int *writeable, char **format, int *ndims, Py_ssize_t **shape, Py_ssize_t **strides, int **isptr) isptr is a flag indicating whether, for a certain dimension, the positision we've strided to so far is a pointer that should be followed before proceeding with the rest of the strides. Now here's what a general get_item_pointer function would look like, given a set of indices: void* get_item_pointer(int ndim, void* buf, Py_ssize_t* strides, Py_ssize_t* derefoff, Py_ssize_t *indices) { char* pointer = (char*)buf; int i; for (i = 0; i ndim; i++) { pointer += strides[i]*indices[i]; if (isptr[i]) { pointer = *(char**)pointer; } } return (void*)pointer; } I don't fully understand the PIL example you gave. Yeah. How about more details. Here is a hypothetical image data object structure: struct rgba { unsigned char r, g, b, a; }; struct ImageObject { PyObject_HEAD; ... struct rgba** lines; Py_ssize_t height; Py_ssize_t width; Py_ssize_t shape_array[2]; Py_ssize_t stride_array[2]; Py_ssize_t view_count; }; lines points to malloced 1-D array of (struct rgba*). Each pointer in THAT block points to a seperately malloced array of (struct rgba). Got that? In order to access, say, the red value of the pixel at x=30, y=50, you'd use lines[50][30].r. So what does ImageObject's getbuffer do? Leaving error checking out: PyObject* getbuffer(PyObject *self, void **buf, Py_ssize_t *len, int *writeable, char **format, int *ndims, Py_ssize_t **shape, Py_ssize_t **strides, int **isptr) { static int _isptr[2] = { 1, 0 }; *buf = self-lines; *len = self-height*self-width; *writable = 1; *ndims = 2; self-shape_array[0] = height; self-shape_array[1] = width; *shape = self-shape_array; self-stride_array[0] = sizeof(struct rgba*); /* yep */ self-stride_array[1] = sizeof(struct rgba); *strides = self-stride_array; *isptr = _isptr; self-view_count ++; /* create and return view object here, but for what? */ } There are three essential differences from a regular, contiguous array. 1. buf is set to point at the array of pointers, not directly to the data. 2. The isptr thing. isptr[0] is true to indicate that the first dimension is an array of pointers, not the actual data. 3. stride[0] is sizeof(struct rgba*), not self-width*sizeof(struct rgba) like it would be for a contiguous array. This is because your first stride is through an array of pointers, not the data itself. So let's examine what get_item_pointer above will do given these values. Once again, we're looking for the pixel at x=30, y=50. First, we set pointer to buf, that is, self-lines. Then we take the first stride: we add index[0]+strides[0], that is, 50*4=200, to poitner. pointer now equals self-lines[50]. Now, we check isptr[0]. We see that it is true. Thus, the position we've strided to is, in fact, a pointer to a subarray where the actual data is. So we follow it: pointer = *pointer. pointer now equals self-lines[50] which equals self-lines[50][0]. Next dimension. We take the second stride: we add index[1]+strides[1], that is, 30*4=120, to pointer. pointer now equals self-lines[50][30]. Now, we check isptr[1]. It's false. No dereferencing this step. We're
Re: [Python-Dev] Extended Buffer Interface/Protocol
Carl Banks wrote: We're done. Return pointer. Thank you for this detailed example. I will have to parse it in more depth but I think I can see what you are suggesting. First, I'm not sure why getbuffer needs to return a view object. The view object in your case would just be the ImageObject. The reason I was thinking the function should return something is to provide more flexibility in what a view object actually is. I've also been going back and forth between explicitly passing all this information around or placing it in an actual view-object. In some sense, a view object is a NumPy array in my mind. But, with the addition of isptr we are actually expanding the memory abstraction of the view object beyond an explicit NumPy array. In the most common case, I envisioned the view object would just be the object itself in which case it doesn't actually have to be returned. While returning the view object would allow unspecified flexibilty in the future, it really adds nothing to the current vision. We could add a view object separately as an abstract API on top of the buffer interface. Second question: what happens if a view wants to re-export the buffer? Do the views of the buffer ever change? Example, say you create a transposed view of a Numpy array. Now you want a slice of the transposed array. What does the transposed view's getbuffer export? Basically, you could not alter the internal representation of the object while views which depended on those values were around. In NumPy, a transposed array actually creates a new NumPy object that refers to the same data but has its own shape and strides arrays. With the new buffer protocol, the NumPy array would not be able to alter it's shape/strides/or reallocate its data areas while views were being held by other objects. With the shape and strides information, the format information, and the data buffer itself, there are actually several pieces of memory that may need to be protected because they may be shared with other objects. This makes me wonder if releasebuffer should contain an argument which states whether or not to release the memory, the shape and strides information, the format information, or all of it. Having such a thing as a view object would actually be nice because it could hold on to a particular view of data with a given set of shape and strides (whose memory it owns itself) and then the exporting object would be free to change it's shape/strides information as long as the data did not change. The reason I ask is: if things like buf and strides and shape could change when a buffer is re-exported, then it can complicate things for PIL-like buffers. (How would you account for offsets in a dimension that's in a subarray?) I'm not sure what you mean, offsets are handled by changing the starting location of the pointer to the buffer. -Travis ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Travis Oliphant wrote: Carl Banks wrote: We're done. Return pointer. Thank you for this detailed example. I will have to parse it in more depth but I think I can see what you are suggesting. First, I'm not sure why getbuffer needs to return a view object. The view object in your case would just be the ImageObject. ITSM that we are using the word view very differently. Consider this example: A = zeros((100,100)) B = A.transpose() In this scenario, A would be the exporter object, I think we both would call it that. When I use the word view, I'm referring to B. However, you seem to be referring to the object returned by A.getbuffer, right? What term have you been using to refer to B? Obviously, it would help the discussion if we could get our terminology straight. (Frankly, I don't agree with your usage; it doesn't agree with other uses of the word view. For example, consider the proposed Python 3000 dictionary views: D = dict() V = D.items() Here, V is the view, and it's analogous to B in the above example.) I'd suggest the object returned by A.getbuffer should be called the buffer provider or something like that. For the sake of discussion, I'm going to avoid the word view altogether. I'll call A the exporter, as before. B I'll refer to as the requestor. The object returned by A.getbuffer is the provider. The reason I was thinking the function should return something is to provide more flexibility in what a view object actually is. I've also been going back and forth between explicitly passing all this information around or placing it in an actual view-object. In some sense, a view object is a NumPy array in my mind. But, with the addition of isptr we are actually expanding the memory abstraction of the view object beyond an explicit NumPy array. In the most common case, I envisioned the view object would just be the object itself in which case it doesn't actually have to be returned. While returning the view object would allow unspecified flexibilty in the future, it really adds nothing to the current vision. We could add a view object separately as an abstract API on top of the buffer interface. Having thought quite a bit about it, and having written several abortive replies, I now understand it and see the importance of it. getbuffer returns the object that you are to call releasebuffer on. It may or may not be the same object as exporter. Makes sense, is easy to explain. It's easy to see possible use cases for returning a different object. A hypothetical future incarnation of NumPy might shift the responsibility of managing buffers from NumPy array object to a hidden raw buffer object. In this scenario, the NumPy object is the exporter, but the raw buffer object the provider. Considering this use case, it's clear that getbuffer should return the shape and stride data independently of the provider. The raw buffer object wouldn't have that information; all it does is store a pointer and keep a reference count. Shape and stride is defined by the exporter. Second question: what happens if a view wants to re-export the buffer? Do the views of the buffer ever change? Example, say you create a transposed view of a Numpy array. Now you want a slice of the transposed array. What does the transposed view's getbuffer export? Basically, you could not alter the internal representation of the object while views which depended on those values were around. In NumPy, a transposed array actually creates a new NumPy object that refers to the same data but has its own shape and strides arrays. With the new buffer protocol, the NumPy array would not be able to alter it's shape/strides/or reallocate its data areas while views were being held by other objects. But requestors could alter their own copies of the data, no? Back to the transpose example: B itself obviously can't use the same strides array as A uses. It would have to create its own strides, right? So, what if someone takes a slice out of B? When calling B.getbuffer, does it get B's strides, or A's? I think it should get B's. After all, if you're taking a slice of B, don't you care about the slicing relative to B's axes? I can't really think of a use case for exporting A's stride data when you take a slice of B, and it doesn't seem to simplify memory management, because B has to make it's own copies anyways. With the shape and strides information, the format information, and the data buffer itself, there are actually several pieces of memory that may need to be protected because they may be shared with other objects. This makes me wonder if releasebuffer should contain an argument which states whether or not to release the memory, the shape and strides information, the format information, or all of it. Here's what I think: the lock should only apply to the buffer itself, and not to shape and stride data at
Re: [Python-Dev] Extended Buffer Interface/Protocol
Carl Banks wrote: It's easy to see possible use cases for returning a different object. A hypothetical future incarnation of NumPy might shift the responsibility of managing buffers from NumPy array object to a hidden raw buffer object. In this scenario, the NumPy object is the exporter, but the raw buffer object the provider. But since the NumPy object has to know about the provider, it can simply pass the release call on to it if appropriate. I don't see how this case necessitates making the release call on a different object. I'm -1 on involving any other objects or returning object references from the buffer interface, unless someone can come up with a use case which actually *requires* this (as opposed to it just being something which might be nice to have). The buffer interface should be Blazingly Fast(tm), and messing with PyObject*s is not the way to get that. This makes me wonder if releasebuffer should contain an argument which states whether or not to release the memory, the shape and strides information, the format information, or all of it. Here's what I think: the lock should only apply to the buffer itself, and not to shape and stride data at all. If the requestor wants to keep its own copies of the data, it would have to malloc its own storage for it. I expect that this would be very rare. Seems to me the lock should apply to *everything* returned by getbuffer. If the requestor is going to iterate over the data, and there are multiple dimensions, surely it's going to want to refer to the shape and stride info at regular intervals while it's doing that. Requiring it to make its own copy would be a burden. -- Greg ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Travis Oliphant wrote: Carl Banks wrote: Tr ITSM that we are using the word view very differently. Consider this example: A = zeros((100,100)) B = A.transpose() You are thinking of NumPy's particular use case. I'm thinking of a generic use case. So, yes I'm using the word view in two different contexts. In this scenario, NumPy does not even use the buffer interface. It knows how to transpose it's own objects and does so by creating a new NumPy object (with it's own shape and strides space) with a data buffer pointed to by A. I realized that as soon as I tried a simple Python demonstration of it. So it's a poor example. But I hope it's obvious how it would generalize to a different type. Having such a thing as a view object would actually be nice because it could hold on to a particular view of data with a given set of shape and strides (whose memory it owns itself) and then the exporting object would be free to change it's shape/strides information as long as the data did not change. What I don't undestand is why it's important for the provider to retain this data. The requestor only needs the information when it's created; it will calculate its own versions of the data, and will not need the originals again, so no need to the provider to keep them around. That is certainly a policy we could enforce (and pretty much what I've been thinking). We just need to make it explicit that the shape and strides provided is only guaranteed up until a GIL release (i.e. arbitrary Python code could change these memory areas both their content and location) and so if you need them later, make your own copies. If this were the policy, then NumPy could simply pass pointers to its stored shape and strides arrays when the buffer interface is called but then not worry about re-allocating these arrays before the buffer lock is released. Another object could hold on to the memory area of the NumPy array but would have to store shape and strides information if it wanted to keep it. NumPy could also just pass a pointer to the char * representation of the format (which in NumPy would be stored in the dtype object) and would not have to worry about the dtype being re-assigned later. Bingo! This is my preference. The reason I ask is: if things like buf and strides and shape could change when a buffer is re-exported, then it can complicate things for PIL-like buffers. (How would you account for offsets in a dimension that's in a subarray?) I'm not sure what you mean, offsets are handled by changing the starting location of the pointer to the buffer. But to anwser your question: you can't just change the starting location because there's hundreds of buffers. You'd either have to change the starting location of each one of them, which is highly undesirable, or to factor in an offset somehow. (This was, in fact, the point of the derefoff term in my original suggestion.) I get better what your derefoff was doing now. I was missing the de-referencing that was going on. Couldn't you still just store a pointer to the start of the array. In other words, isn't your **isptr suggestion sufficient? It seems to be. No. The problem arises when slicing. In a single buffer, you would adjust the base pointer to point at the element [0,0] of the slice. But you can't do that with multiple buffers. Instead, you have to add an offset after deferencing the pointer to the subarray. Hence my derefoff proposal. It dereferenced the pointer, then added an offset to get you to the 0 position in that dimension. Anyways, despite the miscommunications so far, I now have a very good idea of what's going on. We definitely need to get terms straight. I agree that getbuffer should return an object. I think we need to think harder about the case when requestors re-export the buffer. (Maybe it's time to whip up some experimental objects?) I'm still not clear what you are concerned about. If an object consumes the buffer interface and then wants to be able to later export it to another, then from our discussion about the shape/strides and format information, it would have to maintain it's own copies of these things, because it could not rely on the original provider (or exporter) to keep them around once the GIL is released. Right. So, if someone calls getbuffer, it would send its own copies of the buffer information, and not the original exporter's. The values returned by getbuffer can vary for a given buffer, depending on the exporter. Which means the data returned by getbuffer could reflect slicing. Which means the isptr array is not sufficient for the PIL-style multiple buffers. This is the reason we would have to be very clear about the guaranteed persistance of the shape/strides and format memory whose pointers are returned through the proposed buffer interface.
Re: [Python-Dev] Extended Buffer Interface/Protocol
I have developed a split vector type that implements the buffer protocol at http://scintilla.sourceforge.net/splitvector-1.0.zip It acts as a mutable string implementing most of the sequence protocol as well as the buffer protocol. splitvector.SplitVector('c') creates a vector containing 8 bit characters and splitvector.SplitVector('u') is for Unicode. A writable attribute bufferAppearence can be set to 0 (default) to respond to buffer protocol calls by moving the gap to the end and returning the address of all of the data. Setting bufferAppearence to 1 responds as a two segment buffer. I haven't found any code that understands responding with two segments. sre and file.write handle SplitVector fine when it responds as a single segment: import re, splitvector x = splitvector.SplitVector(c) x[:] = The life of brian r = re.compile(l[a-z]*, re.M) print x y = r.search(x) print y.group(0) x.bufferAppearence = 1 y = r.search(x) print y.group(0) produces The life of brian life Traceback (most recent call last): File qt.py, line 9, in module y = r.search(x) TypeError: expected string or buffer It is likely that adding multi-segment ability to sre would complexify and slow it down. OTOH multi-segment buffers may be well-suited to scatter/gather I/O calls like writev. Neil ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Greg Ewing wrote: Travis Oliphant wrote: I'm talking about arrays of pointers to other arrays: i.e. if somebody defined in C float B[10][20] then B would B an array of pointers to arrays of floats. No, it wouldn't, it would be a contiguously stored 2-dimensional array of floats. An array of pointers would be float *B[10]; followed by code to allocate 10 arrays of 20 floats each and initialise B to point to them. You are right, of course, that example was not correct. I think the point is still valid, though. One could still use the shape to indicate how many levels of pointers-to-pointers there are (i.e. how many pointer dereferences are needed to select out an element). Further dimensionality could then be reported in the format string. This would not be hard to allow. It also would not be hard to write a utility function to copy such shared memory into a contiguous segment to provide a C-API that allows casual users to avoid the details of memory layout when they are writing an algorithm that just uses the memory. I can imagine cases like that coming up in practice. For example, an image object might store its data as four blocks of memory for R, G, B and A planes, each of which is a contiguous 2d array with shape and stride -- but you want to view it as a 3d array byte[plane][x][y]. All we can do is have the interface actually be able to describe it's data. Users would have to take that information and write code accordingly. In this case, for example, one possibility is that the object would raise an error if strides were requested. It would also raise an error if contiguous data was requested (or I guess it could report the R channel only if it wanted to). Only if segments were requested could it return an array of pointers to the four memory blocks. It could then report itself as a 2-d array of shape (4, H) where H is the height. Each element of the array would be reported as %sB % W where W is the width of the image (i.e. each element of the 2-d array would be a 1-d array of length W. Alternatively it could report itself as a 1-d array of shape (4,) with elements (H,W)B A user would have to write the algorithm correctly in order to access the memory correctly. Alternatively, a utility function that copies into a contiguous buffer would allow the consumer to not care about exactly how the memory is layed out. But, the buffer interface would allow the utility function to figure it out and do the right thing for each exporter. This flexibility would not be available if we don't allow for segmented memory in the buffer interface. So, I don't think it's that hard to at least allow the multiple-segment idea into the buffer interface (as long as all the segments are the same size, mind you). It's only one more argument to the getbuffer call. -Travis ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
[Python-Dev] Extended Buffer Interface/Protocol
I'm soliciting feedback on my extended buffer protocol that I am proposing for inclusion in Python 3000. As soon as the Python 3.0 implementation is complete, I plan to back-port the result to Python 2.6, therefore, I think there may be some interest on this list. Basically, the extended buffer protocol seeks to allow memory sharing with 1) information about what is in the memory (float, int, C-structure, etc.) 2) information about the shape of the memory (if any) 3) information about discontiguous memory segments Number 3 is where I could use feedback --- especially from PIL users and developers. Strides are a common way to think about a possibly discontiguous chunk of memory (which appear in NumPy when you select a sub-region from a larger array). The strides vector tells you how many bytes to skip in each dimension to get to the next memory location for that dimension. Because NumPy uses this memory model as do several compute libraries (like BLAS and LAPACK), it makes sense to allow this memory model to be shared between objects in Python. Currently, the proposed buffer interface eliminates the multi-segment option (for Python 3.0) which I think was originally put in place because of the PIL. However, I don't know if it is actually used by any extension types. This is a big reason why Guido wants to drop the multi-segment interface option. The question is should we eliminate the possibility of sharing memory for objects that store data basically as arrays of arrays (i.e. true C-style arrays). That is what I'm currently proposing, but I could also see an argument that states that if we are going to support strided memory access, we should also support array of array memory access. If this is added, then it would be another function-call that gets a array-of-array-style memory from the object. What do others think of these ideas? One possible C-API call that Python could grow with the current buffer interface is to allow contiguous-memory mirroring of discontiguous memory, or an iterator object that iterates through every element of any object that exposes the buffer protocol. Thanks for any feedback, -Travis Oliphant ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Attached is the PEP. :PEP: XXX :Title: Revising the buffer protocol :Version: $Revision: $ :Last-Modified: $Date: $ :Author: Travis Oliphant [EMAIL PROTECTED] :Status: Draft :Type: Standards Track :Content-Type: text/x-rst :Created: 28-Aug-2006 :Python-Version: 3000 Abstract This PEP proposes re-designing the buffer API (PyBufferProcs function pointers) to improve the way Python allows memory sharing in Python 3.0 In particular, it is proposed that the multiple-segment and character buffer portions of the buffer API be eliminated and additional function pointers be provided to allow sharing any multi-dimensional nature of the memory and what data-format the memory contains. Rationale = The buffer protocol allows different Python types to exchange a pointer to a sequence of internal buffers. This functionality is *extremely* useful for sharing large segments of memory between different high-level objects, but it is too limited and has issues. 1. There is the little (never?) used sequence-of-segments option (bf_getsegcount) 2. There is the apparently redundant character-buffer option (bf_getcharbuffer) 3. There is no way for a consumer to tell the buffer-API-exporting object it is finished with its view of the memory and therefore no way for the exporting object to be sure that it is safe to reallocate the pointer to the memory that it owns (for example, the array object reallocating its memory after sharing it with the buffer object which held the original pointer led to the infamous buffer-object problem). 4. Memory is just a pointer with a length. There is no way to describe what is in the memory (float, int, C-structure, etc.) 5. There is no shape information provided for the memory. But, several array-like Python types could make use of a standard way to describe the shape-interpretation of the memory (wxPython, GTK, pyQT, CVXOPT, PyVox, Audio and Video Libraries, ctypes, NumPy, data-base interfaces, etc.) 6. There is no way to share discontiguous memory (except through the sequence of segments notion). There are two widely used libraries that use the concept of discontiguous memory: PIL and NumPy. Their view of discontiguous arrays is different, though. This buffer interface allows sharing of either memory model. Exporters will only use one approach and consumers may choose to support discontiguous arrays of each type however they choose. NumPy uses the notion of constant striding in each dimension as its basic concept of an array. With this concept, a simple sub-region of a larger array can be described without copying the data. T Thus, stride information is the additional information that must be shared. The PIL uses a more opaque memory representation. Sometimes an image is contained in a contiguous segment of memory, but sometimes it is contained in an array of pointers to the contiguous segments (usually lines) of the image. The PIL is where the idea of multiple buffer segments in the original buffer interface came from. NumPy's strided memory model is used more often in computational libraries and because it is so simple it makes sense to support memory sharing using this model. The PIL memory model is used often in C-code where a 2-d array can be then accessed using double pointer indirection: e.g. image[i][j]. The buffer interface should allow the object to export either of these memory models. Consumers are free to either require contiguous memory or write code to handle either memory model. Proposal Overview = * Eliminate the char-buffer and multiple-segment sections of the buffer-protocol. * Unify the read/write versions of getting the buffer. * Add a new function to the interface that should be called when the consumer object is done with the view. * Add a new variable to allow the interface to describe what is in memory (unifying what is currently done now in struct and array) * Add a new variable to allow the protocol to share shape information * Add a new variable for sharing stride information * Add a new mechanism for sharing array of arrays. * Fix all objects in the core and the standard library to conform to the new interface * Extend the struct module to handle more format specifiers Specification = Change the PyBufferProcs structure to :: typedef struct { getbufferproc bf_getbuffer releasebufferproc bf_releasebuffer } :: typedef PyObject *(*getbufferproc)(PyObject *obj, void **buf, Py_ssize_t *len, int *writeable, char **format, int *ndims, Py_ssize_t **shape, Py_ssize_t **strides, void **segments) All variables except the
Re: [Python-Dev] Extended Buffer Interface/Protocol
Travis Oliphant: 3) information about discontiguous memory segments Number 3 is where I could use feedback --- especially from PIL users and developers. Strides are a common way to think about a possibly discontiguous chunk of memory (which appear in NumPy when you select a sub-region from a larger array). The strides vector tells you how many bytes to skip in each dimension to get to the next memory location for that dimension. I think one of the motivations for discontiguous segments was for split buffers which are commonly used in text editors. A split buffer has a gap in the middle where insertions and deletions can often occur without moving much memory. When an insertion or deletion is required elsewhere then the gap is first moved to that position. I have long intended to implement a good split buffer extension for Python but the best I have currently is an extension written using Boost.Python which doesn't implement the buffer interface. Here is a description of split buffers: http://www.cs.cmu.edu/~wjh/papers/byte.html Neil ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
On 3/21/07, Neil Hodgson [EMAIL PROTECTED] wrote: Travis Oliphant: 3) information about discontiguous memory segments Number 3 is where I could use feedback --- especially from PIL users and developers. Strides are a common way to think about a possibly discontiguous chunk of memory (which appear in NumPy when you select a sub-region from a larger array). The strides vector tells you how many bytes to skip in each dimension to get to the next memory location for that dimension. I think one of the motivations for discontiguous segments was for split buffers which are commonly used in text editors. A split buffer has a gap in the middle where insertions and deletions can often occur without moving much memory. When an insertion or deletion is required elsewhere then the gap is first moved to that position. I have long intended to implement a good split buffer extension for Python but the best I have currently is an extension written using Boost.Python which doesn't implement the buffer interface. Here is a description of split buffers: http://www.cs.cmu.edu/~wjh/papers/byte.html But there's always a call to remove the gap (or move it to the end). -- --Guido van Rossum (home page: http://www.python.org/~guido/) ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Neil Hodgson wrote: I think one of the motivations for discontiguous segments was for split buffers which are commonly used in text editors. Note that this is different from the case of an array of pointers to arrays, which is a multi-dimensional array structure, whereas a split buffer is a concatenation of two (possibly different sized) one-dimensional arrays. So an array-of-pointers interface wouldn't be a direct substitute for the existing multi-segment buffer interface. -- Greg ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Travis Oliphant wrote: The question is should we eliminate the possibility of sharing memory for objects that store data basically as arrays of arrays (i.e. true C-style arrays). Can you clarify what you mean by this? Are you talking about an array of pointers to other arrays? (This is not what I would call an array of arrays, even in C.) Supporting this kind of thing could be a slippery slope, since there can be arbitrary levels of complexity to such a structure. E.g do you support a 1d array of pointers to 3d arrays of pointers to 2d arrays? Etc. The more different kinds of format you support, the less likely it becomes that the thing consuming the data will be willing to go to the trouble required to understand it. One possible C-API call that Python could grow with the current buffer interface is to allow contiguous-memory mirroring of discontiguous memory, I don't think the buffer protocol itself should incorporate anything that requires implicitly copying the data, since the whole purpose of it is to provide direct access to the data without need for copying. It would be okay to supply some utility functions for re-packing data, though. or an iterator object that iterates through every element of any object that exposes the buffer protocol. Again, for efficiency reasons I wouldn't like to involve Python objects and iteration mechanisms in this. The buffer interface is meant to give you raw access to the data at raw C speeds. Anything else is outside its scope, IMO. -- Greg ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Greg Ewing: So an array-of-pointers interface wouldn't be a direct substitute for the existing multi-segment buffer interface. Providing an array of (pointer,length) wouldn't be too much extra work for a split vector implementation. Guido van Rossum: But there's always a call to remove the gap (or move it to the end). Yes, although its something you try to avoid. I'm not saying that this is an important use-case since no one seems to have produced a split vector implementation that provides the buffer protocol. Numeric-style array handling is much more common so deserves priority. Neil ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Greg Ewing wrote: Travis Oliphant wrote: The question is should we eliminate the possibility of sharing memory for objects that store data basically as arrays of arrays (i.e. true C-style arrays). Can you clarify what you mean by this? Are you talking about an array of pointers to other arrays? (This is not what I would call an array of arrays, even in C.) I'm talking about arrays of pointers to other arrays: i.e. if somebody defined in C float B[10][20] then B would B an array of pointers to arrays of floats. Supporting this kind of thing could be a slippery slope, since there can be arbitrary levels of complexity to such a structure. E.g do you support a 1d array of pointers to 3d arrays of pointers to 2d arrays? Etc. Yes, I saw that. But, it could actually be supported, in general. The shape information is available. If a 3-d array is meant then ndims is 3 and you would re-cast the returned pointer appropriately. In other words, suppose that instead of strides you can request a variable through the buffer interface with type void **segments. Then, by passing the address to a void * variable to the routine you would receive the array. Then, you could handle 1-d, 2-d, and 3-d cases using something like this: This is pseudocode: void *segments; int ndims; Py_ssize_t *shape; char *format; (ndims, shape, format, and segments) are passed to the buffer interface. if strcmp(format, f) != 0 raise an error. if (ndims == 1) var = (float *)segments for (i=0; ishape[0]; i++) # process var[i] else if (ndims == 2) var = (float **)segments for (i=0; ishape[0]; i++) for (j=0; jshape[1]; j++) # process var[i][j] else if (ndims == 3) var = (float ***)segments for (i=0; ishape[0]; i++) for (j=0; jshape[1]; j++) for (k=0; jshape[2]; k++) # process var[i][j][k] else raise an Error. The more different kinds of format you support, the less likely it becomes that the thing consuming the data will be willing to go to the trouble required to understand it. That is certainly true. I'm really only going through the trouble, since the multiple segment already exists and the PIL has this memory model (although I have not heard PIL developers clamoring for support, --- I'm just being sensitive to that extension type). One possible C-API call that Python could grow with the current buffer interface is to allow contiguous-memory mirroring of discontiguous memory, I don't think the buffer protocol itself should incorporate anything that requires implicitly copying the data, since the whole purpose of it is to provide direct access to the data without need for copying. No, this would not be the buffer protocol, but merely a C-API that would use the buffer protocol - i.e. it is just a utility function as you mention. It would be okay to supply some utility functions for re-packing data, though. or an iterator object that iterates through every element of any object that exposes the buffer protocol. Again, for efficiency reasons I wouldn't like to involve Python objects and iteration mechanisms in this. I was thinking more of a C-iterator, like NumPy provides. This can be very efficient (as long as the loop is not in Python). It sure provides a nice abstraction that lets you deal with discontiguous arrays as if they were contiguous, though. The buffer interface is meant to give you raw access to the data at raw C speeds. Anything else is outside its scope, Sure. These things are just ideas about *future* utility functions that might make use of the buffer interface and motivate its design. Thanks for your comments. -Travis ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com
Re: [Python-Dev] Extended Buffer Interface/Protocol
Travis Oliphant wrote: I'm talking about arrays of pointers to other arrays: i.e. if somebody defined in C float B[10][20] then B would B an array of pointers to arrays of floats. No, it wouldn't, it would be a contiguously stored 2-dimensional array of floats. An array of pointers would be float *B[10]; followed by code to allocate 10 arrays of 20 floats each and initialise B to point to them. Yes, I saw that. But, it could actually be supported, in general. Certainly it *can* be supported, but the question is how many different format variations it's reasonable to expect the consumer of the data to be able to deal with. Because each variation requires the consumer to use different code to access the data, if it wants to avoid making a copy. else if (ndims == 3) var = (float ***)segments for (i=0; ishape[0]; i++) for (j=0; jshape[1]; j++) for (k=0; jshape[2]; k++) # process var[i][j][k] This assumes that the 3-dimensional case is using the array-of-pointers implementation at all levels. But there are other possibilities, e.g. a 1d array of pointers to contiguous 2d arrays, or a contiguous 2d array of pointers to 1d arrays. It's hard to deal with all of those using a common piece of code. I can imagine cases like that coming up in practice. For example, an image object might store its data as four blocks of memory for R, G, B and A planes, each of which is a contiguous 2d array with shape and stride -- but you want to view it as a 3d array byte[plane][x][y]. (Actually you'd probably *prefer* to view it as byte[x][y][plane], which would make things even more difficult...) I was thinking more of a C-iterator, like NumPy provides. This can be very efficient (as long as the loop is not in Python). It sure provides a nice abstraction that lets you deal with discontiguous arrays as if they were contiguous, though. Something like that might be useful. -- Greg Ewing, Computer Science Dept, +--+ University of Canterbury, | Carpe post meridiem! | Christchurch, New Zealand | (I'm not a morning person.) | [EMAIL PROTECTED] +--+ ___ Python-Dev mailing list Python-Dev@python.org http://mail.python.org/mailman/listinfo/python-dev Unsubscribe: http://mail.python.org/mailman/options/python-dev/archive%40mail-archive.com