Re: [racket-dev] [racket] newbie: foreign C structure definition question
[redirected to dev again] 15 minutes ago, Hendrik Boom wrote: On Tue, Nov 16, 2010 at 11:37:17PM -0500, Eli Barzilay wrote: Three minutes ago, Jay McCarthy wrote: We could use a syntax parameter in define-cstruct to communicate to the vector syntax to give them the correct size in structs and make my macro's approach the normal behavior. * Types in a cstruct are plain expressions (that should evaluate to ctype values). * Arrays should be available as a type outside of cstructs too. An array of, say, ten elements should be passes the same way as a struct with the same ten elements. Yes, what I said earlier is that it really boils down to being able to define a libffi type descriptor with some manually specified size and alignment. It's also how the current hackarounds all work. That C doesn't do this is a flaw in its calling interface. And that actually is another problem for Sam when he implements this -- dealing with whatever situations make C expect the array to be passed by value (a real one, with the whole array on the stack). But given that it's possible to pass references, it should be possible to pass references to C arrays when the C funtion is coded as if it expects an array (when, of course, it really expects a reference). Right, this works assuming sane C, and IIRC, with small arrays it can expect the whole array to be on the stack. (But maybe the rules for structs are the same, so it works out by size only.) -- ((lambda (x) (x x)) (lambda (x) (x x))) Eli Barzilay: http://barzilay.org/ Maze is Life! _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
Re: [racket-dev] [racket] newbie: foreign C structure definition question
On Thu, Nov 18, 2010 at 10:06:34PM -0500, Eli Barzilay wrote: [redirected to dev again] 15 minutes ago, Hendrik Boom wrote: On Tue, Nov 16, 2010 at 11:37:17PM -0500, Eli Barzilay wrote: Three minutes ago, Jay McCarthy wrote: We could use a syntax parameter in define-cstruct to communicate to the vector syntax to give them the correct size in structs and make my macro's approach the normal behavior. * Types in a cstruct are plain expressions (that should evaluate to ctype values). * Arrays should be available as a type outside of cstructs too. An array of, say, ten elements should be passes the same way as a struct with the same ten elements. Yes, what I said earlier is that it really boils down to being able to define a libffi type descriptor with some manually specified size and alignment. It's also how the current hackarounds all work. That C doesn't do this is a flaw in its calling interface. And that actually is another problem for Sam when he implements this -- dealing with whatever situations make C expect the array to be passed by value (a real one, with the whole array on the stack). But given that it's possible to pass references, it should be possible to pass references to C arrays when the C funtion is coded as if it expects an array (when, of course, it really expects a reference). Right, this works assuming sane C, and IIRC, with small arrays it can expect the whole array to be on the stack. (But maybe the rules for structs are the same, so it works out by size only.) C defines that when an array is the actual parameter, the address of it is passed instead. It depends on the fact that the type of the argunment is an array and nothing else. If you pass a one-element array of integer (note the singular), you're supposed to get its address instead of the array itself. Structures are passed in their entirety. Of course, some compilers when functions with prototypes are called, may do incredibly convoluted things instead to attain greater efficiency. FFI would presumably have to match that complexity in terms of hardware register hacking. -- hendrik _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
Re: [racket-dev] [racket] newbie: foreign C structure definition question
On Wed, Nov 17, 2010 at 2:12 AM, Eli Barzilay e...@barzilay.org wrote: Two hours ago, Sam Tobin-Hochstadt wrote: Python's 'ctypes' library, which also uses libffi under the hood, has arrays: http://docs.python.org/library/ctypes.html#arrays and also unions, which we don't have: http://docs.python.org/library/ctypes.html#structures-and-unions So it seems like this should be possible, and maybe we can even use their code. They guy who did ctypes knows how to deal with the libffi internals. (He did the msvc version of the code.) If you look through the libffi documentation you'll see no mention of arrays, and an explicit mention of no special support for unions. I understand that, but ctypes proves 3 things - it can be done - there is code to read and people to ask to figure out how - Python(!) has a nicer FFI than us for this. -- sam th sa...@ccs.neu.edu _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
Re: [racket-dev] [racket] newbie: foreign C structure definition question
6 hours ago, Jay McCarthy wrote: I may be parsing your responses wrong Eli, but I don't think we should let our use of libffi be a barrier to a good ffi. I see (require ffi) to mean get a good Racket FFI, not get a Racket encoding of libffi. Of course. I even detailed a number of times how it could be done: what you need is to find a hook in libffi that allows you to create new types with a manually specified size and alignment, then use that from the racket side. That does require some digging, but it sounds like Sam has already done some, so you just need to wait for a patch from him. -- ((lambda (x) (x x)) (lambda (x) (x x))) Eli Barzilay: http://barzilay.org/ Maze is Life! _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
Re: [racket-dev] [racket] newbie: foreign C structure definition question
[redirected to d...@] On Tue, Nov 16, 2010 at 11:28 PM, Eli Barzilay e...@barzilay.org wrote: 20 minutes ago, Jay McCarthy wrote: I've complained about this before, which is why I was able to quickly answer your question. I don't endorse the current behavior. (Just a reminder: this is something that libffi doesn't have AFAICT.) Python's 'ctypes' library, which also uses libffi under the hood, has arrays: http://docs.python.org/library/ctypes.html#arrays and also unions, which we don't have: http://docs.python.org/library/ctypes.html#structures-and-unions So it seems like this should be possible, and maybe we can even use their code. -- sam th sa...@ccs.neu.edu _ For list-related administrative tasks: http://lists.racket-lang.org/listinfo/dev
