On Sat, 17 Jan 2009 12:57:24 -0800, Andrei Alexandrescu <[email protected]> wrote:
>Max Samukha wrote: >> On Sat, 17 Jan 2009 09:19:38 -0800, Andrei Alexandrescu >> <[email protected]> wrote: >> >>> Max Samukha wrote: >>>> On Sat, 17 Jan 2009 16:09:43 +0200, Max Samukha >>>> <[email protected]> wrote: >>>> >>>>> rebuild is pretty unusable with dmd 2.023, so I've taken the liberty >>>>> to port a popen implementation to D for Windows and modify rdmd to use >>>>> that. >>>>> >>>>> The handle returned by the popen st >>>> ..ill fails at the end of stream (maybe due to this issue >>>> http://www.digitalmars.com/d/archives/c++/windows/32-bits/274.html), >>>> so the exception is simply ignored for now. >>>> >>>> The rdmd.d patch and popen are here >>>> http://d-coding.com/download/rdmd.zip >>> That's great! Is it ok with you if I take the popen implementation and >>> do the rdmd integration by hand? I made a couple of other changes to >>> rdmd and I'm not sure how patch would cope with that. Also, please let >>> me know if you allow me to change popen a bit to e.g make symbols more >>> local and add documentation. Finally, please let me know if I can >>> publish your implementation in Phobos and use it in other functions such >>> as shell(). >>> >>> Thanks, >>> >>> Andrei >> >> Yep. But there are numerous issues with this implementation >> >> 1. The bad descriptor error. >> 2. pclose is not implemented accoring to the standard. There is a way >> to partially fix this by storing the process ID in the descriptor. >> 3. 0 is passed to the mode parameter of _open_osfhandle. It should >> probably be _O_BINARY if the runtime cares at all. >> 4. This implementaion is based on a number of existing ones so the >> owners of original versions may be bad enough to claim their rights. >> >> So it's far from being a perfect popen :) > >If your implementation relies heavily on others I think there is no >question that you need to ask them for permission and also credit them >in your implementation (whether they ask for that or not). Until then I >can't do much. > >Andrei I wouldn't say it relies on them heavily. It uses the same pretty obvious concept: create a pipe and pass its handles to the child process. I don't think it makes much sense to ask permission from developers of MSVC, cygwin and a dozen more, who came up with similar implementations (imperfect in their own ways). Either I credit them all or nobody. I choose the latter for the sake of brevity and put the code in public domain. Meanwhile, I've updated the implementation to use array slices and a different way of getting to the command interpreter. I tried to use __tmpnum field of _iobuf to store the process handle and it seemed to work but then, having no idea of the field's exact purpose, I simply introduced a global vector that maps file descriptors to process handles, meaning popen is now not thread-safe, which is ok. _open_osfhandle doesn't accept the write end of the pipe, so the function will always fail for "w" mode. As I don't have the sources for digitalmars C runtime I cannot tell what's going on. And it would be really nice if Walter implemented a popen in the C runtime where it belongs. Thanks! The updated version is here http://d-coding.com/download/rdmd.zip. cpopen is for ANSI and UTF-16 versions with C linkage and dpopen contains a wrapper taking UTF-8 strings.
