En Wed, 21 Mar 2007 01:30:34 -0300, John Pye <[EMAIL PROTECTED]> escribió:
> On Mar 21, 3:15 pm, "Gabriel Genellina" <[EMAIL PROTECTED]> > wrote: >> En Wed, 21 Mar 2007 00:46:03 -0300, John Pye >> <[EMAIL PROTECTED]> escribió: >> >> > This is not an option for me, as I want to pass the >> > FILE* from Python and all the way into into a existing shared-library >> > code. I can't change the latter; it must work ok with standard fprintf >> > (etc) functions. >> >> You can get the file descriptor from the Python file object using its >> fileno() method. The file descriptor lives at the OS level, so it's safe >> to pass around. You can regenerate a new FILE struct (using the other >> runtime library) with fdopen. >> >> -- >> Gabriel Genellina > > Hi Gabriel > > Are you sure about this? My understanding is that 'fileno' is just a > FILE* cast to an integer. So it's a pointer to something (check out > stdio.h to see what it points to). Problem is (AFAICT) that Python 2.4 A FILE struct usually *contains* a file descriptor, among other things. But I think the FILE struct is opaque and not specified. Perhaps one should go even at a lower level, using open_osfhandle and get_osfhandle. (From python you can use the msvcrt module). See http://msdn2.microsoft.com/en-us/library/kdfaxaay(VS.71).aspx h=get_osfhandle(f.fileno()) returns a Windows file handle that you can pass around. On the Mingw side: fd=open_osfhandle(h); FILE *f=fdopen(fd) > uses a different version of the C runtime DLL (MSVCRT*.DLL) to that > which MinGW links against. And it turns out that the different C > runtime libraries have incompatible implementations of the FILE > struct. And therefore if you try to pass a FILE* (fileno?) from Python > to MinGW you will get a segfault. Exactly; the idea is not to pass a foreign FILE struct, but to *contruct* another one based on an existing open file; this is what fdopen does. > If there is more to your suggestion that I'm not seeing, please > clarify. Have you tried this with MinGW actually? No, but I'm rather confident it should work... as always, it can fail miserably :) -- Gabriel Genellina -- http://mail.python.org/mailman/listinfo/python-list
