[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [Discuss-gnuradio] anyone else having trouble compiling fusb_linux.c
Re: [Discuss-gnuradio] anyone else having trouble compiling fusb_linux.cc?
Tue, 13 Sep 2005 09:47:45 -0700
On Mon, Sep 12, 2005 at 08:33:09PM -0700, Eric Blossom wrote:
> On Thu, Sep 08, 2005 at 11:04:33AM -0700, Larry Doolittle wrote:
> > The libusb folks maintain a libusb/linux.h that parallels
> > the kernel-space linux/usbdevice_fs.h, but is designed to
> > be used by user-space programs. So I grabbed a copy of
> > that (and called it libusb_linux.h), and #include it instead.
> Which kernel did this stuff blow up on?
In Debian's case, /usr/include/linux/* comes not from the kernel,
but from the linux-kernel-headers package. The broken situation
uses linux-kernel-headers-2.6.13+0rc3-1.1, the stable working
case is linux-kernel-headers-2.5.999-test7-bk-17. Sounds scary
already, since the kernel versions actually running on those two
systems are 2.6.11 and 2.6.8, respectively.
> Although I'm all for solving the problem, importing an uninstalled
> file from libusb into the gnuradio code base that "better" defines the
> user/kernel interface seem like a path to future fragility. Not that
> I'm sooo pure that I wouldn't do this to fix the problem ;)
> It seems like the right answer is a fix to the kernel header, or
> create a supported user-space include that ships with the kernel and
> defines the ioctl interface.
This category of problems including Linux kernel headers is not new.
Several are tracked as Debian bugs to the linux-kernel-headers package,
Our particular problem might be AMD64 specific, as reported in
I recall discussions of similar problems back in the kernel 2.0 days.
> Can we make the problem go away by (conditionally) including some
> other kernel include file before linux/usbdevice_fs.h? (Yes, I know
> it's spreading evil...)
I doubt if there is a method which is reliable in the face
of all the various flavors of Linux development system people
use these days. User-space inclusion of Linux header files is
apparently not in Linus's regression suite.
Ask yourself, why did the libusb decide to make and use their
own version of this include file?
Description: Digital signature