(Re)move locale bpf definitions in Packet32.h

[sgeto]

Some of the bpf* definitions stolen from libpcap and not really guarded by BPF_MAJOR_VERSION in Packet32.h are also available in pcap/bpf.

They should be (re)moved or better guarded. So for example #ifndef BPF_MAJOR_VERSION should say #if !defined (BPF_MAJOR_VERSION) && defined (PCAP_DONT_INCLUDE_PCAP_BPF_H)

[sgeto]

They should be (re)moved or better guarded. So for example #ifndef BPF_MAJOR_VERSION should say #if !defined (BPF_MAJOR_VERSION) && defined (PCAP_DONT_INCLUDE_PCAP_BPF_H)

Update: Didn't work as expected: https://ci.appveyor.com/project/sgeto/npcap/build/0.10.1 😠

[Dysl3xik]

bpf.h actually has a note about this:

* We do not check for BPF_MAJOR_VERSION, as that's defined by
 * <linux/filter.h>, which is directly or indirectly included in some
 * programs that also include pcap.h, and <linux/filter.h> doesn't
 * define stuff we need.
 *
 * This also provides our own multiple-include protection.
 */
#if !defined(_NET_BPF_H_) && !defined(_NET_BPF_H_INCLUDED) && !defined(_BPF_H_) && !defined(_H_BPF) && !defined(lib_pcap_bpf_h)
#define lib_pcap_bpf_h

I modified Packet32.h to #ifndef lib_pcap_bpf_h and that fixed it well enough for now.

[guyharris]

The stuff at the beginning of libpcap's pcap/bpf.h, in the master branch, is currently

/*
 * This is libpcap's cut-down version of bpf.h; it includes only
 * the stuff needed for the code generator and the userland BPF
 * interpreter, and the libpcap APIs for setting filters, etc..
 *
 * "pcap-bpf.c" will include the native OS version, as it deals with
 * the OS's BPF implementation.
 *
 * At least two programs found by Google Code Search explicitly includes
 * <pcap/bpf.h> (even though <pcap.h>/<pcap/pcap.h> includes it for you),
 * so moving that stuff to <pcap/pcap.h> would break the build for some
 * programs.
 */

/*
 * If we've already included <net/bpf.h>, don't re-define this stuff.
 * We assume BSD-style multiple-include protection in <net/bpf.h>,
 * which is true of all but the oldest versions of FreeBSD and NetBSD,
 * or Tru64 UNIX-style multiple-include protection (or, at least,
 * Tru64 UNIX 5.x-style; I don't have earlier versions available to check),
 * or AIX-style multiple-include protection (or, at least, AIX 5.x-style;
 * I don't have earlier versions available to check), or QNX-style
 * multiple-include protection (as per GitHub pull request #394).
 *
 * We trust that they will define structures and macros and types in
 * a fashion that's source-compatible and binary-compatible with our
 * definitions.
 *
 * We do not check for BPF_MAJOR_VERSION, as that's defined by
 * <linux/filter.h>, which is directly or indirectly included in some
 * programs that also include pcap.h, and <linux/filter.h> doesn't
 * define stuff we need.  We *do* protect against <linux/filter.h>
 * defining various macros for BPF code itself; <linux/filter.h> says
 *
 *  Try and keep these values and structures similar to BSD, especially
 *  the BPF code definitions which need to match so you can share filters
 *
 * so we trust that it will define them in a fashion that's source-compatible
 * and binary-compatible with our definitions.
 *
 * This also provides our own multiple-include protection.
 */
#if !defined(_NET_BPF_H_) && !defined(_NET_BPF_H_INCLUDED) && !defined(_BPF_H_) && !defined(_H_BPF) && !defined(lib_pcap_bpf_h)
#define lib_pcap_bpf_h

The bit about <linux/filter.h> is

#ifndef __LINUX_FILTER_H__

and

#endif

around a bunch of the definitions of the macros for BPF instructions.

Perhaps libpcap's pcap/bpf.h should protect against Packet32.h being included first.