2 =======================
4 Important stuff (to be done before the next release)
9 - configure should not be in the CVS. Most open source projects have an
10 autogen.sh script to run autoconf etc. after checkout. I think we
11 should stick to the standard.
13 - The source files should be better documented. There is no official
14 design guideline for what is done where. There should be a common coding
15 style (okay, you can guess that by looking at the code) and a guide for
16 what needs to be documented.
18 Linux kernel interface
20 - Currently there is a race condition in that a socket is activated at the
21 same time it is opened - before applying a filter. This has to
22 be corrected so that capture starts when pcap_read is called for the
28 - Better documentation and cleanup of the interface. I am seeing a few
29 problems at the first glance which needs fixing:
30 + pcap_lookupnet makes little to no sense with protocols != IPv4
31 + not very well suited for interactive programs (think ethereal). There
32 should be a way for the application to get a file descriptor which it
33 has to monitor and a callback in pcap which has to be called on
34 activity (XXX - "pcap_fileno()" handles the first part, although
35 "select()" and "poll()" don't work on BPF devices on most BSDs, and
36 you can call "pcap_dispatch()" as the dispatch routine after putting
37 the descriptor into non-blocking mode)
38 + too many functions. There are a lot of functions for everything which
39 violates the KISS principle. Why do we need pcap_strerror, pcap_perror
41 + the manpage has a brief description of each function but where is the
42 big picture? Seems like you need to buy UNP for that...