Nping calls route_dst at least twice: once with a NULL device, and again
with the device learned from the first time. This interfered with the
code that automatically chooses a loopback interface for dests that are
the same as an interface address. For example, if you are 192.168.0.1,
and you are scanning 192.168.0.1, route_dst will tell you to use
interface "lo0" even though the interface of 192.168.0.1 may be "eth0".
route_dst was returning failure because the device name check was being
done before "eth0" got changed to "lo0".
This problem didn't show up on Linux because Linux uses
route_dst_netlink, which delegates the work to the kernel. But I could
reproduce on Linux by forcing the use of route_dst_generic.
The existing tests here didn't make much sense, and didn't cover all the
situations we care about. The 8 tests now check each of
client !--send-only EOF on stdin
client --send-only EOF on stdin
server !--send-only EOF on stdin
server --send-only EOF on stdin
client !--recv-only EOF on socket
client --recv-only EOF on socket
server !--recv-only EOF on socket
server --recv-only EOF on socket
The other thing we care about is whether the client/server keeps running
after the EOF. That is now checked in each of the tests too.
RFC 3986 section 3.1:
Although schemes are case-insensitive, the canonical form is lowercase
and documents that specify schemes must do so with lowercase letters.
An implementation should accept uppercase letters as equivalent to
lowercase in scheme names (e.g., allow "HTTP" as well as "http") for the
sake of robustness but should only produce lowercase scheme names for
consistency.
RFC 3986 says that these URLs are equivalent:
http://example.com/http://example.com:/
url.parse was returning port="" for the latter. Make it instead return
port=nil like the former.
To me, this is a relative URL with a path of "". I.e., it refers to the
same document as wherever the URL appeared, like when people do
<form action="" enctype="multipart/form-data" method="POST">
This code was quite old (dating from r854 in 2000) and may not serve a
purpose anymore. Let's try removing it.
Apart from being obsolete, the code had a race condition where the
connection could be closed with a RST before getpeername was called,
leading to a fatal error message.
This used to be used to hold the original argv before quashing with -q.
Now that -q is gone, it was just a copy of argv.
fakeargv was also causing Valgrind to complain about leaked memory. it
wasn't a real leak, but now it's out of the report anyway.
