GitLab

Tried using this function in a new project and didn't work. My guess: it wasn't tested. <looks at Deuce>

Tweak the poll() implementation to match.

Should fix #248

Specifically, have errors mean the fd is "ready".

Also, fix some extra revent clearing and a spy socket issue while
we're looking at this code.

Should fix issues with CGI scripts that close stdin/stdout but do
not terminate, and maybe deal with some other corner cases at the
same time.

This appears to work and the event handler *should* work on other
event types already.

Note, this is *nix-only due to the use of poll().  select() will
need to be used for Windows to keep XP compatability.

Rather than depending on _WIN32 which is silly, make it clear what
exactly is being done and why.

Note that the spy socket stuff and some external stuff on *nix have
had the select() implementation removed.  To support using select()
on !Win32 will require de-refactoring that's not really worth doing.
Instead, just use #error and reference the commit where the old
code can be found.

Not all implementations set POLLHUP when a TCP socket is half-closed.

Was returning disconnected when it wasn't at times.

This won't impact Synchronet as it has a separate signal handling
thread, but we still need to behave properly for processes that
don't.  I'm also saying that ENOMEM does not indicate a disconnection,
though it may be better to pretend it was disconnected...

socket_readable() returns TRUE if recv() will not block if called
on the specified socket.

socket_writable() returns TRUE if send() will not block if called
on the specified socket.

socket_recvdone() returns TRUE if all data has been recv()ed an
the socket is closed.

"will not block" includes "will return an error".

This matches the three main use-cases for select() of a single socket
in Synchronet.

Use them in the SyncTERM RLogin stuff so they can be tested easily.

A couple of stragglers here, the isalnum() call caught by an MSVC exception.

xp_inet_pton() now sets the socket errno value, doing as instructed by the TODO.

Rename the portable inet_pton() to xp_inet_pton() and define a wrapper macro (for all Widows builds) to use it instead of inet_pton() as not all supported Windows systems (e.g. WinXP) have an inet_pton() implementation.

This allows the HAProxy mod to be re-enabled by default without preventing executing on WinXP.

Sorry guys. :(

This one should build even on Win32.

Needed by MinGW32... we'll see if MSVC needs it as well soon.

Socket error descriptions on Windows can be quite long (e.g. > 128 chars) and
include punctuation, so include at the end of logged socket error messages.

If FormatMessage() fails on Windows (in socket_strerror()) return the
GetLastError() value in the error description.

Use Microsoft magic to "ignore regular line breaks in the message definition text." of socket error strings on Windows:
strips the embedded line-feeds from the socket error description.

that support large file offsets (e.g. 64-bit Linux).
Considered returning ssize_t (like sendfile does), but opted for off_t. Could
be convinced otherwise.

descriptions in bind failure error messages in a cross-platform
(i.e. WinSock-compatible) manner.

address (family and address value, ignoring ports and other fields).

failed login attempt client IP addresses on Windows.

(only impacts active FTP transfers via IPv6).

Specifically, if select(),  getsockopt(), or connect() fails, return
ERROR_VALUE instead of SOCKET_ERROR.

Found by Coverity since we never checked the return value of getsockopt()

Win32 make connect() failure an exceptional condition.