Not a perfect solution (we'd ideally want to go to another plan
immediately, but that would involve a malloc).
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Don't call through io_loop, but have it pass the connection back
and call manually.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Don't call from the plan-construction function, call after it returns.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Without this, closing an fd results in a spin...
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Overloading io_close() as a callback is ugly: create an explicit
io_close_cb().
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Use a -1 for error codes.  This makes it easier to write your own io funcs.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Rather than insisting on supplying them on every call to io_new_conn().
Also, this way it can be changed on a connection.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

And rename debug_io_plan() to io_plan_debug() so it can be exposed.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Debugging an async library is a pain: it's nice to force it into a
linear call chain to try to track problems.

Ugly code, though.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

In particular, make sure that idle connections don't get closed.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Have it set the plan itself, rather than passing it back.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Pass fd and plan explicitly, so they don't need to know the definition
of struct io_conn, and return a bool instead of an enum.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Use a NULL next pointer instead to indicate a closing connection.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Use a NULL io pointer instead to indicate an idle connection.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

It was only used for initial connections which hadn't started I/O.  Now
they are initialized with an io_plan, it can be eliminated.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Rather than going via a callback, which tends to just set up I/O, do
any setup before the call to io_new_conn(), then pass it the io_plan
directly.

The patch shows how much this simplifies our test code.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Instead of assuming they want a connection made from the new fd, hand
the fd to a callback.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

No longer needed.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

This simplifies some things: in particular, we can construct an
io_plan without needing the current io_conn.

On the other hand, we need to expose the structure now.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

We only use it in one place: for wakeup.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Ready for exposure.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Explicit callbacks are slower, but more flexible.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Sure, both listener and conn need them, but for different things
(listener uses them simply to set up conn).  Putting them in the
common union was a mistake.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Weaning off enum io_state, to allow custom ones.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

This is a better description, since it's I/O we plan to do next.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Upgrade license, since timer is LGPL.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

It's probably a bug if we're waiting for nothing.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Cleaner model for I/O, with cost of complexity if you really want bidir.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

More stress test than benchmarks, but provides a beginning.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Designed for async I/O.
Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Rusty Russell <rusty@rustcorp.com.au>

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

Remove the assembly file that checks if the CPUID instruction is
supported and have it in inline assembly as suggested by Rusty.
Signed-off-by: Ahmed Samy <f.fallen45@gmail.com>

If the list of addrinfos give to net_bind() contains only IPv4 or only
IPv6 addresses, then the ipv4 and ipv6 variables may be used uninitialized.
Signed-off-by: David Gibson <david@gibson.dropbear.id.au>