Netplay sync incorrectly checked whether the replay pointer was behind
the unread pointer twice, when in the second check it should only have
been checking if it was behind the current execution pointer. Because of
how resimulation works with device sharing, I THINK this could affect
sync. Even if it can't, it's wrong.
An unfortunate hack for cores that translate the keyboard input device
into keydown/keyup events without saving that in the savestate. We
simply replay the previous frame's input before loading the rewound
frame's state, to assure that both the state and the keyup/down state
are correct.
Ideally, cores would save this as part of the state, but it's a bit
proximal and a fairly significant change for those that use it, so it's
easier to fix in netplay.
This abstracts away the details of particular input devices for netplay,
and adds support for mice and (similar) lightguns. Unfortunately, in
practice, no core handles mice or lightguns in a savestate-safe way, so
they need to be used in stateless mode to be reliable, but they do work.
The new input handling makes adding and removing players more
complicated, since data can be present that's not expected from the
connected clients list, or absent that's expected in the list but
actually shouldn't be there.
Netplay state demotions, i.e. changes from playing to spectating or
disconnected states, could cause chain disconnections of all other
clients. This was due to a bug in when MODE change messages were sent.
Clients rely on the server sending all messages in its own order, and as
a consequence, the server typically holds messages for retransmission
until they can be retransmitted at the correct time. MODE messages were
not held, so could be sent early. When they were sent early, this caused
other clients to panic and disconnect.
A smaller but much stupider secondary bug was also fixed, in which the
first connection could be dropped due simply to writing connections[0]
instead of connections[i] somewhere.
This commit adds support for temporary desync in netplay. When frontend
features that can't be truly synced, in particular rewind, are used,
netplay is momentarily disabled. As soon as the feature finished, e.g. a
rewind ending, netplay resumes with a state load. For rewind, netplay
peers won't actually experience the effect of rewind, but they will load
the rewound state.
This commit makes spectator mode and slave mode in netplay always stay
ahead of the input, thereby avoiding rewinds, which is sort of the
point. This also changes catch-up detection to be a bit less eager, so
that they hopefully don't flap between stalling for server input and
catching up with that input.
This patch transfers core_reset across netplay. Resets effectively
worked before thanks to check_frames, but this makes resets work even
without check_frames, and in particular should allow resets to force
sync in savestateless cores, bringing them one step closer to actually
being usable by non-experts.
Previously, if two clients were connected to the same server and one of
them was ahead of the server, the only way to rectify that situation was
for the client to get so far ahead that it stalled, as the server could
only catch up with an ahead client if all clients were ahead. That's
unrealistic. This gives the server the alternate option of demanding
that a client stall. This keeps things nicely in line even with >2
players.
In the previous catch-up system, we would only try to catch up if we
were falling behind the farthest-behind peer. However, as they would
also only try to catch up to us, everyone basically agreed to the
worst-case latency. It makes more sense to try to be in parity with your
direct peer than with indirect connections.
Previously, we could be stalled by one player but still reading data
from another, which would wedge the client because we would never act
upon the newly-read data. Now we act upon data even if we're stalled.
Fixes bugs in initial connection with high latency.
The idea:
* Use a fixed number of delay_frames (eventually to be fixed at 120,
currently still uses the config variable, 0 will still be an option)
* Determine how long it takes to simulate a frame.
* Stall only if resimulating the intervening frames would be
sufficiently annoying (currently fixed at three frames worth of
time)
Because clients always try to catch up, the actual frame delay works out
automatically to be minimally zero and maximally the latency. If one
client is underpowered but the other is fine, the powerful one will
automatically take up the slack. Seems like the most reasonable system.
