Creating an address record without having heard from the remote is
new, and setting it to an old version was preventing using new msg
format. So set to new by default. Not sure how it'll get downgraded
facing an old client, but a 100-game upgrade test passes.....
Looks like invitations become unsendable before they're deleted
sometimes and so block sending real messages. This fixes device
accepting invitation but never hearing back from host.
Duplicate messages early on, which happened only in the test script
but could have anywhere, broke connectivity. So don't kill address
records when a duplicate shows up. Dupes only escape message ID
checking early (before channel is established). I used to remove
address records when a message was rejected, but don't understand why
so removed that, though asserts show it's not mattering except for
those early messages.
There was some confusion around host and self addresses, where they're
created, default values, removing conTypes from defaults that are not
in received host addr, etc. I left in some asserts to help understand
if code that seems wrong but hard to fix is still getting called.
Current networking, based on invitations rather than a relay that
plays matchmaker, allows host to know its address when a game is
created, and for guest to know its host's address in
addition. Enforcing this makes inviting and rematching in common
code (coming soon) easier. Big change on Android is I used to create a
new game prior to passing it to GameConfigDelegate, but now I have to
wait for user to configure (including choosing how to communicate)
before I can create it.
Remove legacy relay-inspired logic around comms addressing. Now when a
device creates a game it's required to provide its "self address," and
if it's a client, the address of the host (which it presumably got
through the invitation in response to which the game is being
created.) Then as registration messages come in from clients, the host
gathers their addresses as always.
I want receiver to know when message was originally created. This adds
timestamp to messages and passes it via send proc. Client needs to
send it where possible. So far, MQTT format can't include it without
change, so I'm adding a new proto version. This change can read the
new version. Once that's well-enough distributed I can start sending
using it. Other transmission types than MQTT are for later.
Attempting to stop calling the relay, but to let relay-only games finish
and issue invitations. Big changes are that relay is removed from games
if they have viable mqtt connections, and relay timers fire less often,
then eventually stop getting set if there are no active games. Result is
that a relay-only invitation won't likely be received, but there should
be few or none of those now.
augmentAddrIntrnl() sets the has-mqtt bit in comms->addr but not the
address data (has none). If that address had been loaded from stream
the address bits will be random, not 0, and so get treated as an
address.
Sending in 32 bits what can probably fit in 8 -- twice -- is dumb. As is
sending the gameID (as connID) when it's already known. In 2016 I added
a fixed marker, so I can count on all current code sending that. If I
fail to find that, I know the sender is new (current release or better.)
In that case, use 12 bits instead of 32 for two fields, drop the marker
and the connID. Result is that an ACK's size, including MQTT headers,
drops from 28 bytes to 21.
It's simpler this way, and I'm tired of stuff not happening because the
OS chooses not to schedule e.g. an invitation send for minutes. Goal's
to be running BluetoothServerSocket.accept() as much as possible when
there are active BT games in play OR when the game's in the foreground.
If that's happening, sent invitations and moves will be received when
users expect. When there's no traffic and app isn't being brought to
foreground, backoff will ensure I don't try to run accept() too often.
FWIW, BTLE seems to offer a better way to do this (to have an app be
responsive to incoming invitations when it hasn't run in the foreground
in a while), but it requires users to accept FINE_LOCATION
permission. I'm hoping I can make this work to avoid asking for that
permission.
Not at all tested, but now the game's timestamp is kept and passed in to
where it can be used to determine, e.g., which of two Bluetooth device
names to keep for a given opponent.
Got as far as having gtk client display list of previously harvested
known players to be invited. Their addresses, or at least mqtt ids are
saved. Next is to actually invite one.
I'm fixing android client not showing stats for or allowing to disable
mqtt after it's added automatically to a game that connects
otherwise. Problem was that only the channel got the mqtt address
flag. So now add the flag for any type that's added.
If a configured-as-host game joined an existing game the relay would
make it a guest. The android util_ callback for that change was only
implemented in BoardDelegate and so the change was dropped unless the
game was open/visible. Because comms recorded the change, though, the
callback would never be called again and so the game never learned to
behave as a guest and never registered: permanent failure to join game!
Implemented with a new server state so initClientConnection can be
called from server_do() instead of inside comms while processing an
incoming packet.
To avoid inviteless relay games not having an mqtt channel (and so
yielding a crappy experience) pass the mqtt devid when device registers
and when server replies with all-here.
This is meant to replace the relay eventually, but for now it's a new
option, like BT or SMS, to be chosen. Protocol is handled in common/
code for the first time, meaning that linux and android interact without
the need to keep two platforms in sync. Linux uses lib-mosquitto, and
Android uses eclipse's Paho client (the generic java version, not the
one that uses four-year-old Service patterns and so crashes for SDK >=
26.)
I was getting an occasional crash using a stale env to delete a dict's
resources because the dict was cacheing the env that created it. Dumb!
Using the thread->env mapping stuff worked, but that felt risky and so I
tried just passing it in. It's safe, and involves an amount of change I
can tolerate. So likely going that way.
With these changes the test script no longer produces games that don't
finish. I think they are recovering from the problems produced by one
device doing an undo while another is making a move, or other problems
produced by undo being allowed to happen on any device at any time, but
haven't analyzed the games beyond the test script's verifying that they
all finish with a Winner after all tiles are consumed.
There are still some stalls turning up when running the test script,
especially with more than two players in a game, but but it'll be
easier to do the final debugging off my pathetic travel-only
chromebook. And things are already much better. :-)
Lots of changes adding a games-list view to the app from which you
create new games, open and delete existing ones, etc. There's still
plenty that's unimplemented, but it's already more useful for testing
and development. Which is the point.
Use low-level NFC, a combination of emulated card and reader mode, to
work around Google's removal of "beaming" support from Android 10. App
emulates a card by declaring support in its AndroidManifest. When a game
is open that has data to send, it goes periodically into read mode. If
two devices are touched while one is in read mode and the other isn't,
they handshake and open a connection that should last until they're
separated. The devices loop, sending messages back and forth with or
without data (as available.)
