toward tracking down relay stalls, log a bit more in resend process and
add UI to support an option where there's no reliance on
polling (i.e. FCM only)
I got bitten when creating a new variant leaving the permission out of
the manifest but using a config flag to claim it was there. Better to
have just one place to configure that. So check for presence in the
manifest and, for now, assert that that matches the isBanned flag. Soon
the isBanned flag can be removed.
Some people will decide to leave SMS comms in while I look for an
alternative. Maybe. Don't show them the same alert every time they open
the game. In the process removed what seemed to be a duplicate ivar.
There's no one way to launch an SMS app with a ready-to-send
invitation. So try the two that work on our two phone/app combinations,
iterating until startActivity() succeeds. The code's set up to have new
recipes added easily.
Lots of work dealing with banned permissions (SEND_SMS and
RECEIVE_SMS). First, if they're banned and NBSProxy is installed, just
use it as if the permissions had been granted. When it's not there,
explain at various points where users will otherwise be confused: when
they try to invite using data sms, or when they open a game that already
uses it.)
It's an error to use an env with the wrong thread, but it IS possible
for more than one thread to be associated with the same env. So don't
assert to the contrary.
Seeing the occasional crash and trying to plug leaks that might be
contributing. I think I fixed the Gameptr.finalize() assertions I've
been seeing forever but the thread match assertion remains. Logs will
help catch that.
Didn't think it was required, but once the app started getting messages
in the background it stoped working or crashed. Changes are minimal
because this service isn't responsible for receiving messages: it
doesn't have the long-lived threads of the BT and Relay services.
Lots of threads call the smsproto code now so the existing check was
failing. Replace it with one that ensures that only one thread at a time
is running the code. Debug-builds only.
What I've been calling SMS is now "data sms" (or "nbs", in
code). There's a new way of inviting, called "User Sms" or somesuch,
that launches the user's SMS app with a URL and phone number, much like
email (save that no addressing is required by the inviter.) This new way
won't be involved in the ban on SMS permissions. (But play by NBS is
still awesome and will stay where it can.)
Get rid of the single sender thread and the complexity of querying
individual queues for their state (and the slowdown that happened when
successful writes had to wait for those to devices that weren't
responding). Instead each PacketAccumulator does its own waiting with
timeouts and backoffs, wakes itself when appropriate, and periodically
sends if it can make a connection. Now when there are a bunch of
messages ready they'll get sent pretty quickly once connections to the
remote device start to be successful.
A device where the OS isn't servicing enqueueWork() requests won't write
back to sockets it accept()s, causing the sender's writer to block
forever. So give it 30 seconds then kill the socket.
Add support for the old protocol, and define to use it. After the next
release has been adopted and everybody can read the new protocol another
release will turn it on.
Fix race condition that would have cleared all messages including those
added after a send began. In the process, move to storing outbound
messages individually rather than concatenated in their stream format.
Set a boolean every time app's upgraded, and clear it on first launch of
MainActivity. In between, if we try to launch the BTService but the OS
doesn't schedule it, post a notification asking user to launch the app.
Lots of changes. Old BT proto is no longer compatible. New one
batches messages per device and sends all on a single connection.
The queue (now a single buffer) is now a static global and the thread
that services it doesn't die (but gloms onto whatever Service instance
is most recent.) Code to pack and unpack the protocol can probably be
reused for wifi-direct or any other pass-through transport.
I'm seeing an assertion failure in cleanup that could be explained by
the cleanup happening when initGame() has been called but no subsequent
jni method that would have added the env to the map. So let's add it
immediately so that can't happen.
JNIThread is somehow sticking around sometimes and holding the lock for
a game so that that game can never be opened again. On the theory that
there's some place retain() was called but not release(), use the thing
in try-with-resources wherever possible. Which is pretty much
everywhere.
Also added age to the lock-holder report being uploaded.
I'm seeing something permanently lock a game so it can't be opened. So
add code to report the stack of the owner to Crashlytics when opening
fails 3 times in a row. It's stubbed out for non-debug builds.
I didn't like having to re-open a game to archive it after
rematching, so now there's a checkbox that lets you archive it after
either rematching OR just using OK to dismiss the alert. Also fixed
rematch of a solo game causing new game to be created in Archive
group if that's where the source game was.
squash me
Got tired of the space the forground-service notification icons were
taking. So now BT sends and receives are done via static threads and
onHandleWork(). The send thread times itself out quickly. The receive
thread doesn't yet. We'll see how long the OS lets it run and what
needs to be done to deal with that.
It's useful when testing to have the remote device play without human
interaction. So add an option, and UI to trigger it, for the players
created remotely in response to an invitation to be robots. There are
guards in place to catch the feature slipping into a release build.
Eliminate the writer thread, since onHandleWork() is already getting
called on a thread and can process the outbound queue itself. And get
rid of reader thread needing to wait for an instance of the RelayService
to be available to process a packet (which was taking 5-10 seconds all
the time): just post received packets via an enqueueWork() call using
the app context that's always available.
GameListItem was clearing its summary when kicking off a replacement
load. Instead keep the old around in case somebody wants it (e.g. to
figure out what menus to enable) until the reload finishes.
Also add logging of how long loading takes. I think a cache might be
called for.
Darned thing was dropping packets, failing to connect games built in
response to invitations, and otherwise misbehaving. First was due to not
resheduling when exited with outbound packets in queue; second to not
overriding relayNoConnProc() (due to signature change.) Though it still
happens occasionally.... Also added timestamps to track how long it
takes a packet to be sent and ACK'd.
Not sure why I was doing this, but it's not worth the
trouble (e.g. stopping updates of other data to fix an android bug
sending too many reregistration messages.) So now just use the first
element of arrays, replacing what's there instead of prepending. Ideally
those columns would not be arrays, but that's a harder change.
It's too useful to live without. So: tap into the static list of live
Delegates, and if any of them has an Activity available use it to run
Toast on UI thread. Otherwise there will be no display.
There's likely a bug in Android now where I'm registering over and
over though the fcm id hasn't been changed. That's wrong, but it's
also wrong (and filling up the db) to register a duplicate as if it
were new. So stop that.
My VSIZE is no longer legal, and apparently there's no workaround (no
way to safely figure the length of an array whose size is known at
compile time.) To avoid the risk of duplicating little constants, added
macros that define length in a way a new VSIZE can pick it up. Couldn't
make that work for struct field arrays, however, so there I'm using
constants.
Looks like certain launches that went through DispatchNotify could
create a second "instance" of the app. Giving that Activity a launchMode
of singleTask fixes this (singleTop does not.)
Launching a game in response to a wordlist it needing being downloaded
went through a different Intent-creation path from the rest, and
resulted in a second instance of the app showing in the launcher. This
shoudl fix that. I don't *think* it's the only way I'm getting second
instances or stacked GamesLists, but it should get fixed.
I modified existing translated strings, adding the new clause. Not doing
that would cause the existing strings to be stripped because they'd no
longer have the matching set of format specifiers.
Fix a race condition introduced by making initing jnithread be
asynchronous: all the layout that was supposed to happen after listeners
were added instead happened before, so that they weren't there to be
installed as part of layout. So now, after adding them, get them hooked
up to the UI. (The complexity of this is all historical: at some point
the listeners were getting added BEFORE there were views to attach them
to, so they were cached and added later. Probably now they could simply
be installed as part of adding them. But I'm not doing that now.)
Was seeing frequent failure to open games as JNIThread.getRetained() was
unable to get a lock without waiting. Which it can't do on UI thread. So
added a method to GameLock that takes a callback to be called when the
lock's obtained, with the actual waiting done on a local thread. Then
fixed BoardDelegate a bit to not crash while waiting for the callback.
Once built for SDK 26 the old GCM support code crashes as it's calling
startService() from background. Duh! Will have to bring in Firebase's
replacement at some point. For now if I'm to release it has to be
without GCM.
more dev-only code, likely: if the game that's sending the invitation is
open when the invitation arrives, should still check if it's a dupe or
should be accepted.
Was creating a second game in some cases in response to duplicated
invitations. Add a test for forceChannel: if there's already a game with
the same gameID *and* forceChannel assume the invitation is a dupe and
drop it.
So I don't have to open a game to see e.g. how it's connecting, refactor
the network status stuff so it can be called from both board and
gameslist views. Then add new context menuitem that calls it.
I'm seeing a rare case where a game connectes to relay specifying a room
and somehow gets both slots, having provided different gameSeeds the two
times. This means an opponent won't connect, the room being full in that
game. I can't reproduce, so am logging seed changes better and switching
linux client to leave seed generation to comms as Android does.
Instead, keep them forever (for now), sorted by how long since they were
last seen. A Delete button's probably needed to prevent ex-partners from
sticking around too long. :-)
Adding a second permission a while back resulted in two Actions being
received and acted on. Needed instead to group the two and receive a
single Action IFF both were granted.
Put up an error message if too many tiles selected for trade
(a condition that couldn't exist when the pool was guaranteed to
have at least 7 in it.) (It's a hack: there's not even an enum
giving Spanish's code, and the lang_locale stuff in info.txt isn't
making it into the .xwd format.)
I've learned a bit about java since writing that class. Should be faster
now without the sleep/polling. Also making it illegal to use blocking
lock() call on UI thread. There may be some assertions to fix in the
next few days.
Was seeing race condition on slowest device where JNIThread init hadn't
finished before the event loop tried to use the gamePtr. So use
wait()/notifyAll() to fix.
And: Don't start the foreground service with its user-visible
notification, as that's too obtrusive and the ACL_CONNECTED stuff seems
to work well enough. Launching timers on a new install is mostly for my
own dev use, but won't hurt user experience either.
On O and beyond it's possible to tweak notification channels separately,
meaning users can hide the new BT-is-running notification and still get
game event notifications. So add a button that takes you to the right
Settings app page.
Separate processing of sockets from accepting them so that when an ACL
CONN notification is received and we open a socket (but don't yet have a
Service running because the ACL thing is most likely for some other app)
we can set it aside to be processed once we do have a service. Use the
same block-until-non-null thing as in RelayService to keep that thread
free of NPEs.
Rather than just dropping it and going back to a wait that will likely
last forever. I *think* wait() throwing that exception means the
thread's being killed, in which case the exception should be thrown up
to the containing loop that will then exit.
I thought I had to stop using a service instance before returning from
its onDestroy(), but that made the UI incredibly laggy AND appears not
to be necessary. At least in a bit of testing things still work.
Oreo's creating a new service instance for every single intent passed to
enqueueWork, meaning a brand new set of threads with a new queue, empty
set of messages to be retried, etc. was created every time, and all
attempts to optimize and retry were broken. So: make the threads a
static singleton that are given a Service instance from onCreate() and
told to drop it from onDestroy(). The threads proceed until they need an
instance, then block until one's available. Seems to work on Oreo and an
older Android as well.
Apparently one of the newer Android SDK levels adds the requirement to
have RECEIVE_SMS permission in order for a broadcast receiver to get
called. Meaning receipt didn't work even if SEND_SMS had been
granted. Since they're both in the same group (for now) the OS will
grant the second silently if the first has been granted, but it still
has to be requested. So request both at the same time. This still leaves
the problem that a user who's never tried to create an SMS game won't
have been asked for either permission and so won't receive SMS
invitations, but fixing that is for a later release.
Other code will take care of duplicates. This was meant to avoid a race
condition, for which 5 seconds is enough. Blocking forever complicates
testing.
Receiver is created and installed by a non-UI thread sometimes so can't
create the Handler there. onReceive() is called on an ok thread however
so create it there on demand. Fixes crash that showed when receiving
relay messages in background.
Fire up the receiver thread, and start the service, on receipt of this
ACTION (if they're not already running.) On start, the service takes
over the thread and begins dispatching messages. Works to launch the app
when it's not running and in most cases, though messages received before
the service launches are currently dropped, things seem to work.
reset timer on receiving meaningful data and on moving app to
background. If it fires and we're in background, kill the service.
(There's currently no way to restart it except bringing the app into the
foreground. Fixing that's coming.)
Moving toward a better BT invite experience: use BTService to scan for
ourselves on all paired devices, and only allow selecting from among
those on which we're running (and so likely to respond to an
invitation.)
It's a pain to have to change code to run in Genymotion, and to have
upgrade not testable using Genymotion. Consider verting this change
before next release.
Did a bunch of stuff to inherit from JobIntentService and use enqueue(),
but doesn't work yet. OS is unable to bind, with this error:
09-21 17:20:51.678 3050 3050 W JobServiceContext: Time-out while trying to bind 2edee28 #u0a277/1111 org.eehouse.android.xw4dbg/org.eehouse.android.xw4.BTService, dropping.
Notifications don't work on Oreo without this change, which includes a
new Support Library in order to get NotificationChannel and creates and
uses that as docs describe. Requires that MinSDK be raised from 8 to 14,
which may lock some users out. It *should* be possible not to do this in
the fdroid variant since their app store doesn't requires SDK 26, but
I'll look at that later.
For some reason my laptop wouldn't build without this change. No idea
what happened to the newer version I was using or if the change
works (beyond compiling). Should be easy to find the change later if
it's a problem.
Some devices, including my Moto, are apparently calling server_do() more
than most. When the game's supposed to be asking the user to pick tiles
that resulted in stacked TilePickAlerts. The stack of these
sending (taken together) too many picked tiles to the game made it
crash. So modify server to have only one pending tile-pick request going
at a time. Because the server can't know when the user dismisses the
alert in Android and so won't post again, respond to the dismissal by
closing the game. Reopening will put it in a state where the tile picker
can get called again.
Logic error meant it was never sent. Now always send on receipt of a
well-formed invitation, even if e.g. the recipient's missing a wordlist
and the game can't be started immediately.
Had inadventently changed how NetLaunchInfo was transmitted, and crashed
on receiving from older builds. Fix to not crash, and then fix to send
and recieve in the old format.
It's done for other comm types too, and is necessary to get linux test
script to pass once I make SMS delivery unreliable. But it may duplicate
the tickle that Android's doing and result in duplicate messages: test.
add new stream getters that return false if reach EOS and use them to
exit early and safely if incoming SMS msg is misformatted. I'm getting
random garbage meant for other apps perhaps.
Had an assert fire when a message ID was reused with a different count.
It was likely because of messages crossing between two variants, but
still, fix: delete what's been saved so far when a new count shows up.
Handles case where the app receives only a subset of the SMS messages
into which a larger game-level message has been broken. Now when it
restarts and the remaining parts come in the whole can be reassembled.
And use in linux client. Goal here is to reproduce then improve the
Android SMS pre- and post-processing stuff with a common/ implementation
that can be tested on linux and used wherever.
My linux sms hack used inotify and didn't check for messages that were
there when the app launched. Replace inotify with a simple glib periodic
timer. A bit of latency mimics SMS better anyway. Update test script to
support SMS, and add params to and otherwise fix linux client so
everything works.
So now all jni code uses a single dutil context, but also a single
mempool and jniutil instance instead of new instances of the latter two
per game and dict-iteration.
Trying to separate what's game-specific from what can be app/device
specific (i.e. with a long lifespan, and available when a game isn't
open.)
Android will be broken after this commit and fixed after the next
A number of jni calls were "stateless", which meant they allocated their
own vtmgr and mpool instances each time invoked. Instead invoke them
with the global jni closure and add to it vtmgr (already has mpool) and
use these instead of allocating/freeing each time. To make sure no race
conditions are introduced (mpool, though debug-only, is probably not
thread-safe), guard these new uses with an in-use flag. If that fires
I'll need a mutex or something.
Once the pool count drops to 0, start showing the number of tiles left
in the user's tray. This prevents there being a long time when nothing
seems to be changing *and* the script from exiting early because it
thinks all games are hung.
Somehow I got a wordlist into a location different from what was
recorded in the DB table and since the delete command matched on
location as well as name it was never deleted (which meant the checksum
was never updated and so upgrading never seemed to succeed.) Removing
the match on location fixed that problem, and since I don't see any harm
in cacheing only one version of a wordlist will simply leave it that
way.
Did a bunch of cleanup as well.
I'm seeing several IllegalStateException crashes due to e.g. having an
alert posted when app's in the background. Need to fix them, but the
debug build crashing isn't helpful. Log.e() instead.
The pesky thing is back. When app's in the background with an
unconnected game open displaying the "resend/wait" alert and the game
connects get IllegalStateException because the fragment stack's being
modified after onSaveInstanceState() (or, because the dialog fragment,
saved as an instance variable in BoardDelegate, dates from an earlier
state. Anyway, catching and dropping the exception and elsewhere failing
silently to rebuild the alert seems to fix the problem, but the right
fix is likely different. I suspect hanging onto that iVar is wrong, and
that the dialog should go away when onStop() is called and then be
rebuilt later from saved state. But for now, not crashing is good.
Got a report of crashes due to corrupt move records. Given I rarely see
them I wondered if it's because the hint- and robot-generated moves I
work with have tiles in order. So now on debug builds tiles in moves
from those sources are randomly rearranged (as if the user had formed
the word in random order.) The bug isn't showing up, but I figure the
test's worth keeping.
Stuff I'm doing with invitation resends is making the relay loop
inifintely. Let's assert a small loop count instead: better to crash and
restart than loop forever unable to process requests.
Remove a bunch of duplicated code, replacing with implementation in
XWService. Fixes duplicate invitation and game opening policies being
slightly different.
The fix I made earlier for this relied on a callback that was skipped in
release builds. Now always take the path that involves making the
callback when one is provided. Also remove an optimization that was
trying to eliminate possible moves based on scores prior to doing the
more expensive full check. In 2018 I prefer simplicity, and can make the
remaining code faster if that's required.
Try some funky layout shite to get, within a horizontal linear layout,
the first text field trucated if necessary so that the second (holding
the score) can be fully displayed. Tested on exactly one emulator so
far.
Try some funky layout shite to get, within a horizontal linear layout,
the first text field trucated if necessary so that the second (holding
the score) can be fully displayed. Tested on exactly one emulator so
far.
I think there's a bug in Weblate because I've seen this before: where
the English provides only an <other> the translation comes back with
only a <one>. That's wrong. Try adding a <one> in the English case to
see if that makes a difference.
I think there's a bug in Weblate because I've seen this before: where
the English provides only an <other> the translation comes back with
only a <one>. That's wrong. Try adding a <one> in the English case to
see if that makes a difference.
Otherwise it takes too long to scroll if you have hundreds of
games. To make this work had to move scroller to left side of games
list display as otherwise the scroller steals events from the expander
thingies.
Likely because of something in the jni world unset per-player dict name
is empty string rather than null, so test for that too. Fixes dicts
popup in newly-created game have an empty first line " (in use)".
I like this better than the previous fix: rather than share a
thread->env map with the game world allocate a new one for each
iterator. This could cause problem if the iterator is used on threads
that don't currently call map_thread(), or if there are callbacks that
need to look up the env that I'm not aware of. Needs testing...
Stumbled on a NPE opening up the wordlist browser configuring the first
game on a new install. So now test for null there and init early if
necessary. Seems to work, and won't do anything in places were not
needed.
Remake the min and max spinners every time either value changes so they
can't be used to set nonsensical values. (Which leads to immediate
crashes.) I'm sure this wasn't always a problem, but...
As reported to google, dict iterator destruction was crashing due to a
race condition if it happened after a game using the same dict had been
closed since it needed a mapped env that the game closure would
remove. Fixed in two ways, one by adding the mapping prior to the code
that uses it (a common pattern: add happens many times, whenver it might
be needed, but remove only once), and second by passing env into the
code that was crashing.
The mapping stuff remains inherently racy and I'm not sure now how to
fix that. It depends on there being a place to unmap after which it's
guaranteed the mapped value won't be needed again. When two
objects (game and dict_iter in this case) map the same env/thread combo
there's a race.
Looks like the assertion was left in when adding support for dual-pane
mode, as all other onPosButton() implementations called super rather
than assert. Which this one does now too.
Getting ANRs because (I think) the main thread's waiting for the write
thread to die and now the write thread's doing a ton of work
sometimes. So move the threads into a standalone object that can be
allowed to die on its own time without anybody waiting.
I *think* the reason I'm occasionally seeing toasts about not finding a
move is that when the engine's interrupted by there being a UI event in
the queue that error is posted. Instead try posting only when at the end
of the search nothing's been found.
Having reconfigured to use non-existent relay port as a test of falling
back to the web apis, tweak stuff: send the packets that have been
accumulated when an EOQ is found (rather than dropping all of them
immediately) before exiting the write thread; and start the threads up
when posting a packet in case they aren't (they may not be when the post
happens via timer firing.)
Seemed to be causing ANRs. Integrate instead into outgoing message queue
by using poll(timeout) then checking for unack'd packets every time
through the loop (but not more than once/3 seconds or so.)
Presence of timestamp instead of a boolean determines whether packet
should next via Web. Timestamps might also allow to process a larger
number of unacked packets in a single timer fire....
Track ack'd and unack'd packets. When there are ten more of the latter,
skip the UDP-send step. This is probably not the algorithm I'll settle
on (an explicit PING to the relay over UDP might be simpler), but it's
simple and easy.
Send each packet via UDP if that's thought to be working (always is,
now) and start a 10-second timer. If it hasn't been ack'd by then,
resend via Web API. Tested by configuring to use a UDP socket that the
relay isn't listening on. Only problem is that the backoff timers are
broken: never stops sending every few seconds.
The plan's to use the native relay protocol first, then to fall back to
the slower but more reliable (esp. on paranoid/block-everything wifi
networks) webAPI. This is the name change without behavior
change (except that the native kill() to report deleted games is gone.)
When grouping to allow multiple packets per outbound API call I forgot
that some are there to mark the end-of-queue: can't be sent! Trying
caused a NPE. Now if any EOQ is found in the queue that batch is dropped
and the thread's exited.
AddrInfo now has ref()/unref() and keeps a global socket->refcount map
(since actual AddrInfo instances come and go.) When the
count drops to 0, the existing CloseSocket() method is called. This
seems to fix a bunch of race conditions that had a socket being closed
and reused while old code was still expecting to write to the device
attached to the socket the first time (along with lots of calls to close()
already-closed sockets, attempts to write() to closed sockets, etc.)
Use of mutex logging recurses infinitely if config uses mlock, so remove
that. And don't free sockets after handling their messages as they may
be in use elsewhere. This likely introduces a leak of sockets.
Haven't seen it happen, but I think there was a bug that could have led
to all the sockets coming back as ready from poll() being dropped. Fixed
that and added/cleaned up some logging.
translate the most-used features of my too-big-for-bash script into python3,
which is clearly much better suited. Tried to keep the structure and variable
names intact so that diff has a chance of showing something, but when a class
replaces a bunch of arrays that were being kept in sync there's only so much
you can to. Currently doesn't support stuff like app upgrades and switching
from tcp to udp, but those should be relatively easy to bring over from the
.sh when/if I need them.
translate the most-used features of my too-big-for-bash script into python3,
which is clearly much better suited. Tried to keep the structure and variable
names intact so that diff has a chance of showing something, but when a class
replaces a bunch of arrays that were being kept in sync there's only so much
you can to. Currently doesn't support stuff like app upgrades and switching
from tcp to udp, but those should be relatively easy to bring over from the
.sh when/if I need them.
I'm not quite sure why it was firing, but the pattern of a FindGame
failing due to a race condition and requiring a retry exists elsewhere
in the code. Lots of tests pass once this change is made. :-)
The hack I came up with for storing them in memory isn't working. Even
if it's just that I don't understand C++ maps they need to be cleared
when the DB's wiped (my favorite test these days) and I don't want to
spend the time now.
It breaks rematch that "dict" is being passed to the Android client from
the linux side, and this is easier than figuring out how and when to
dereference the link.
Making the right_side elem match its parent height prevents the
lower-right region of game list items from falling through and
triggering a toggle-selection event.
Making the right_side elem match its parent height prevents the
lower-right region of game list items from falling through and
triggering a toggle-selection event.
As expected, moves are no longer received instantly because the UDP
socket isn't available for the relay to write too once the URL
handler (relay.py) finishes.
Ideally the comms module wouldn't go through its connecting routine in
order to join a game. To that end I added a join() method to relay.py
and code to call it. Joins happen (pairing games, starting new ones,
etc.), but after that communication doesn't. First part of fixing that
would be to make cookieID persistent and transmit it back with the rest
of what join sends (since it's used by all the messages currently sent
in a connected state), but I suspect there's more to be done, and even
that requires a fair number of changes on the relay side. So all that's
wrapped in #ifdef RELAY_VIA_HTTP (and turned off.)
Was coming in as a string when called via curl. This may be a problem
for other ints if I go with lots of params instead of a json, which is
looking less likely.
join's how devices will create or join existing games. It more
compilicated than I'd like but seems to work except that once a slot's
assigned it's unavailable to anybody else even if the other fails ever
to respond (i.e. needs the ACK function of the c++ relay.)
