Hover highlight was visible even if the mouse pointer was in a different
window, and the window refreshed itself every frame if the mouse pointer
was not over its matrix widget.
Resolved by using OnMouseOver() instead of OnMouseLoop(), and only
redrawing if the hover position has changed.
docs/admin_network.md promised that information in an admin packet
is never removed. It does allow the possibility of using a new packet type
for changed data in combination with a bump of the admin port version.
As the recent command handling changes modified the contents of
ADMIN_PACKET_SERVER_CMD_LOGGING, do exactly that.
If the arguments of the callback proc don't match with the command parameters,
we can't do the proper command execution anyway. As such, don't even generate
an unpack function in the first place, saving a bit of unnecessary code bloat.
Validate on receive that the cmd/callback combination is supported, rejecting
clients that try to send invalid values.
The data will be transmitted as the length followed by the serialized data. This allows the command
data to be different for every command type in the future.
A race condition happens when an IPv6 connection takes more than
250ms to report an error, but does return before the IPv4 connection
is established.
In result, an invalid socket might be used for that connection.
This gives user visual feedback that the refresh is still pending, and
prevents people from clicking again and again thinking nothing is
happening. This is especially true for connections that fall back to
TURN, as that takes a few seconds to kick in.
Additionally, prevent clicking on the button again while a refresh
is pending. This is only delaying a successful result.
The function clears all stun-handlers. This causes all of those
objects to be destroyed.
A handler can have a pending connecter, which was only killed in
case CloseConnection() was called. This is never the case when
the object is destroyed. In result, the connecter could finish
and cause a use-after-free by calling into the (now deleted)
handler.
When you are query several servers at once, it is rather unclear
for which server you got a popup. Instead, show any errors on the
server itself.
This is only true for the query-part. Joining a server still gives
an error popup to tell you about any issue.
You can now still query a full server, as long as the maximum
amount of allowed connections isn't reached. This means that as
long as there are not 255 clients connected to a server, you can
always connect to query.
Old servers don't tell the GameScript they are running, so nothing
should be shown.
All values in NetworkGameInfo initialize as 0/empty, except for GS
version. Someone has to be different from the rest, I guess.
Nobody really paid attention to the lobby window, and it completely
missed its purpose. Most people don't even wait for companies to
show up, but just hit "New Company".
This in turn means people create a lot of unneeded companies, while
they "just want to watch the game" or join another company.
Instead, "Join Game" now just joins the game as spectators.
Soon we will make "join game" join the game as spectator first,
so limiting the amount of spectators makes no sense anymore in
that context. Not sure it ever did make sense.
"my_client" wasn't always free'd when a game ended. "my_client"
keeps a reference inside the PT_NCLIENT pool. The rest of the
code assumes that when you are not in a game, it can freely
reset this pool.
In result: several ways to trigger a use-after-free.
TURN is a last resort, used only if all other methods failed.
TURN is a relay approach to connect client and server together, where
openttd.org (by default) is the middleman.
It is very unlikely either the client or server cannot connect to
the STUN server, as they are both already connected to the Game
Coordinator. But in the odd case it does fail, estabilishing the
connection fails without any further possibility to recover.
Before 8a2da49 the NewGRF names were synchronized using UDP packets, however
those have been removed. With this a new version of the GameInfo packet is
introduced that allows to specify the type of serialisation happens for
NewGRFs. Either only the GRF ID and checksum, or those two plus the name of
the NewGRF.
On this request for local servers will send the NewGRFs names.
The Game Coordinator will get the names on the first registration, and after
that only the GRF ID and checksum.
These were filled with "<Unknown>" (before 8a2da49) and later their name would get filled via UDP requests to the server. These UDP packets do not exist anymore, so they will always remain "<Unknown>".
Remove that logic and just use the generic translated error GRF UNKNOWN string instead.
This method doesn't require port-forwarding to be used, and works for
most common NAT routers in home setups. But, for sure it doesn't work
for all setups, and not everyone will be able to use this.
Every outgoing connection, either TCP or UDP, triggered
NetworkInitialize(), which triggered NetworkUDPInitialize() which
first closes all connections.
Now the problem was that "Search LAN games" found a server, added
it to the list, after which (over TCP) it queries the server. This
closes all UDP sockets (as that makes sense, I guess?), while the
UDP was still reading from it.
Solve this by simply stop initializing UDP every time we make an
outgoing TCP connection; instead only do it on start-up.
In this mode you do register to the Game Coordinator, but your
server will not show up in the public server listing. You can give
your friends the invite code of the server with which they can
join.
This removes the need to know a server IP to join it. Invite codes
are small (~7 characters) indentifiers for servers, which can be
exchanged with other players to join the servers.
Normally TCPConnecter will do a DNS resolving of the connection_string
and connect to it. But for SERVER_ADDRESS_INVITE_CODE this is different:
the Game Coordinator does the "resolving".
This means we need to allow TCPConnecter to not setup a connection
and allow it to be told when a connection has been setup by an external
(to TCPConnecter) part of the code. We do this by telling the (active)
socket for the connection.
This means the rest of the code doesn't need to know the TCPConnecter
is not doing a simple resolve+connect. The rest of the code only
cares the connection is established; not how it was established.
OTTD_COORDINATOR_CS for the game coordinator defaults to coordinator.openttd.org:3976
OTTD_CONTENT_SERVER_CS for the content server defaults to content.openttd.org:3978
OTTD_CONTENT_MIRROR_CS for the content mirror server defaults to binaries.openttd.org:80
Rename the zero-parameter NetworkValidateClientName to NetworkValidateOurClientName to make it clearer it is performed on our client name, and to make it a non-overloaded function to aid with the variant being added a few commits later
* Codechange: [Network] split CloseSocket and CloseConnection more clearly
- CloseSocket now closes the actual OS socket.
- CloseConnection frees up the resources to just before CloseSocket.
- dtors call CloseSocket / CloseConnection where needed.
In the destructors of many of the network related classes Close() is called, just like the
top class in that hierarchy. However, due to virtual functions getting resolved statically
in the destructor it would always call the empty Close() of the top class.
Document the other cases where a virtual call is resolved statically.
It now follows very simple rules:
0 - Fatal, user should know about this
1 - Error, but we are recovering
2 - Warning, wrong but okay if you don't know
3 - Info, information you might care about
4 -
5 - Debug #1 - High level debug messages
6 - Debug #2 - Low level debug messages
7 - Trace information
The code mixed up "client has quit but we already told everyone"
with "client lost connection, handle this".
Split up those two signals:
- CLIENT_QUIT means we told everyone and the connection is now dead
- CONNECTION_LIST means we should tell everyone we lost a client
This meant that on opening the Multiplayer window, if you had more
than one server configured, it would one by one cancel all pending
queries and send a new. Result: only the last server was updated.
The most common case never needs access to it anymore. Make the
one exception to this explicit. This means the fact that we
store it is now an implementation detail.
Hostnames like "content.openttd.org" resolve into multiple IPv4 and IPv6.
It is possible that either of the IPs is not working, either due to
a poorly configured OS (having IPv6 but no valid route), broken network
paths, or a service that is temporary unavailable.
Instead of trying the IPs one by one, waiting for a 3s timeout between
each, be a bit more like browsers, and stack attempts on top of each
other with slight delays. This is called Happy Eyebells.
Initially, try the first IPv6 address. If within 250ms there is no
connection yet, try the first IPv4 address. 250ms later, try the
second IPv6 address, etc, till all addresses are tried.
If any connection is created, abort all the other (pending) connections
and use the one that is created. If all fail 3s after the last connect(),
trigger a timeout for all.
We now resolve the connection_string to a NetworkAddress in a much
later state. This means there are fewer places constructing a NetworkAddress.
The main benefit of this is in later PRs that introduce different types
of NetworkAddresses. Storing this in things like NetworkGameList is
rather complex, especially as NetworkAddress has to be mutable at all
times.
Additionally, the NetworkAddress is a complex object to store simple
information: how to connect to this server.
Split the updating in a "static" version that only needs to be called when a new map is loaded or some settings are changed, and a "dynamic" version that updates everything that changes regularly such as the current game date or the number of spectators.
Under normal circumstances the server's ID is 32 characters excluding '\0', however this can be changed at the server. This ID is sent to the server for company name hashing. The client reads it into a statically allocated buffer of 33 bytes, but fills only the bytes it received from the server. However, the hash assumes all 33 bytes are set, thus potentially reading uninitialized data, or a part of the server ID of a previous game in the hashing routine.
It is still reading from memory assigned to the server ID, so nothing bad happens, except that company passwords might not work correctly.
One could join a network game from within an already running network game. This would call a NetworkDisconnect, but keeps the UI alive. If, during that process the join is aborted, e.g. by cancelling on a password dialog, you would still be in your network game but also get shown the server list.
Solve all the underlying problems by falling back to the main UI when (re)connecting to a(nother) server.
This means that pressing Refresh button and adding servers manually
now uses TCP.
The master-server and initial scan are still UDP as they will be
replaced by Game Coordinator; no need to change this now.
If we query a server that is too old, show a proper warning to the
user informing him the server is too old.
The idea is that if you query an older server that does not support
this packet yet, the client receives an error. The assumption was
that on every "illegal packet" the connection would be closed. This
turns out to be false.
Now CLIENT_GAME_INFO aligns with the old PACKET_CLIENT_NEWGRFS_CHECKED,
which does a pre-check (which fails), and an error is sent back
and the connection is closed.
This is not a nice solution, but it is the best we got.
The lobby of a server requested some parts via UDP and some via
TCP. This is strictly seen fine, but for future extensions it
is a lot easier if just one protocol is used.
Currently we use default OS timeout for TCP connections, which
is around 30s. 99% of the users will never notice this, but there
are a few cases where this is an issue:
- If you have a broken IPv6 connection, using Content Service is
first tried over IPv6. Only after 30s it times out and tries
IPv4. Nobody is waiting for that 30s.
- Upcoming STUN support has several methods of establishing a
connection between client and server. This requires feedback
from connect() to know if any method worked (they have to be
tried one by one). With 30s, this would take a very long time.
What is good to mention, is that there is no good value here. Any
value will have edge-cases where the experience is suboptimal. But
with 3s we support most of the stable connections, and if it fails,
the user can just retry. On the other side of the spectrum, with 30s,
it means the user has no possibility to use the service. So worst case
we annoy a few users with them having the retry vs annoying a few
users which have no means of resolving the situation.
Especially if there are many players online, trying to chat with
the right one can be a visual challenge. This can be solved by
highlighting the row you are on. This visual cue is often enough
for humans to find the right row.
The GUI now more clearly shows some basic information about the
server you joined, your client name (and the ability to change it),
and what players are in which company.
It also contains useful buttons to press to join companies, chat
with other people, and for admins to kick/ban people.
Additionally, renamed "advertised" to "visibility"; this has to
do with future additions, but also because it is more clear in
wording.
This so names from other clients are known valid in the client as well, instead allowing some compromised/bad server to potentially crash clients upon certain expectations.
Strictly seen the comment is true, as it says 'e.g.', but it is
misleading. The server name is just that: the name of the server
as configured. No need to mention advertising.
When ever you saw this debug lines (which you never should), they
showed an empty address. It is also not very useful to have, as it
always points to a known server anyway.
The original idea was that people could find a server they could
talk in their native language on. This isn't really used in that
way. There are several reasons for removing this:
- the client also sends his "language" to the server, but nothing
is doing anything with this.
- flags are a bad way to represent languages, and over the years
we had several (rightfully) complaints about this.
- most servers have their language set to "All", and prefix the
servername with the language it is about. This is a much more
efficient way to do the same.
All in all, this feature should go back to the drawing board.
Maybe it could work in another form, but this form is not it.
The idea back in the days was nice, but it never resulted in
anything useful. Most servers either read "(loaded game)" or
"Random Map", neither being useful. It was meant for heightmaps,
so you could find a server that was using a specific one .. but
there are many things wrong with that idea. Mostly, servers tend
to save and load savegames from time to time, after which the
original heightmap used was lost.
All in all, removing map_name all together is just better.
The server information panel was scaled by GUI scale, which could result in a panel that is longer than the server list. This height difference is then maintained when the window is resized to fill the screen.
Instead, specify the minimum size by number of text lines and (summed total) padding.
Use status >= STATUS_AUTHORIZED as the state criteria for all cases
where updates about other clients are sent.
This avoids the case where a client is informed that another client
has joined but not informed when it later quits, resulting in
stale entries in the client list window.
Basically, modal windows had their own thread-locking for what
drawing was possible. This is a bit nonsense now we have a
game-thread. And it makes much more sense to do things like
NewGRFScan and GenerateWorld in the game-thread, and not in a
thread next to the game-thread.
This commit changes that: it removes the threads for NewGRFScan
and GenerateWorld, and just runs the code in the game-thread.
On regular intervals it allows the draw-thread to do a tick,
which gives a much smoother look and feel.
It does slow down NewGRFScan and GenerateWorld ever so slightly
as it spends more time on drawing. But the slowdown is not
measureable on my machines (with 700+ NewGRFs / 4kx4k map and
a Debug build).
Running without a game-thread means NewGRFScan and GenerateWorld
are now blocking.
In other words, it should only (!) return true if A comes for B.
This promise was broken for the situation where two values are
identical. It would return true in these cases too. This is of
course not possible: if two values are identical, neither come
before the other. As such, the sorter was not imposing strict
weak ordering relations.
libstdc++ handled this scenario just fine, but libc++ crashes
badly on this, as it allowed comparing of [begin, end] instead
of [begin, end).
libc++ considered this not a bug (and by specs, they are correct;
just this way of crashing is of course a bit harsh):
https://bugs.llvm.org/show_bug.cgi?id=47903
Strictly seen, there are "N" people -waiting- in front of you
in the queue, but it is nicer to show "N + 1" for the person that
is currently downloading the map. Avoids it showing:
"0 clients in front of you". That just feels a bit off.
SendError() notifies all clients of the disconnect. This calls
CloseConnection() at the end, which also notified the clients
of the disconnect. Really no need to do it twice.
The status NETWORK_RECV_STATUS_SERVER_ERROR is only set by
SendError(), so in case that is the status, don't let
ClientConnection() send another notification.
If a server is compatible, it falls back to sorting by clients.
This used to be in reverse, so full servers are on top. With
the codechange commit, this was removed by accident, and as
such empty servers were on top. This is silly.
This is a much better location for this button, as you send
money from one company to another company, not from player
to player.
This is based on work done by JGRPP in:
f820543391
and surrounding commits, which took the work from estys:
https://www.tt-forums.net/viewtopic.php?p=1183311#p1183311
We did modify it to fix several bugs and clean up the code while
here anyway.
The callback was removed, as it meant a modified client could
prevent anyone from seeing money was transfered. The message
is now generated in the command itself, making that impossible.
Check if the job is still running two date fract ticks before it is due
to join, and if so pause the game until its done.
When loading a game, check if the game would block immediately due to
a job which is scheduled to be joined within two date fract ticks,
and if so pause the game until its done.
This avoids the main thread being blocked on a thread join, which appears
to the user as if the game is unresponsive, as the UI does not repaint
and cannot be interacted with.
Show if pause is due to link graph job in status bar, update network
messages.
This does not apply for network clients.
Apple Clang version 12 (bundled with Xcode 12) complaints about copying
small objects in range loop (-Wrange-loop-analysis introduced by -Wall).
This warning can be easily avoided by removing the reference from
the const pointer type.
Emscripten compiles to WASM, which can be loaded via
HTML / JavaScript. This allows you to play OpenTTD inside a
browser.
Co-authored-by: milek7 <me@milek7.pl>
Remove static buffer form of NetworkAddress::GetAddressAsString.
This is used in multiple threads concurrently, and is not thread-safe.
Replace it with a form returning std::string.
CMake works on all our supported platforms, like MSVC, Mingw, GCC,
Clang, and many more. It allows for a single way of doing things,
so no longer we need shell scripts and vbs scripts to work on all
our supported platforms.
Additionally, CMake allows to generate project files for like MSVC,
KDevelop, etc.
This heavily reduces the lines of code we need to support multiple
platforms from a project perspective.
Addtiionally, this heavily improves our detection of libraries, etc.
Previously the internal content list was invalidated and sorted for
every new item added. Now the sorting is delayed until the GUI is
drawn, which means we only sort once per GUI tick.
Since the amount of incoming items per GUI tick is not controlled by
the GUI but rather by network speed, we were previously doing a lot
of duplicate work per tick, causing the mouse cursor to lag while
the list was initialized.
This commit adds the missing feature of allowing the server owner to
provide a reason for kicking/banning a client, which the client sees in
a pop-up window after being kicked. The implementation extends the
network protocol by adding a new network action called
NETWORK_ACTION_KICKED that is capable of having an error string, unlike
the other network error packages. Additionally, the kick function
broadcasts a message to all clients about the kicked client and the
reason for the kick.
If the game was started loading a savegame or scenario, auto-restart will load a new random map.
This is inconsistent with the case in which a heightmap was loaded, as in that case the heightmap is kept as a basis for a new game.
This proposal solves this heterogeneity be considering the originally loaded resource shall be kept, hence savegames & scenarios shall be reloaded
Peter Nelson