Green Sky
aae086cc65
b03b571272 fix: flaky tcp test This only fixes the symptoms, not the real problem. Sometimes or consistently on some platforms a socket might need a moment before it can be written to. 32e67ab4c2 cleanup: use typedef for private message ID's in callback 7b1db6adc1 feat: add message IDs to private group messages 99e0bcc27d refactor: Observers/ignored peers can now send and receive custom packets b3c3c49d26 fix: Disable IPv6 in Windows cross-compilation tests e742deddff feat: Check hashes of Windows dependencies when cross-compiling dfb9a0b02b fix: Test the current Windows Dockerfile, not an old Dockerhub image 14de93ccec chore: Use WineHQ's Wine as Debian Bookworm's crashes ed37616249 docs: Update the Windows cross-compilation section 9bb79c174f cleanup: Remove a couple of unnecessary misc_tools dependencies 19475adb70 chore: Statically link OpenMP into the cracker fun util on Windows 1be311e51f feat: Build the fun utils when cross-compiling to Windows 88133f8446 chore: Strip Windows binaries 3cc0ae7535 refactor: Copy over all of the required static dependencies c4fa8f7fb1 feat: Generate .def, .exp and .lib files when building for Windows 74bbac5363 feat: Let CMake create the dll instead of doing so ourselves 246642e9ae feat: Harden Windows cross-compilation 8d431c0d11 chore: Bump Windows build dependency versions e519f7998b fix: Remove unnecessary wsock32 dependency on Windows ed2b60c217 chore: Use a specific non-broken slimcc version. d7f21010a1 chore: Update github actions. e71a68b7f2 docs: Update the list of CMake options 77e08876ff chore: Remove mod and founder from group API naming scheme 12bc042767 docs: add the experimental api build option to INSTALL.md e1fa5cae96 refactor: Rename Queries to Query to align with other enums. be82a3ea30 fix: Correct type for conference offline peer numbers. 0627c36716 test: Add pkgsrc build. 92578afe4b test: Add FreeBSD VM action on GitHub. 52ece0f57b test: Build toxcore on NetBSD (VM). 3fe8ee2c11 chore: Only install tox_private.h on request. 9a8dfa06ab fix: save_compatibility_test failing on big-endian systems 86f5e55578 fix: Don't serve files from websockify. 710eb674a5 fix: Correctly pass extended public keys to group moderation code. 021db7031c refactor: Use `struct`s for extended public/secret keys. a1e999fd80 chore: Compile libsodium reference implementation with compcert. fbe3c19cf5 cleanup: correct a few nullable annotations 623e3ee5c3 cleanup: Don't use `memcpy` to cast arbitrary `struct`s to `uint8_t[]`. c71567dc18 fix: Pass array, not array pointer, to `memcmp`. 9b46a08144 cleanup: Never pass `void*` directly to `memcpy`. 5d7b7a7bbc refactor: Use tox rng to seed the keypair generation. 961891d568 cleanup: Small improvements found by PVS Studio. 8201019f0d chore: Disable NGC saving by default, enable through Tox_Options. 5dd9ee3f65 cleanup: Replace pointer arithmetic with explicit `&arr[i]`. ca4606d49d refactor: Use strong typedef for NGC peer id. 442213b722 cleanup: Simplify custom packet length check in NGC. 08d3393def fix: Correct a few potential null derefs in bootstrap daemon. b9877b32b0 fix: Add missing memunlock of local variable when it goes out of scope. dab5fe44b9 fix: Zero out stack-allocated secret key before return. f058103299 refactor: Make prune_gc_sanctions_list more obviously correct. 3ba7a0dec9 docs: Add static analysis tool list to README. 8d0811a0f3 docs: Run prettier-markdown on markdown files. 969e3a2bfc refactor: Fix network test not using the strong typedef 93c83fbc7c refactor: Use strong typedef instead of struct for `Socket`. 9fe18b176f fix: Fix some false positive from PVS Studio. 7c44379ccb cleanup: Check that WINXP macro exists before comparing it. 5c93231bef refactor: Make tox mutex non-recursive. aacff73939 docs: Fix up doxyfile. d55fc85ff5 docs: Add more documentation to crypto_core. 5bdaaaedb6 refactor: Remove `Tox *` from `tox_dispatch`. e202341e76 refactor: Don't rely on tox_dispatch passing tox in tests. 34df938f52 chore: Use C++ mode for clang-tidy. 8b05296a78 chore: Check that both gtest and gmock exist for tests. 42010660e1 test: Add slimcc compiler compatibility test. b473630321 chore: Add some comments to the astyle config. b7404f24f6 cleanup: Remove implicit bool conversions. 4e2dba4d9f chore: Reformat sources with astyle. 4359e3a6bc chore: Rename C++ headers to .hh suffixes. 0c05566e58 cleanup: Further `#include` cleanups. 8d29935b7a chore: Only check the bootstrap daemon checksum on release. f70e588bc6 cleanup: Add more `const` where possible. 511bfe39c8 cleanup: Use Bazel modules to enforce proper `#include` hygiene. 1710a0d091 refactor: Move pack/unpack `IP_Port` from DHT into network module. a975943564 chore: Really fix coverage docker image build. c08409390f chore: Fix post-submit coverage image. 39aadf8922 fix: Don't use `memcmp` to compare `IP_Port`s. d94246a906 fix: partially fix a bug that prevented group part messages from sending. eeaa039222 chore: Fix rpm build; add a CI check for it. 8328449c1a chore: Speed up docker builds a bit by reducing layer count. d6d67d56f3 cleanup: Add `const` where possible in auto tests. 6aa9e6850d cleanup: Minor cleanup of event unpack code. bdf460a3a9 refactor: Rename `system_{memory,...}` to `os_{memory,...}`. 203e1af81e fix: a few off by one errors in group autotests 5c093c4888 cleanup: Remove all uses of `SIZEOF_VLA`. 662c2140f3 test: Add goblint static analyser. 8f07755834 cleanup: Use `memzero(x, s)` instead of `memset(x, 0, s)`. a7258e40cf cleanup: Use explicit 0 instead of `PACKET_ID_PADDING`. 6370d0f15d cleanup: Expand the `Tox_Options` accessor macros. 14a1a0b9bd cleanup: Remove plan9 support. a05dccad13 test: Add a simple new/delete test for Tox. 1cdcf938b9 cleanup: Add comment after every `#endif`. ba99d4dc4b test: Fix comment I broke in the events test PR. e07248debb refactor: Migrate auto_tests to new events API. bdd42b5452 refactor: Add common msgpack array packer with callback. 3c659f5288 cleanup: Rename group to conference in groupav documentation. 89957be230 cleanup: Ensure handler params are named after callback params. c650d9d345 refactor: Pass `this` pointer as first param to s11n callbacks. e7fb91ddb8 refactor: Allow NULL pointers for byte arrays in events. 5e2c8cabc1 cleanup: make some improvements to group moderation test 259de4867e cleanup: Remove `bin_pack_{new,free}`. 21a8ff5895 cleanup: skip a do_gc iteration before removing peers marked for deletion 16809dc36e feat: Add dht_get_nodes_response event to the events system. git-subtree-dir: external/toxcore/c-toxcore git-subtree-split: b03b5712720de9a9901ea12fd741f177327a7021
134 lines
5.5 KiB
Markdown
134 lines
5.5 KiB
Markdown
# Persistent conferences
|
|
|
|
This document describes the "minpgc" simple persistent conferences
|
|
implementation of PR #1069.
|
|
|
|
Many of the ideas derive from isotoxin's persistent conferences implementation,
|
|
PR #826.
|
|
|
|
## Specification of changes from pre-existing conference specification
|
|
|
|
We add one new packet type:
|
|
|
|
Rejoin Conference packet
|
|
|
|
| Length | Contents |
|
|
| :----- | :-------------------- |
|
|
| `1` | `uint8_t` (0x64) |
|
|
| `33` | Group chat identifier |
|
|
|
|
A peer times out from a group if it has been inactive for 60s. When a peer times
|
|
out, we flag it as _frozen_. Frozen peers are disregarded for all purposes
|
|
except those discussed below - in particular no packets are sent to them except
|
|
as described below, they are omitted from the peer lists sent to the client or
|
|
in a Peer Response packet, and they are not considered when determining closest
|
|
peers for establishing direct connections.
|
|
|
|
A peer is considered to be active if we receive a group message or Rejoin packet
|
|
from it, or a New Peer message for it.
|
|
|
|
If a frozen peer is seen to be active, we remove its 'frozen' flag and send a
|
|
Name group message. (We can hold off on sending this message until the next
|
|
`tox_iterate`, and only send one message if many frozen peers become active at
|
|
once).
|
|
|
|
If we receive a New Peer message for a peer, we update its DHT pubkey.
|
|
|
|
If we receive a group message originating from an unknown peer, we drop the
|
|
message but send a Peer Query packet back to the peer who directly sent us the
|
|
message. (This is current behaviour; it's mentioned here because it's important
|
|
and not currently mentioned in the spec.)
|
|
|
|
If we receive a Rejoin packet from a peer we update its DHT pubkey, add a
|
|
temporary groupchat connection for the peer, and, once the connection is online,
|
|
send out a New Peer message announcing the peer, and a Name message.
|
|
|
|
Whenever we make a new friend connection, we check if the public key is that of
|
|
any frozen peer. If so, we send it a Rejoin packet, add a temporary groupchat
|
|
connection for it, and, once the connection is online, send the peer a Peer
|
|
Query packet.
|
|
|
|
We do the same with a peer when we are setting it as frozen if we have a friend
|
|
connection to it.
|
|
|
|
The temporary groupchat connections established in sending and handling Rejoin
|
|
packets are not immediately operational (because group numbers are not known);
|
|
rather, an Online packet is sent when we handle a Rejoin packet.
|
|
|
|
When a connection is set as online as a result of an Online packet, we ping the
|
|
group.
|
|
|
|
When processing the reply to a Peer Query, we update the DHT pubkey of an
|
|
existing peer if and only if it is frozen or has not had its DHT pubkey updated
|
|
since it last stopped being frozen.
|
|
|
|
When we receive a Title Response packet, we set the title if it has never been
|
|
set or if at some point since it was last set, there were no unfrozen peers
|
|
(except us).
|
|
|
|
## Discussion
|
|
|
|
### Overview
|
|
|
|
The intention is to recover seamlessly from splits in the group, the most common
|
|
form of which is a single peer temporarily losing all connectivity.
|
|
|
|
To see how this works, first note that groups (even before the changes discussed
|
|
here) have the property that for a group to be connected in the sense that any
|
|
peer will receive the messages of any other peer and have them in their
|
|
peerlist, it is necessary and sufficient that there is a path of direct group
|
|
connections between any two peers.
|
|
|
|
Now suppose the group is split into two connected components, with each member
|
|
of one component frozen according to the members of the other. Suppose there are
|
|
two peers, one in each component, which are using the above protocol, and
|
|
suppose they establish a friend connection. Then each will rejoin the other,
|
|
forming a direct group connection. Hence the whole group will become connected
|
|
(even if all other peers are using the unmodified protocol).
|
|
|
|
The Peer Query packet sent on rejoining hastens this process.
|
|
|
|
Peers who leave the group during a split will not be deleted by all peers after
|
|
the merge - but they will be set as frozen due to ping timeouts, which is
|
|
sufficient.
|
|
|
|
### Titles
|
|
|
|
If we have a split into components each containing multiple peers, and the title
|
|
is changed in one component, then peers will continue to disagree on the title
|
|
after the split. Short of a complicated voting system, this seems the only
|
|
reasonable behaviour.
|
|
|
|
### Implementation notes
|
|
|
|
Although I've described the logic in terms of an 'frozen' flag, it might
|
|
actually make more sense in the implementation to have a separate list for
|
|
frozen peers.
|
|
|
|
## Saving
|
|
|
|
Saving is implemented by simply saving all live groups with their group numbers
|
|
and full peer info for all peers. On reload, all peers are set as frozen.
|
|
|
|
Clients needs to support this by understanding that groups may exist on
|
|
start-up. Clients should call `tox_conference_get_chatlist` to obtain them. A
|
|
group which is deleted (with `tox_conference_delete`) is removed permanently and
|
|
will not be saved.
|
|
|
|
## Limitations
|
|
|
|
If a peer disconnects from the group for a period short enough that group
|
|
timeouts do not occur, and a name change occurs during this period, then the
|
|
name change will never be propagated.
|
|
|
|
One way to deal with this would be a general mechanism for storing and
|
|
requesting missed group messages. But this is considered out of scope of this
|
|
PR.
|
|
|
|
If a peer changes its DHT pubkey, the change might not be properly propagated
|
|
under various circumstances - in particular, if connections do not go down long
|
|
enough for the peer to become frozen.
|
|
|
|
One way to deal with this would be to add a group message announcing the sending
|
|
peer's current DHT pubkey, and treat it analogously to the Name message.
|