- Retry was broken for the common case: retry_selected_download removed
the failed download only from the UI list, not the client's
DownloadStore. Since Client::download re-derives the same md5 token
from the unchanged filename, retrying a still-Failed download left a
duplicate-token entry; the token-migration path then deleted both,
misrouting completion so the retried file downloaded to disk but the
UI showed 'Queued' forever. Add Client::remove_download /
DownloadStore::remove_by_file (status-agnostic) and clear the stale
entry before re-issuing. Regression-tested.
- Chat-room member selection could point out of range: RoomEvent::UserLeft
shrank the active room's member list without clamping user_selected, so
the highlight silently shifted to the wrong user (or vanished, making
'b'/'m' no-ops, and blanking the member pane). Clamp user_selected after
every applied room event. Regression-tested.
Address the gap that the port mapper's UPnP path and gateway discovery
had no automated coverage (only NAT-PMP did):
- UPnP e2e: split map_via_gateway out of discovery and drive it through
igd-next against an in-process mock IGD (a std HTTP server returning a
canned SOAP envelope). Verifies the real add_port SOAP round-trip and
that we request the exact external/internal port + LAN client.
- default_gateway parsing is now pure, testable functions with unit
tests for the macOS and Linux formats (including the 0.0.0.0 skip).
- New 'portmap' CLI subcommand (needs no login): synchronously probes
UPnP/NAT-PMP against the real router and reports whether the port was
opened + the external address, then removes the test mapping — so a
user can verify reachability on their own network.
9 port_mapping tests now cover the NAT-PMP protocol, the UPnP SOAP
round-trip, gateway parsing, and the map/renew/cleanup lifecycle.
- Shutdown no longer freezes: Drop signalled stop but then blocked on a
thread join, and nothing observed stop during the blocking UPnP/NAT-PMP
discovery (up to ~9s). Drop now waits on a done-channel with a 1s cap
and detaches otherwise, so quitting is always prompt; the finite lease
cleans up if we detach mid-discovery. Regression-tested against an
unresponsive gateway.
- Linux default_gateway() no longer returns 0.0.0.0 for a gateway-less
(directly-connected) default route — it keeps scanning, matching the
macOS branch, instead of firing pointless NAT-PMP at the unspecified
address.
- UPnP now retries AddPortMapping with a permanent (lease 0) mapping when
the router rejects a timed lease (IGDs that only support permanent
leases), and renews/removes with the accepted lease.
Log lines were painting over the alternate-screen TUI because some
messages bypassed the buffered logger and wrote straight to the
terminal:
- a live println! in GetPeerAddressHandler fired on every peer-address
lookup (i.e. constantly while browsing/downloading) — routed to the
buffered debug! logger.
- the TUI's error handlers used eprintln! in the render loop and in
background search/download worker threads — routed to the buffered
warn! logger so failures no longer smear the screen.
The logger already defers to a buffer while the TUI holds the screen; the
fix is simply to go through it instead of print!/eprintln!.
Behind a home router the client is unreachable for incoming peer
connections, so browsing/downloading firewalled peers (and the
server-brokered connect-back) times out. Add a best-effort port mapper
to the binary crate that, when listening is enabled, makes us reachable:
- NAT-PMP (RFC 6886) hand-rolled on std::net::UdpSocket, zero deps, with
an injectable gateway address so it is fully testable.
- UPnP-IGD via the blocking igd-next crate as the primary path.
- A background PortMapper that discovers off the main thread (never
blocks startup), renews at half the lease, and removes the mapping on
drop. Any failure is non-fatal and logged.
- Maps the EXACT listen port only (we advertise it verbatim via
SetWaitPort, so a different external port would be unreachable) and
warns the user to forward manually otherwise.
Wired into the TUI, search, and CLI browse (the P2P modes); chat/rooms/
messaging are server-mediated and skip it. e2e-tested against an
in-process mock NAT-PMP gateway: the map request bytes, response parsing,
result-code/timeout handling, and the full spawn -> map -> drop -> unmap
service lifecycle.
- Retry actually retries: 'r' in the browse popup re-requests a
timed-out user (previously the 'Press b to retry' hint did nothing —
b wasn't handled inside the popup). BrowseTabs::retry_active is
unit-tested.
- Friendlier timeout copy that explains the likely cause: "Couldn't
reach <user>. They may be offline, or their connection can't be
reached (both of you may be behind a router/firewall). Press r to try
again."
- Replace protocol terminology in user-facing labels: 'Peer slots' ->
'Free slots', 'Peer upload speed' -> 'Upload speed'.
Add an e2e that forces the server-brokered (firewalled) browse path —
the path that matters on the real network. The sharer does NOT listen,
so the browser's direct dial fails and the server must broker the peer
connecting back to the browser's listener. This exercises
PeerConnectFailed -> ConnectToPeer -> inbound PierceFirewall ->
PeerConnected flush, which the localhost direct-dial test never hits.
Also soften user-facing download failure reasons from protocol jargon
("Peer …") to plain language ("The user went offline", etc.).
Replace the single browse view with BrowseTabs (one tab per user),
mirroring the chat-room tabs. Browsing a new user (from search results
or a chat room's member list) opens or focuses their tab; Tab/Shift-Tab
switch between open users, 'w' closes the active tab, Esc hides the
popup while keeping the tabs. A tab bar shows the open users when more
than one is open, and each tab polls/loads independently. Re-browsing a
timed-out user retries. BrowseTabs is unit-tested.
The member list in an open room is now selectable with the arrow keys.
Press 'b' to browse the highlighted user's shared files (opens the
browse tree) or 'm' to send them a private message (opens the compose
bar pre-filled with the recipient). Selection resets when switching
rooms; RoomsState navigation is unit-tested.
The same unbounded-count DoS class fixed in the room parsers also exists
in three pre-existing parsers that read a length-prefixed vector count
straight off the wire: FileSearchResponse (SearchResult), the browse
SharedFileListResponse, and ExcludedSearchPhrases. A malicious/MITM
server or peer could send a tiny frame with a ~4-billion element count
and spin the actor thread into an OOM allocation loop.
Bound each loop (and the inner attribute loops, where read_int32 does
not advance past the buffer end) by the remaining payload. Regression
tests feed a hostile count and assert prompt, empty parses.
Adversarial review of the chat-rooms feature surfaced seven confirmed
issues; all fixed with regression tests:
lib (DoS hardening — a malicious/MITM server could send a tiny frame
with a ~4-billion element count):
- room_list: bound the name/count vector loops by remaining payload.
- join_room: bound the member-list loop the same way.
TUI (soulseek-rs/src/models/rooms.rs, ui):
- a message for a room we already left no longer resurrects its tab
(route via open_index, not ensure_open).
- closing a tab now clears the unread badge on the room we land on.
- a room-list refresh keeps the highlight on the same room across a
re-sort (track selection by name, not index).
- Esc in the room list clears a lingering filter before closing, matching
the title's 'Esc: clear' hint.
- chat log renders one row per line so the newest message is never
clipped by wrapping.
Add a chat-rooms popup (open with 'c') satisfying the requested UX:
- Browsable, /-filterable room list showing per-room user counts,
busiest first; already-open rooms are marked.
- Multiple rooms open at once as tabs; Tab/Shift-Tab switch, 'x' leaves.
- Per-room chat log with a live member list and an inline compose line
(Enter to type, Enter to send; the server echo renders the message).
- Unread indicators: bold tab name + 'room (n)' badge per room, plus
'chat (n)' and 'inbox (n)' counters in the shortcut bar; viewing a
room clears its badge.
RoomsState is unit-tested (routing, unread accounting, tab lifecycle,
filtering).
Add `rooms` (list public rooms with user counts, busiest first) and
`chat <room> [message]` (join a room; say a message and exit, or listen
and print incoming messages/joins/leaves for a while). Factor a shared
connect_and_login helper.
Implement the Soulseek chat-room server messages against soulfind's
wire format:
- MessageFactory builders: room-list request (64), join (14),
leave (15), say (13).
- Handlers parsing the server side: RoomList (64) names+counts,
JoinRoom (14) membership, SayChatroom (13), UserJoinedRoom (16),
UserLeftRoom (17), LeaveRoom (15).
- Surface everything to the client as a drained RoomEvent stream plus
a room-list snapshot; new Client API: request_room_list, join_room,
leave_room, say_in_room, room_list, take_room_events.
Unit tests cover each builder and handler; server-gated e2e tests
exercise message delivery and room listing against soulfind.
Carry a human-readable reason on DownloadStatus::Failed so it flows
through the existing status channel to the TUI. The download Info pane
now shows an Error line for a failed download (connect failure, peer
disconnect, truncated transfer, peer-reported upload failure, or the
underlying transfer error). Failures without a known reason render as
before.
In the Downloads pane: 'r' re-queues the selected failed/timed-out download,
and 'c' clears all completed/failed/timed-out entries from the list. Shortcuts
bar updated.
Wire browsing a user's shared files into the TUI as a modal tree popup:
- press 'b' on a highlighted search result to browse that result's owner, or
'b' elsewhere to type a username (new CommandBarMode::Browse);
- a new models/browse.rs synthesizes the flat SharedDirectory list into a
nested, collapsible tree (intermediate folders reconstructed from the
backslash paths; file download paths kept byte-exact from the original names),
with expand/collapse state and a flatten-to-rows renderer — fully unit-tested;
- the popup polls take_browse_result each tick (with a timeout), navigates with
arrows/hjkl, expands/collapses folders, and queues downloads: Enter on a file,
'd' on a file or an entire folder subtree (recursively) — all funneled into
the existing Downloads pane;
- shortcuts bar advertises browse keys and surfaces our own sharing status.
Add Client::browse_user (sends GetShareFileList, code 4, connecting first if
needed via the same queue-and-flush path as search replies) and
take_browse_result. A new SharedFileListResponseHandler parses the peer's
code-5 reply into SharedDirectory entries, forwarded as
ClientOperation::BrowseResult and stored per user. Verified end to end: a
second real client browses a sharer and receives its file listing.
GetShareFileList (peer code 4) previously replied with a malformed message (a
server SharedFolders stats message over a peer connection). It now emits
ShareListRequested, and the client builds a real SharedFileListResponse
(code 5) from its shares grouped by directory. Adds a std-only code-5
encoder/decoder (build_shared_file_list / parse_shared_file_list) with a
roundtrip test, and Shares::directories() to group files. The decoder is the
foundation for the Browse feature.
End-to-end proof of the whole sharing stack with no mock peer: a sharer indexes
a temp file, the leecher's search is distributed by soulfind, the sharer replies
with a FileSearchResponse, and the leecher downloads the real bytes that the
sharer serves over a peer-to-peer transfer. Asserts the search result and that
the downloaded bytes match the source. Server-gated (SOULSEEK_E2E_REQUIRED).
Implement the upload path so a peer can download our shared files:
- fix the mis-coded PlaceInQueueResponse (43 -> 44) and add a real QueueUpload
handler (code 43) -> PeerMessage::IncomingQueueUpload;
- the peer actor forwards it as ClientOperation::QueueUpload; the client looks
the file up in its shares, mints an upload token (high half of the token
space, disjoint from md5 download tokens), and offers it via ServeUpload ->
build_upload_transfer_request (peer code 40, direction=upload);
- a TransferResponse for one of our serving tokens is handled by a surgical
early-return that emits StartUpload, leaving the download path untouched;
- StartUpload resolves the downloader's address (GetPeerAddress, cached) and
streams the file via a new peer::upload_peer::serve_file over an F connection.
Tested: build_upload_transfer_request decodes via the production Transfer
parser; serve_file streams a file correctly over a loopback F connection.
Wire the incoming server FileSearch (code 26): the handler now forwards
ServerMessage::FileSearchRequest -> ClientOperation::IncomingSearch instead of
just logging. The client matches the query against its shares and, if anything
matches, sends a FileSearchResponse to the searcher — connecting to them first
(GetPeerAddress -> outbound P) and flushing the response on PeerConnected when
not already connected. GetPeerAddressResponse now also caches peer addresses
for the serve path. Adds build_search_response + a unit test.
Add build_file_search_response + a borrowed FileEntry view: the zlib-compressed
reply advertising our matching shared files to a searcher. The payload is the
exact inverse of SearchResult::new_from_message (username, token, per-file
name/size/ext/attributes, slots, speed), compressed with the stored-block
encoder. Round-trip tested through the production decoder.
The post-login handshake (SharedFolders / HaveNoParent / SetStatus /
SetWaitPort) lived only in ServerActor::login(), which nothing ever called —
so a live client never advertised its listen port or share counts. Move it to
a testable post_login_messages() fired from the LoginStatus(true) arm (once, on
success, on the live path), and delete the dead login().
Client::connect now scans ClientSettings.shared_directory once into
ClientContext.shares and passes the real folder/file counts to the server.
This is the prerequisite for two live clients to resolve each other's
addresses (GetPeerAddress needs the peer to have sent SetWaitPort).
Unit-tested post_login_messages; the login/listener e2e still passes, proving
soulfind accepts the handshake.
Add a std-only shares module: Shares::scan walks the shared directory (skipping
symlinks and unreadable entries) into a read-only snapshot keyed by peer-facing
virtual path (root name + backslash-separated relative path). Exposes
search (case-insensitive multi-term AND), get (O(1) by virtual path), and
file/folder counts for the login handshake. Unit-tested against a temp tree.
Soulseek peer messages (SharedFileListResponse, FileSearchResponse) are
zlib-compressed on the wire, but the library only had a decoder. Add
compress_stored(): a valid zlib stream built entirely from uncompressed
DEFLATE 'stored' blocks (0x78 0x01 header, blocks of <= 0xFFFF bytes, adler32),
which needs no real compressor and is accepted by standard decoders and by the
existing inflate path. Roundtrip-tested through the decoder at block
boundaries, plus header/adler32 vectors.
When a healthy peer actor was replaced under the same username (both sides
connecting at once, or two overlapping downloads to the same not-yet-connected
peer), its graceful shutdown still emitted PeerDisconnected, and the client
evicted by username — removing and stopping the NEWER actor now occupying the
slot, tearing down the live connection and stalling that peer's downloads.
Give each peer actor a unique id (assigned by the registry) and echo it in
PeerDisconnected/PeerConnectFailed. The client now evicts via
remove_peer_if(username, id), which removes the entry only when the reporting
actor still owns the slot, so a stale actor's shutdown is a no-op. The brokered
timeout no longer removes an actor by username either, avoiding the same
aliasing. Add a registry unit test for the identity guard.
When the server asks us to connect to a peer (a ConnectToPeer we receive,
carrying the server's correlation token), we dial that peer and must announce
ourselves with a PierceFirewall quoting the token — not a PeerInit — so the
remote peer can correlate its own indirect request (search-result / upload
delivery). The outbound handshake now sends PierceFirewall when the peer
carries a server token and PeerInit only for direct GetPeerAddress dials
(token None). Direct and firewalled e2e paths are unchanged.
Two related lifecycle defects in the direct/firewalled download path:
- On a failed direct connect, ClientOperation::PeerConnectFailed left the dead
outbound actor registered and running. The actor pinned a thread-pool worker
forever (its loop only exits on Stop), so a handful of unreachable-peer
downloads could exhaust the fixed pool and wedge the whole client; and a
later download() to the same peer saw registry.contains()==true and queued
into the dead, streamless actor, hanging silently. The arm now removes and
stops the dead actor before brokering, mirroring PeerDisconnected.
- The server-brokered fallback had no timeout, so a peer that never connects
back (offline / double-firewalled / unroutable address) left the download
Queued forever and its Receiver blocked. Arm a bounded timeout: if no
PierceFirewall consumes the correlation token within BROKER_CONNECT_TIMEOUT,
fail the peer's queued downloads (on the channel and in the store) and
reclaim the token. A successful pierce consumes the token first, so the
timeout is a no-op on the happy path.
Add a unit test for the queued-download failure helper; the direct and
firewalled e2e tests still pass (pierce arrives well within the timeout).
GetPeerAddressResponse narrowed obfuscated_port (u16) into a u8 via
try_into().unwrap(), panicking for any real obfuscated port (>255) and
unwinding the client's single ops-loop thread — hanging all downloads,
searches and peer events. Since download() now auto-resolves peer addresses
via GetPeerAddress, a download to any obfuscation-enabled peer could crash the
whole client. Widen Peer::obfuscated_port to u16 and stop truncating.
Implement the FIREWALLED connection mode: when a direct outbound connection to
a peer fails (peer unreachable), fall back to asking the server to broker it.
- Add MessageFactory::build_connect_to_peer (server code 18).
- A pre-established outbound failure now emits ClientOperation::PeerConnectFailed
instead of a normal disconnect; the client registers a correlation token and
sends ConnectToPeer so the peer connects back.
- The listener accepts the peer's PierceFirewall (peer code 0) connect-back,
matches the pending token, registers it as the P control connection, and
fires PeerConnected to flush the queued download.
Fix a latent aliasing bug this surfaced: a peer actor could emit two terminal
notifications (a connect-failure, then a disconnect when later replaced and
stopped), and the second removed the *newer* actor registered under the same
username, killing the brokered connection mid-transfer. An actor now reports a
terminal outcome at most once (disconnect_reported guard).
Verified end to end against soulfind: the mock advertises an unlistened port so
the direct connect is refused, the server brokers the connection, the mock
pierces back, and the file transfers — bytes match. A companion test
(soulfind_brokers_connect_to_peer_between_users) pins the server behaviour the
mode relies on.
Implement the DIRECT connection-initiation mode so a download can target a
peer the client is not already connected to:
- Fix the outbound peer handshake. on_connection_established sent
build_watch_user() (a server WatchUser message, wrongly framed with
get_data()) for outbound connections; it now sends a proper length-prefixed
PeerInit (peer code 1) with our own username. own_username is plumbed in
via PeerRegistry -> PeerActor; inbound peers (outbound=false) are unchanged.
- download() now auto-connects: if there is no control connection to the peer
it sends GetPeerAddress and defers the QueueUpload until the outbound stream
handshakes, signalled by a new ClientOperation::PeerConnected. A never-
connected client still resolves the download to Failed (unit test).
- Add Client::connect_peer(username) to open/pre-warm a control connection.
Verified end to end against soulfind: a_file_downloads_from_a_peer_via_direct_connection
logs a mock peer into the server, advertises its port via SetWaitPort, and the
client resolves the address, dials it, and downloads the file — bytes match.
The F-serve leg is shared with the passive-download mock.
A peer actor multiplexes socket reads with its mailbox on one thread: tick()
reads the socket while outgoing messages (e.g. a queued QueueUpload that starts
a download) are delivered through the mailbox between ticks. Inbound
connections accepted by the listener were left in blocking mode, so once such a
peer went idle tick() parked inside read() and never drained the mailbox — a
download queued against that peer never sent its QueueUpload and hung. Outbound
connections already set non-blocking; do the same for inbound P connections.
Covered by a new server-optional e2e test that downloads a file end to end from
an in-process mock peer over real P (control) and F (transfer) connections,
asserting the bytes written match the source. The mock speaks the library's own
public Message wire format, so no dependencies are added.
Add private messaging to the TUI:
- 'm' opens the command bar in message mode; type '<recipient> <text>' and
Enter sends it via the client;
- incoming messages are polled each tick into an inbox;
- 'i' toggles a centered inbox popup listing sent/received messages.
The shared command bar now carries a mode (search vs message) so the prefix,
submit action and shortcut hints follow what is being typed.
Wire private messaging through the client:
- MessageFactory::build_message_user (code 22) and build_message_acked
(code 23);
- the incoming MessageUser handler now acknowledges freshly delivered
messages and forwards them to the client instead of only logging them;
- ClientContext buffers received UserMessages;
- Client::send_private_message and Client::take_private_messages expose the
feature, and UserMessage (with accessors) is re-exported from the crate.
Verified end to end against soulfind: a_private_message_is_delivered_between_users
logs in two clients and asserts one receives the other's message.
Cover implemented, server-verifiable paths that were previously untested:
- assert an issued search is registered in client state, not just that
results are empty;
- a_search_is_forwarded_to_a_connected_peer: a second client exercises the
incoming FileSearch handler (the server distributes searches to it), and we
prove that session survives by round-tripping its own search afterwards;
- login_succeeds_with_listener_enabled: drives the post-login handshake with
the peer listener on (SetWaitPort), which every other e2e test skips.
Peer downloads / file sharing remain out of scope here: they need a second
peer that actually serves files, which the client does not implement yet.
Extend the existing CI workflow with an ubuntu-only "End-to-end (soulfind)"
job that builds soulfind from source (LDC + `dub build :server`), points
SOULFIND_BIN at it, and runs the e2e suite with SOULSEEK_E2E_REQUIRED=1.
Also widen the clippy job to the whole workspace.
The e2e harness gains SOULSEEK_E2E_REQUIRED: when set, a missing/unstartable
server is a hard failure instead of a silent skip, so CI genuinely exercises
the client end-to-end. Add e2e coverage for two existing behaviours a single
server can verify: wrong-password rejection and concurrent logins.
Add a --shared-dir / SOULSEEK_SHARED_DIR option and a SOULSEEK_DOWNLOAD_DIR
env for the download directory. The shared directory is resolved and
validated up front (tilde-expanded, must exist and be a directory); a
misconfigured value is a warning, not a fatal error.
The library gains an optional shared_directory on ClientSettings/Client
with a getter, keeping soulseek-rs-lib dependency-free. Resolution lives
in a new soulseek-rs/src/directories.rs with unit tests.
Adds soulseek-rs-lib/tests/e2e.rs, which drives the client's public API
(connect, login, search) against a real Soulseek server. The harness is
server-optional so cargo test stays green everywhere: it uses
SOULSEEK_TEST_SERVER if set, otherwise spawns a soulfind binary (SOULFIND_BIN
or a soulfind/bin/soulfind checkout in a parent dir) on an ephemeral port with
a throwaway database, and skips with a notice when no server is available.
Uses only std and the library's public API — no new dependencies. Documents how
to run it (including the macOS soulfind SQLite workaround) in the README.
Enable the full clippy all + pedantic + nursery groups as workspace warnings,
with a small documented allow-list for the pedantic/nursery lints this project
intentionally does not follow (rustdoc-style requirements, deliberate numeric
casts in the protocol/UI layers, and a few noisy nursery lints). Also enable
the rust unused_lifetimes and unsafe_op_in_unsafe_fn lints.
Fixed every resulting warning (527 -> 0) mechanically per clippy's suggestions:
inline format args, const fns, item ordering, let-else, redundant clones,
map_or, cast_lossless, Eq derives, associated-fn conversions for methods that
never used self, and removing unnecessary Result wrappers whose call sites could
never propagate an error. No runtime behavior changed; cargo fmt applied.
cargo clippy --workspace --all-targets: 0 warnings. All tests pass.
MainTui::run only disabled mouse capture and called ratatui::restore() after
the event loop exited normally; any error propagated by ? from draw/poll/read
returned early and left the terminal in raw mode, on the alternate screen, with
mouse capture still enabled. The loop body moved into a helper whose result is
returned only after the terminal is unconditionally restored.
Michel de Graaf