pub enum ConnectionError {
Show 19 variants
Endpoint(EndpointError),
Codec(CodecError),
Transport(TransportError),
VarInt(VarIntError),
NoControlStream,
UnexpectedEnd,
StreamFinished,
InvalidAddress(String),
TlsConfig(String),
DataStreamState(&'static str),
ControlMessageNarrowing,
RequestOnControlStream(MessageType),
RequestStreamMessageOnControlStream(MessageType),
PayloadOnStatusDatagram {
object_id: u64,
payload_len: usize,
status: Option<ObjectStatus>,
},
PropertiesOnNonNormalStatus {
object_id: u64,
properties_len: usize,
status: ObjectStatus,
},
NonRequestOnRequestStream(MessageType),
RespondedToOwnRequest(u64),
FinBeforeResponse(u64),
FinBeforePublishDone(u64),
}Expand description
Errors from the connection layer.
Variants§
Endpoint(EndpointError)
Endpoint state machine error.
Codec(CodecError)
Wire codec error.
Transport(TransportError)
Transport-level error.
VarInt(VarIntError)
Variable-length integer decoding error.
NoControlStream
Control stream was not opened.
UnexpectedEnd
Stream ended before a complete message was read.
StreamFinished
Stream was finished by the peer.
InvalidAddress(String)
Invalid server address string.
TlsConfig(String)
TLS configuration error.
DataStreamState(&'static str)
Data stream used out of order (e.g. object before header).
ControlMessageNarrowing
A control message this build decoded for draft-21 and then could not narrow to draft-21’s own message type.
Unreachable, and that is not the same as harmless. read_control
decodes with this connection’s own draft, so the AnyControlMessage it
hands back can only carry this draft’s variant — but the narrowing arm
is compiled in every configuration anyway, under
#[allow(unreachable_patterns)] rather than a cfg naming the other
drafts, because such a list has to be edited in every per-draft
module whenever a draft is added and a copy that omits one leaves the
match non-exhaustive.
Spelled as CodecError::UnknownMessageType(0) it would not stay inert:
every draft’s
codec_session_error_code
answers that variant Some(PROTOCOL_VIOLATION). So the day the
narrowing did fail, this build’s own defect would reach a caller as the
peer sent a control message type this draft does not assign, and the
session must be closed with a Protocol Violation — carrying 0x00 as
the codepoint that proved it. A conformance report reading that
publishes a named, well-evidenced accusation against a relay for
something no relay did.
A variant of its own is what stops that.
draft_specific_cause answers it
LocalRefusal, the facade turns that into ErrorCause::Facade, and
nothing downstream can read a rule out of a cause that says nothing
reached the wire. What is pinned is the consequence rather than the
unreachability: nothing pins the arm’s reachability, which is exactly
why the consequence must not be an accusation.
RequestOnControlStream(MessageType)
A message that begins a request stream was handed to
Connection::send_control. Nothing was written.
RequestStreamMessageOnControlStream(MessageType)
A message draft-21 places on a request stream was handed to
Connection::send_control. Nothing was written.
Distinct from ConnectionError::RequestOnControlStream, which is about
a message that would begin a request stream of its own. This one is
about a message that belongs on a request stream already open, and so has
no meaning without one around it.
PayloadOnStatusDatagram
A datagram carrying an Object Status arrived with bytes after its header. Draft-21 Section 11.1.2: “An Object MUST have an empty payload unless its Object Status value is registered as permitting a payload in the Object Status registry” Section 11.2.1 says the same thing about the framing: a datagram with the STATUS bit set “is present and there is no Object Payload.”
The codec cannot see this. DatagramHeader::decode stops at the end of
the header and never owns the datagram’s tail, so the only layer that
holds both the status and the bytes after it is this one. Without the
check the application is handed, say, an End-of-Group object carrying
four bytes of payload — a combination the draft forbids outright.
Recoverable: the drafts state the rule as a property of a conforming object, not as one of the “MUST close the session” cases, so the datagram is refused and the session left running.
Fields
status: Option<ObjectStatus>The status the header declared.
Spelled out in full because the glob import of moqtap_codec::types
brings a different ObjectStatus into this module.
PropertiesOnNonNormalStatus
An Object arrived carrying properties on a status that is not Normal.
Draft-21 Section 11.1.3: “Any Object with status Normal can have properties (Section 8.4). If an endpoint receives properties on an Object with status that is not Normal, it MUST close the session with a PROTOCOL_VIOLATION.”
The codec decodes such an Object rather than refusing it — the frame is well formed, and a tool that reports non-conforming traffic has to be able to read it. Being an endpoint rather than an observer is what turns it into an error, so it is raised here, on the receive path, and not in the decoder.
Fields
status: ObjectStatusThe Object’s status, resolved through the encoding’s elision rule.
Spelled out in full because the glob import of moqtap_codec::types
brings a different ObjectStatus into this module.
NonRequestOnRequestStream(MessageType)
A bidirectional stream the peer opened began with a message type that does not open a request stream.
Draft-21 Section 6.3: “Bidirectional streams MUST NOT begin with any other message type unless negotiated. If they do, the peer MUST close the Session with a PROTOCOL_VIOLATION.” The session has already been closed on the wire by the time this is returned, and the offending stream reset.
RespondedToOwnRequest(u64)
A respond_* helper was handed a request stream this endpoint opened.
Nothing was written and no state moved.
FinBeforeResponse(u64)
RequestStream::finish was called on a request stream the peer opened
before a response had been written on it. Nothing was written.
Draft-21 Section 6.4.2.2: “An endpoint MUST NOT send a FIN on a direction of a request stream until it has sent all required messages on that direction for its request type. In particular, an endpoint sending a response to a request MUST send the corresponding response message … before sending a FIN.” The section arrived in draft-19 and draft-21 keeps it verbatim; drafts 17 and 18 have no such rule and no such guard.
FinBeforePublishDone(u64)
RequestStream::finish was called on a subscription this endpoint is
the publisher of, before PUBLISH_DONE. Nothing was written.
The second half of the same sentence in draft-21 Section 6.4.2.2: “the
publisher of an Established subscription MUST send PUBLISH_DONE, before
sending a FIN.” It applies to the accepted SUBSCRIBE a responder
answered with SUBSCRIBE_OK and to the PUBLISH this endpoint sent, which
is why Connection::publish_done and
Connection::publish_done_on are the only routes that clear it.
Trait Implementations§
Source§impl Debug for ConnectionError
impl Debug for ConnectionError
Source§impl Display for ConnectionError
impl Display for ConnectionError
Source§impl Error for ConnectionError
impl Error for ConnectionError
Source§fn source(&self) -> Option<&(dyn Error + 'static)>
fn source(&self) -> Option<&(dyn Error + 'static)>
1.0.0 · Source§fn description(&self) -> &str
fn description(&self) -> &str
use the Display impl or to_string()
Source§impl From<CodecError> for ConnectionError
impl From<CodecError> for ConnectionError
Source§fn from(source: CodecError) -> Self
fn from(source: CodecError) -> Self
Source§impl From<ConnectionError> for AnyConnectionError
Available on crate feature draft21 only.
impl From<ConnectionError> for AnyConnectionError
draft21 only.Source§fn from(err: ConnectionError) -> Self
fn from(err: ConnectionError) -> Self
Source§impl From<DialError> for ConnectionError
impl From<DialError> for ConnectionError
Source§fn from(e: DialError) -> Self
fn from(e: DialError) -> Self
Maps a dial failure onto the variants this error already has, so a
caller matches InvalidAddress or TlsConfig.
§LocalSocket joins InvalidAddress, and that is the answer being kept
A socket this machine would not open has a variant of its own on
DialError, and it still arrives here.
Not laziness about the churn — InvalidAddress is one of the
variants the facade reads as
ErrorCause::Facade, which
is_local answers true for, and a failed bind is this side’s by
definition. Routing it to Transport would read better in prose and
would publish this machine’s missing IPv6 stack as the relay’s doing.
The phase is not lost, only unread on this path. A caller measuring
which stage of a dial died reads
DialError::phase off the dial
itself; a caller who arrived at this type named a host:port and asked
for a connection, not for a measurement, and a public variant here for
a distinction nothing on this path reads is churn with no reader, which
is why this impl stays flat.