Previously the exported Disconnect and Shutdown functions called each other
and therefore needed to release and reacquire the locks which could
potentionally allow a request to sneak in between the lock and unlock.
This commit changes the exported Shutdown function so that everything is
done under the request lock to prevent this possibility. In order to
support this, the shared code between the Disconnect and Shutdown
functions has been refactored into unexported functions and the locking
has been altered accordingly.
Also, replace the sendMessage function with a select over the send and
disconnect channels to prevent any issues with queued up messages.
joint debugging effort between myself and @jrick
Previously, requests could still be sent to a shutdown client and
added to the client's internal data structures, without ever
responding to the future with an error for a shutdown client (causing
hangs when blocking on the future receive). This change fixes this by
performing a non-blocking read of the client's shutdown channel before
adding a request, and responding with the shutdown error if the client
has begun or completed its shutdown.
ok @davecgh
This commit modifies the error handling for websocket connections to fall
back to returning the status text returned from the server if the
handshake fails and it's not due to an authentication or invalid endpoint.
This change set equips the RPC client with handling of non successful
HTTP responses. An HTTP response is considered non successful when its
status code is not in the range 200..299
The string needs to be unmarshaled rather than returned directly otherwise
it will have an extra set of quotes plus it's safter to properly unmarshal
it.
Pointed out by @jrick.
The submitblock RPC returns a string containing an error if it failed for
any reason. This comment detects a non-null return from submitblock and
converts that string to an error which is returned.
This commit modifies the TLS setup to only override the RootCAs for the
TLS connection if certificates are specified. This allows the
Certificates parameter to be ommitted from the connection config to use
the system CAs.
If connecting to a bitcoin RPC server as an HTTP POST client, full
response objects rather than just the raw result bytes were being
passed to the specific result unmarshalers, causing unmarshal errors
for the incorrect JSON types. To fix this, first unmarshal the
response body into a rawResponse, and pass only the raw result bytes
(or an error) to the specific handlers.
This was reported by nskelsey on IRC.
ok @davecgh
There are certain RPCs where an address should be sent as the raw string
instead of the encoded bitcoin address since the RPC handler on the other
end expects "keys" instead of Bitcoin addresses.
For example, the multisignature RPCs addmultisigaddress and createmultisig
can work with pubkeys (compressed, uncompressed, or hybrid) as well as
Bitcoin addresses.
The original issue which prompted these changes was report by Paul Snow on
IRC.
Rather than using bytes.NewBuffer, which is a read/write entity
(io.ReadWriter), use bytes.NewReader which is only a read entitiy
(io.Reader). Benchmarking shows it's slightly faster and it's also
technically more accurate since it ensures the data is read-only.
The btcutil package recently exposed a WIF type to provide more
functionality and simplify working with WIF strings. This commit changes
the DumpPrivKey and ImportPrivKey RPCs to use the new WIF type.
There are several RPCs which accept a pointer to a hash, transaction,
block, etc. Previously not all RPCs handled being passed a nil pointer
consistently.
Closes#4.
This commit adds logic to track all registered notifications that have
been registered by the client in a notification state when the default
automatic reconnect is enabled.
The notification state is then used to reregister for all previously
registered notifications on reconnect. This allows the caller to
continue receiving notifications across reconnect cycles.
Also, since the new package exposes more connection related error
information, add a new ErrInvalidEndpoint which is returned if the
specified enpoint does not appear to be a valid websocket provider and
only return the ErrInvalidAuth error when HTTP authorization failure
status codes are detected.
Closes#1.
