This commit implements a reimagining of the way the btcjson package
functions based upon how the project has evolved and lessons learned while
using it since it was first written. It therefore contains significant
changes to the API. For now, it has been implemented in a v2 subdirectory
to prevent breaking existing callers, but the ultimate goal is to update
all callers to use the new version and then to replace the old API with
the new one.
This also removes the need for the btcws completely since those commands
have been rolled in.
The following is an overview of the changes and some reasoning behind why
they were made:
- The infrastructure has been completely changed to be reflection based instead
of requiring thousands and thousands of lines of manual, and therefore error
prone, marshal/unmarshal code
- This makes it much easier to add new commands without making marshalling
mistakes since it is simply a struct definition and a call to register that
new struct (plus a trivial New<foo>Cmd function and tests, of course)
- It also makes it much easier to gain a lot of information from simply
looking at the struct definition which was previously not possible
such as the order of the parameters, which parameters are required
versus optional, and what the default values for optional parameters
are
- Each command now has usage flags associated with them that can be
queried which are intended to allow classification of the commands such
as for chain server and wallet server and websocket-only
- The help infrastructure has been completely redone to provide automatic
generation with caller provided description map and result types. This
is in contrast to the previous method of providing the help directly
which meant it would end up in the binary of anything that imported the
package
- Many of the structs have been renamed to use the terminology from the
JSON-RPC
specification:
- RawCmd/Message is now only a single struct named Request to reflect the fact
it is a JSON-RPC request
- Error is now called RPCError to reflect the fact it is specifically an RPC
error as opposed to many of the other errors that are possible
- All RPC error codes except the standard JSON-RPC 2.0 errors have been
converted from full structs to only codes since an audit of the codebase
has shown that the messages are overridden the vast majority of the time
with specifics (as they should be) and removing them also avoids the
temptation to return non-specific, and therefore not as helpful, error
messages
- There is now an Error which provides a type assertable error with
error codes so callers can better ascertain failure reasons
programatically
- The ID is no longer a part of the command and is instead specified at the time
the command is marshalled into a JSON-RPC request. This aligns better with
the way JSON-RPC functions since it is the caller who manages the ID that is
sent with any given _request_, not the package
- All <Foo>Cmd structs now treat non-pointers as required fields and pointers as
optional fields
- All New<Foo>Cmd functions now accept the exact number of parameters, with
pointers to the appropriate type for optional parameters
- This is preferrable to the old vararg syntax since it means the code will
fail to compile if the optional arguments are changed now which helps
prevent errors creep in over time from missed modifications to optional args
- All of the connection related code has been completely eliminated since this
package is not intended to used a client, rather it is intended to provide
the infrastructure needed to marshal/unmarshal Bitcoin-specific JSON-RPC
requests and replies from static types
- The btcrpcclient package provides a robust client with connection management
and higher-level types that in turn uses the primitives provided by this
package
- Even if the caller does not wish to use btcrpcclient for some reason, they
should still be responsible for connection management since they might want
to use any number of connection features which the package would not
necessarily support
- Synced a few of the commands that have added new optional fields that
have since been added to Bitcoin Core
- Includes all of the commands and notifications that were previously in
btcws
- Now provides 100% test coverage with parallel tests
- The code is completely golint and go vet clean
This has the side effect of addressing nearly everything in, and therefore
closes#26.
Also fixes#18 and closes#19.
* Previously both searchrawtransaction and getrawtransaction would omit
the 'Confirmations' field by default if the transaction was found in
the mempool.
* By removing `omitempty`, a mempool transaction will now have
'Confirmations:0'
Our policy is to only ensure the code compiles for the latest Go release
minus one version. Since Go 1.4 has now been released, this commit
updates TravisCI to remove Go 1.2 from the build matrix and add Go 1.3.
Also, go vet changed locations in between Go 1.3 and Go 1.4, so a check
has been added to download the appropriate version based on which version
of Go is running the integration tests.
Finally, the previous removed Go tip entry in the build matrix has been
reintroduced since it will now work again.
Since the result might not be set when the response does not contain one,
such as in the case when the command is not implemented by the server,
don't attempt to modify it until ensuring it exists.
This commit adds a new result type for the getblocktemplate RPC which
provides the fields as defined by BIP0022. The extension fields defined
by BIP0023 are not included yet.
ok @jcvernaleo
BIP0022 defines optional fields in a getblocktemplate request for long
polling and template tweaking.
In addition, for template tweaking, there are two fields, sigoplimit and
sizelimit, which are atypical in that they are allowed to be either
booleans or numeric. This requires the fields to be represented as
interfaces which means any code making use of the struct will need to use
type assertions or a type switch.
This commit updates GetBlockTemplateCmd accordingly.
ok @jcvernaleo
This reasons for this change follow:
- All instances of the same key should be consistent amongst the commands
and returns
- Output indices can't be negative, so rather than adding more code to
check for a negative after unmarshal, just allow the unmarshal to weed
out negatives
ok @jcvernaleo
