* trie: store nodes as pointers
This avoids memory copies when unwrapping node interface values.
name old time/op new time/op delta
Get 388ns ± 8% 215ns ± 2% -44.56% (p=0.000 n=15+15)
GetDB 363ns ± 3% 202ns ± 2% -44.21% (p=0.000 n=15+15)
UpdateBE 1.57µs ± 2% 1.29µs ± 3% -17.80% (p=0.000 n=13+15)
UpdateLE 1.92µs ± 2% 1.61µs ± 2% -16.25% (p=0.000 n=14+14)
HashBE 2.16µs ± 6% 2.18µs ± 6% ~ (p=0.436 n=15+15)
HashLE 7.43µs ± 3% 7.21µs ± 3% -2.96% (p=0.000 n=15+13)
* trie: close temporary databases in GetDB benchmark
* trie: don't keep []byte from DB load around
Nodes decoded from a DB load kept hashes and values as sub-slices of
the DB value. This can be a problem because loading from leveldb often
returns []byte with a cap that's larger than necessary, increasing
memory usage.
* trie: unload old cached nodes
* trie, core/state: use cache unloading for account trie
* trie: use explicit private flags (fixes Go 1.5 reflection issue).
* trie: fixup cachegen overflow at request of nick
* core/state: rename journal size constant
This commit replaces the deep-copy based state revert mechanism with a
linear complexity journal. This commit also hides several internal
StateDB methods to limit the number of ways in which calling code can
use the journal incorrectly.
As usual consultation and bug fixes to the initial implementation were
provided by @karalabe, @obscuren and @Arachnid. Thank you!
In this commit, contract bindings and their backend start using the
Ethereum Go API interfaces offered by ethclient. This makes ethclient a
suitable replacement for the old remote backend and gets us one step
closer to the final stable Go API that is planned for go-ethereum 1.5.
The changes in detail:
* Pending state is optional for read only contract bindings.
BoundContract attempts to discover the Pending* methods via an
interface assertion. There are a couple of advantages to this:
ContractCaller is just two methods and can be implemented on top of
pretty much anything that provides Ethereum data. Since the backend
interfaces are now disjoint, ContractBackend can simply be declared as
a union of the reader and writer side.
* Caching of HasCode is removed. The caching could go wrong in case of
chain reorganisations and removing it simplifies the code a lot.
We'll figure out a performant way of providing ErrNoCode before the
1.5 release.
* BoundContract now ensures that the backend receives a non-nil context
with every call.
ValidateFields was introduced before the rlp decoder disallowed nil
values. Decoding RLP will never return nil values, there is no need
to check for them.
This CL makes several refactors:
- Define a Tracer interface, implementing the `CaptureState` method
- Add the VM environment as the first argument of
`Tracer.CaptureState`
- Rename existing functionality `StructLogger` an make it an
implementation of `Tracer`
- Delete `StructLogCollector` and make `StructLogger` collect the logs
directly
- Change all callers to use the new `StructLogger` where necessary and
extract logs from that.
- Deletes the apparently obsolete and likely nonfunctional 'TraceCall'
from the eth API.
Callers that only wish accumulated logs can use the `StructLogger`
implementation straightforwardly. Callers that wish to efficiently
capture VM traces and operate on them without excessive copying can now
implement the `Tracer` interface to receive VM state at each step and
do with it as they wish.
This CL also removes the accumulation of logs from the vm.Environment;
this was necessary as part of the refactor, but also simplifies it by
removing a responsibility that doesn't directly belong to the
Environment.
Support for legacy version 0.9.x is gone. The compiler version is no
longer cached. Compilation results (and the version) are read directly
from stdout using the --combined-json flag. As a workaround for
ethereum/solidity#651, source code is written to a temporary file before
compilation.
Integration of solc in package ethapi and cmd/abigen is now much simpler
because the compiler wrapper is no longer passed around as a pointer.
Fixes#2806, accidentally
The account manager was previously created by packge cmd/utils as part
of flag processing and then passed down into eth.Ethereum through its
config struct. Since we are starting to create nodes which do not have
eth.Ethereum as a registered service, the code was rearranged to
register the account manager as its own service. Making it a service is
ugly though and it doesn't really fix the root cause: creating nodes
without eth.Ethereum requires duplicating lots of code.
This commit splits utils.MakeSystemNode into three functions, making
creation of other node/service configurations easier. It also moves the
account manager into Node so it can be used by those configurations
without requiring package eth.
Consensus rules dictate that objects can only be removed during the
finalisation of the transaction (i.e. after all calls have finished).
Thus calling a suicided contract twice from the same transaction:
A->B(S)->ret(A)->B(S) results in 2 suicides. Calling the suicided
object twice from two transactions: A->B(S), A->B, results in only one
suicide and a call to an empty object.
Our current debug tracing functionality replays all transaction that
were executed prior to the targetted transaction in order to provide
the user with an accurate trace.
As a side effect to calling StateDB.IntermediateRoot it also deletes any
suicides objects. Our tracing code never calls this function because it
isn't interested in the intermediate root. Becasue of this it caused a
bug in the tracing code where transactions that were send to priviously
deleted objects resulted in two suicides rather than one suicide and a
call to an empty object.
Fixes#2542
We used to have reporting of bad blocks, but it was disabled
before the Frontier release. We need it back because users
are usually unable to provide the full RLP data of a bad
block when it occurs.
A shortcoming of this particular implementation is that the
origin peer is not tracked for blocks received during eth/63
sync. No origin peer info is still better than no report at
all though.
Shutting down geth prints hundreds of annoying error messages in some
cases. The errors appear because the Stop method of eth.ProtocolManager,
miner.Miner and core.TxPool is asynchronous. Left over peer sessions
generate events which are processed after Stop even though the database
has already been closed.
The fix is to make Stop synchronous using sync.WaitGroup.
For eth.ProtocolManager, in order to make use of WaitGroup safe, we need
a way to stop new peer sessions from being added while waiting on the
WaitGroup. The eth protocol Run function now selects on a signaling
channel and adds to the WaitGroup only if ProtocolManager is not
shutting down.
For miner.worker and core.TxPool the number of goroutines is static,
WaitGroup can be used in the usual way without additional
synchronisation.
Context keys must have a unique type in order to prevent
any unintented clashes. The code used int(1) as key.
Fix it by implementing the pattern recommended by package context.
- Manager.Accounts no longer returns an error.
- Manager methods take Account instead of common.Address.
- All uses of Account with unkeyed fields are converted.
The account management API was originally implemented as a thin layer
around crypto.KeyStore, on the grounds that several kinds of key stores
would be implemented later on. It turns out that this won't happen so
KeyStore is a superflous abstraction.
In this commit crypto.KeyStore and everything related to it moves to
package accounts and is unexported.
rpc: be less restrictive on the request id
rpc: improved documentation
console: upgrade web3.js to version 0.16.0
rpc: cache http connections
rpc: rename wsDomains parameter to wsOrigins
Exposes some core methods to transition and compute new state
information and adds an additional return value to the transition db
method to fetch required gas for that particular message (excluding gas
refunds from any SSTORE[X] = 0 and SUICIDE.
Fixes#2395
The chain maker and the simulated backend now run with a homestead phase
beginning at block 0 (i.e. there's no frontier).
This commit also fixes up #2388
Added chain configuration options and write out during genesis database
insertion. If no "config" was found, nothing is written to the database.
Configurations are written on a per genesis base. This means
that any chain (which is identified by it's genesis hash) can have their
own chain settings.
When a block is queried for retrieval we should add a check whether the
block falls within the frontier rules. If we'd always use `From`
retrieving transaction might fail. This PR temporarily changes
everything to `FromFrontier` (safe!).
* change gas cost for contract creating txs
* invalidate signature with s value greater than secp256k1 N / 2
* OOG contract creation if not enough gas to store code
* new difficulty adjustment algorithm
* new DELEGATECALL op code
Pending logs are now filterable through the Go API. Filter API changed
such that each filter type has it's own bucket and adding filter
explicitly requires you specify the bucket to put it in.
This PR fixes a regression of the RPC where the default gas price that
was being used for transaction wasn't properly using the GPO. This PR
adds the GPO back to suggest gas prices rather than the hardcoded
default of 10000000000000.
Closes#2194
Out of Bound log events are events that were removed due to a fork. When
logs are received the filtering mechanism should check for the `removed`
field on the json structure.
The test chain generated by makeChainFork included invalid uncle
headers, crashing the generator during the state commit.
The headers were invalid because they used the iteration counter as the
block number, even though makeChainFork uses a block with number > 0 as
the parent. Fix this by introducing BlockGen.Number, which allows
accessing the actual number of the block being generated.
State and receipt deliveries from a previous eth/62+ sync can hang if
the downloader has moved on to syncing with eth/61. Fix this by also
draining the eth/63 channels while waiting for eth/61 data.
A nicer solution would be to take care of the channels in a central
place, but that would involve a major rewrite.
Unexpected deliveries could block indefinitely if they arrived at the
right time. The fix is to ensure that the cancellation channel is
always closed when the sync ends, unblocking any deliveries. Also remove
the atomic check for whether a sync is currently running because it
doesn't help and can be misleading.
Cancelling always seems to break the tests though. The downloader
spawned d.process whenever new data arrived, making it somewhat hard to
track when block processing was actually done. Fix this by running
d.process in a dedicated goroutine that is tied to the lifecycle of the
sync. d.process gets notified of new work by the queue instead of being
invoked all the time. This removes a ton of weird workaround code,
including a hairy use of atomic CAS.
This removes the burden on a single object to take care of all
validation and state processing. Now instead the validation is done by
the `core.BlockValidator` (`types.Validator`) that takes care of both
header and uncle validation through the `ValidateBlock` method and state
validation through the `ValidateState` method. The state processing is
done by a new object `core.StateProcessor` (`types.Processor`) and
accepts a new state as input and uses that to process the given block's
transactions (and uncles for rewords) to calculate the state root for
the next block (P_n + 1).
There are a bunch of changes required to make this work:
- in miner: allow unregistering agents, fix RemoteAgent.Stop
- in eth/filters: make FilterSystem.Stop not crash
- in rpc/comms: move listen loop to platform-independent code
Fixes#1930. I ran the shell loop there for a few minutes and didn't see
any changes in the memory profile.
XEth.gpo was being initialized as needed. WithState copies the XEth
struct including the gpo field. If gpo was nil at the time of the copy
and Call or Transact were invoked on it, an additional GPO listenLoop
would be spawned.
Move the lazy initialization to GasPriceOracle instead so the same GPO
instance is shared among all created XEths.
Fixes#1317
Might help with #1930
* xeth, rpc: implement eth_getNatSpec for tx confirmations
* rename silly docserver -> httpclient
* eth/backend: httpclient now accessible via eth.Ethereum init-d via config.DocRoot
* cmd: introduce separate CLI flag for DocRoot (defaults to homedir)
* common/path: delete unused assetpath func, separate HomeDir func
* lines with leading space are ommitted from history
* exit processed even with whitespace around
* all whitespace lines (not only empty ones) are ignored
add 7 missing commands to admin api autocomplete
registrar: methods now return proper error if reg addresses are not set. fixes#1457
rpc/console: fix personal.newAccount() regression. Now all comms accept interactive password
registrar: add registrar tests for errors
crypto: catch AES decryption error on presale wallet import + fix error msg format. fixes#1580
CLI: improve error message when starting a second instance of geth. fixes#1564
cli/accounts: unlock multiple accounts. fixes#1785
* make unlocking multiple accounts work with inline <() fd
* passwdfile now correctly read only once
* improve logs
* fix CLI help text for unlocking
fix regression with docRoot / admin API
* docRoot/jspath passed to rpc/api ParseApis, which passes onto adminApi
* docRoot field for JS console in order to pass when RPC is (re)started
* improve flag desc for jspath
common/docserver: catch http errors from response
fix rpc/api tests
common/natspec: fix end to end test (skipped because takes 8s)
registrar: fix major regression:
* deploy registrars on frontier
* register HashsReg and UrlHint in GlobalRegistrar.
* set all 3 contract addresses in code
* zero out addresses first in tests
Log filtering is now using a MIPmap like approach where addresses of
logs are added to a mapped bloom bin. The current levels for the MIP are
in ranges of 1.000.000, 500.000, 100.000, 50.000, 1.000. Logs are
therefor filtered in batches of 1.000.
* Moved `vm.Transfer` to `core` package and changed execution to call
`env.Transfer` instead of `core.Transfer` directly.
* core/vm: byte code VM moved to jump table instead of switch
* Moved `vm.Transfer` to `core` package and changed execution to call
`env.Transfer` instead of `core.Transfer` directly.
* Byte code VM now shares the same code as the JITVM
* Renamed Context to Contract
* Changed initialiser of state transition & unexported methods
* Removed the Execution object and refactor `Call`, `CallCode` &
`Create` in to their own functions instead of being methods.
* Removed the hard dep on the state for the VM. The VM now
depends on a Database interface returned by the environment. In the
process the core now depends less on the statedb by usage of the env
* Moved `Log` from package `core/state` to package `core/vm`.
Moved the filtering system from `event` to `eth/filters` package and
removed the `core.Filter` object. The `filters.Filter` object now
requires a `common.Database` rather than a `eth.Backend` and invokes the
`core.GetBlockByX` directly rather than thru a "manager".
Added a `Difference` method to `types.Transactions` which sets the
receiver to the difference of a to b (NOTE: not a **and** b).
Transaction pool subscribes to RemovedTransactionEvent adding back to
those potential missing from the chain.
When a chain re-org occurs remove any transactions that were removed
from the canonical chain during the re-org as well as the receipts that
were generated in the process.
Closes#1746
Dev mode enabled some debugging flags such as:
* VM debugging mode
* Simpler proof of work
* Whisper enabled by default
* Datadir to a tmp datadir
* Maxpeers set to 0
* Gas price of 0
* Random listen port
Fixed database merge strategy to use the correct database. Due to a copy
paste fail when doing type evaluation the same database was being
iterated (chain), all others were ignored.
Removed state prefixing because {H(code): code} is stored in the same
database as the rest of the state.
* resolver -> common/registrar
* global registrar name registry interface
* add Call to resolver backend interface
* the hashReg and UrlHing contracts now initialised from global registry
* initialization of contracts uniform
* improve errors and more econsistent method names
* common/registrar/ethreg: versioned registrar
* integrate new naming and registrar in natspec
* js console api: setGlobalRegistrar, setHashReg, setUrlHint
* js test TestContract uses mining - tests fixed all pass
* eth/backend: allow PoW test mode (small ethash DAG)
* console jsre refers to resolver.abi/addr,
* cmd/geth/contracts.go moved to common/registrar
* Miners do now verify their own header, not their state.
* Changed old putTx and putReceipts to be exported
* Moved writing of transactions and receipts out of the block processer
in to the chain manager. Closes#1386
* Miner post ChainHeadEvent & ChainEvent. Closes#1388
* multiple passwords allowed in password file
* split on "\n", sideeffect: chop trailing slashes. fixes common mistake <(echo 'pass')
* remove accounts.Primary method
* do not fall back to primary account for mining
TestMadeupParentBlockChainAttack has been deleted because it was too
hard to port and the attack that it checks the prevention of is being
averted in a different way (through a protocol change).
Nodes that are out of sync will queue many transactions, which causes
the initial transactions message to grow very large. Larger transactions
messages can make communication impossible if the message is too big to
send. Big transactions messages also exhaust egress bandwidth, which
degrades other peer connections.
The new approach to combat these issues is to send transactions in
smaller batches. This commit introduces a new goroutine that handles
delivery of all initial transaction transfers. Size-limited packs of
transactions are sent to one peer at a time, conserving precious egress
bandwidth.
You can set the nonce of the block with `--genesisnonce`. When the
genesis nonce changes and it doesn't match with the first block in your
database it will fail. A new `datadir` must be given if the nonce of the
genesis block changes.
