Implement a safety check mechanism on object_database, based on the
safety_check_policy abstract interface. Create two standard
implementations of the safety_check_policy interface, one
(null_safety_check) which allows all modifications unconditionally, and
the other (database_lock_safety_check) which allows modifications only
when unlocked.
Integrate these safety checks into chain::database and plugins, so that
the consensus databases are locked at all times except during core
consensus code pathways. Also ensures that databases are re-locked when
calling code outside of consensus pathways from consensus pathways.
To make this work, it was necessary to move two objects from the
consensus object spaces to a new API object space. The
operation_history_object and account_transaction_history_object were
moved to the API object space, as they are not actually used by
consensus and are maintained by a plugin (which can no longer modify the
consensus object spaces, due to the safety checks).
Finally, add a mechanism to application and chain::database, which
allows the chain to start in "unit testing mode" and allows unchecked
actions upon the database within delimited scopes. This was necessary
because many tests edit the database directly to set up the environment
for their respective tests. This mode is activated by database_fixture so
tests can utilize it conveniently, but it is architecturally difficult to
enable this mode in production, i.e. from a plugin.
Pursuant to the requested safety checks on the database, to ensure that
plugin code (i.e., third party code) cannot modify the database, we
implement evaluator tagging so the chain can distinguish between
consensus evaluators and third party evaluators. Also, define a new kind
of evaluator base class, third_party_evaluator, so that fees are not
charged multiple times for operations with multiple evaluators.
Next step, implement the actual safety check mechanism on the database.
Create a new cli_wallet command, run_custom_operation, which makes it
convenient to run custom_operation transactions which invoke third party
contracts (i.e., dapps)
When using an external genesis file, it doesn't make sense to calculate
the chain ID as the literal text contents of the file, which are quite
volatile. Rather, it should be calculated based on the logical content
of the file. Serialize the genesis object to get a canonical
representation, and calculate the chain ID off of that.
The count_objects_in_space function, while providing the necessary
functionality, was clumsy and inconvenient, and was not idiomatic.
Replace it with inspect_all_indexes which resolves these shortcomings.
Add a way to create a secondary index with the space/type ID of the
object rather than the compile-time type, which may not always be
available.
Also, add a way to delete a secondary index (whether by compile-time
type or by object space/type ID).
Add the ability to dynamically unload an auxiliary evaluator (any
evaluator for a given operation type except the first one registered for
that operation type), and as an added convenience, add an idiomatically
consistent way to check if a given index is already registered or not.
In order to support dynamically linked nodes based on Graphene, add
support for building dynamic libraries for the core Graphene
modules: chain, db, protocol, net, and utilities.
This commit adds the ability to register multiple evaluator types for a
given operation type. This will be used to allow plug-ins to define their
own evaluators for various operation types (most notably, custom
operations) so that they may be used to carry app-specific data which can
be parsed by a plug-in.
This adds the most important updates to Graphene from BitShares. Most notably,
https://github.com/bitshares/bitshares-core/issues/1506
Second most notably, it updates Peerplays' FC to be in sync with BitShares FC.
This is a squash commit of several subcommits. The subcommit messages are
reproduced below:
Replace fc::uint128 with boost::multiprecision::uint128_t
replace smart_ref with shared_ptr
Fixes/Remove Unused
Remove NTP time
Remove old macro
This macro is now in FC, so no need to define it here anymore
Replaced fc::array with std::array
Separate exception declaration and implementation
Adapted to fc promise changes
Fixes
Add back in some of Peter's fixes that got lost in the cherry pick
_hash endianness fixes
Remove all uses of fc/smart_ref
It's gone, can't use it anymore
Replace improper static_variant operator overloads with comparators
Fixes
Remove boost::signals from build system; it's header-only so it's not
listed in cmake anymore.
Also remove some unused hashing code
Impl. pack/unpack functions for extension class
Ref #1506: Isolate chain/protocol to its own library
Ref #1506: Add object_downcast_t
Allows the more concise expression `object_downcast_t<xyz>` instead of
the old `typename object_downcast<xyz>::type`
Ref #1506: Move ID types from db to protocol
The ID types, object_id and object_id_type, were defined in the db
library, and the protocol library depends on db to get these types.
Technically, the ID types are defined by the protocol and used by the
database, and not vice versa. Therefore these types should be in the
protocol library, and db should depend on protocol to get them.
This commit makes it so.
Ref #1506: Isolate chain/protocol to its own library
Remove commented-out index code
Wrap overlength line
Remove unused key types
Probably fix Docker build
Fix build after rebase
Ref #1506/#1737: Some requested changes
Ref #1506/#1737: Macro-fy ID type definitions
Define macros to fully de-boilerplate ID type definitions.
Externalities:
- Rename transaction_object -> transaction_history_object
- Rename impl_asset_dynamic_data_type ->
impl_asset_dynamic_data_object_type
- Rename impl_asset_bitasset_data_type ->
impl_asset_bitasset_data_object_type
The first is to avoid a naming collision on transaction_id_type, and the
other two are to maintain consistency with the naming of the other
types.
Ref #1506/#1737: Fix clean_name()
Ref #1506/#1737: Oops
Fix .gitignore
Externalized serialization in protocol library
Fix compile sets
Delete a couple of ghost files that were in the tree but not part
of the project (I accidentally added them to CMakeLists while
merging, but they're broken and not part of the Peerplays code), and
add several files that got dropped from the build during merge.
General fixes
Fix warnings, build issues, unused code, etc.
Fix#1772 by decprecating cli_wallet -H
More fixes
Fix errors and warnings and generally coax it to build
Fix test
I'm pretty sure this didn't break from what I did... But I can't build
the original code, so I can't tell. Anyways, this one now passes...
Others still fail...
Small fix
Fix crash in auth checks
Final fixes
Last round of fixes following the rebase to Beatrice
Rename project in CMakeLists.txt
The CMakeLists.txt declared this project as BitShares and not Peerplays,
which makes it confusing in IDEs. Rename it to be clear which project is
open.
Resolve#374
Replace all object refs in macros with IDs, and fix affected tests to look
up objects by ID rather than using invalidated refs.
A full audit of all tests should be performed to eliminate any further
usage of invalidated object references.
Resolve#373: Add object notifiers
Various fixes
Fixes to various issues, primarily reflections, that cropped up
during merge conflict resolution
Fix startup bug in Bookie plugin
Bookie plugin was preventing the node from starting up because it
registered its secondary indexes to create objects in its own primary
indexes to track objects being created in other primary indexes, and did
so during its `initialize()` step, which is to say, before the database
was loaded from disk at startup. This caused the secondary indexes to
create tracker objects when the observed indexes were loading objects
from disk. This then caused a failure when these tracker indexes were
later loaded from disk, and the first object IDs collided.
This is fixed by refraining from defining secondary indexes until the
`startup()` stage rather than the `initialize()` stage. Primary indexes
are registered in `initialize()`, secondary indexes are registered in
`startup()`.
This also involved adding a new method, "add_secondary_index()", to
`object_database`, as before there was no way to do this because you
couldn't get a non-const index from a non-const database.
I have no idea how this was working before I got here...
Fix egenesis install
Fixes after updates
Rebase on updated develop branch and fix conflicts
Replace all object refs in macros with IDs, and fix affected tests to look
up objects by ID rather than using invalidated refs.
A full audit of all tests should be performed to eliminate any further
usage of invalidated object references.
The CMakeLists.txt declared this project as BitShares and not Peerplays,
which makes it confusing in IDEs. Rename it to be clear which project is
open.
I'm pretty sure this didn't break from what I did... But I can't build
the original code, so I can't tell. Anyways, this one now passes...
Others still fail...
Delete a couple of ghost files that were in the tree but not part
of the project (I accidentally added them to CMakeLists while
merging, but they're broken and not part of the Peerplays code), and
add several files that got dropped from the build during merge.
Define macros to fully de-boilerplate ID type definitions.
Externalities:
- Rename transaction_object -> transaction_history_object
- Rename impl_asset_dynamic_data_type ->
impl_asset_dynamic_data_object_type
- Rename impl_asset_bitasset_data_type ->
impl_asset_bitasset_data_object_type
The first is to avoid a naming collision on transaction_id_type, and the
other two are to maintain consistency with the naming of the other
types.
The ID types, object_id and object_id_type, were defined in the db
library, and the protocol library depends on db to get these types.
Technically, the ID types are defined by the protocol and used by the
database, and not vice versa. Therefore these types should be in the
protocol library, and db should depend on protocol to get them.
This commit makes it so.
