nimbus-eth1/nimbus/db/aristo/aristo_api.nim

943 lines
32 KiB
Nim
Raw Normal View History

# nimbus-eth1
# Copyright (c) 2024 Status Research & Development GmbH
# Licensed under either of
# * Apache License, version 2.0, ([LICENSE-APACHE](LICENSE-APACHE) or
# http://www.apache.org/licenses/LICENSE-2.0)
# * MIT license ([LICENSE-MIT](LICENSE-MIT) or
# http://opensource.org/licenses/MIT)
# at your option. This file may not be copied, modified, or distributed
# except according to those terms.
## Stackable API for `Aristo`
## ==========================
import
std/times,
eth/common,
results,
./aristo_desc/desc_backend,
./aristo_init/memory_db,
"."/[aristo_delete, aristo_desc, aristo_fetch, aristo_hashify,
aristo_init, aristo_merge, aristo_path, aristo_profile, aristo_tx]
export
AristoDbProfListRef
const
AutoValidateApiHooks = defined(release).not
## No validatinon needed for production suite.
AristoPersistentBackendOk = AutoValidateApiHooks # and false
## Set true for persistent backend profiling (which needs an extra
## link library.)
when AristoPersistentBackendOk:
import ./aristo_init/rocks_db
# Annotation helper(s)
{.pragma: noRaise, gcsafe, raises: [].}
type
AristoApiCommitFn* =
proc(tx: AristoTxRef;
): Result[void,AristoError]
{.noRaise.}
## Given a *top level* handle, this function accepts all database
## operations performed through this handle and merges it to the
## previous layer. The previous transaction is returned if there
## was any.
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiDeleteAccountRecordFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
): Result[void,AristoError]
{.noRaise.}
## Delete the account leaf entry addressed by the argument `path`. If
## this leaf entry referres to a storage tree, this one will be deleted
## as well.
AristoApiDeleteGenericDataFn* =
proc(db: AristoDbRef;
root: VertexID;
path: openArray[byte];
): Result[bool,AristoError]
{.noRaise.}
## Delete the leaf data entry addressed by the argument `path`. The MPT
## sub-tree the leaf data entry is subsumed under is passed as argument
## `root` which must be greater than `VertexID(1)` and smaller than
## `LEAST_FREE_VID`.
##
## The return value is `true` if the argument `path` deleted was the last
## one and the tree does not exist anymore.
AristoApiDeleteGenericTreeFn* =
proc(db: AristoDbRef;
root: VertexID;
): Result[void,AristoError]
{.noRaise.}
## Variant of `deleteGenericData()` for purging the whole MPT sub-tree.
AristoApiDeleteStorageDataFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
stoPath: openArray[byte];
): Result[bool,AristoError]
{.noRaise.}
## For a given account argument `accPath`, this function deletes the
## argument `stoPath` from the associated storage tree (if any, at all.)
## If the if the argument `stoPath` deleted was the last one on the
## storage tree, account leaf referred to by `accPath` will be updated
## so that it will not refer to a storage tree anymore. In the latter
## case only the function will return `true`.
AristoApiDeleteStorageTreeFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
): Result[void,AristoError]
{.noRaise.}
## Variant of `deleteStorageData()` for purging the whole storage tree
## associated to the account argument `accPath`.
AristoApiFetchLastSavedStateFn* =
proc(db: AristoDbRef
): Result[SavedState,AristoError]
{.noRaise.}
## The function returns the state of the last saved state. This is a
## Merkle hash tag for vertex with ID 1 and a bespoke `uint64` identifier
## (may be interpreted as block number.)
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiFetchAccountRecordFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
): Result[AristoAccount,AristoError]
{.noRaise.}
## Fetch an account record from the database indexed by `accPath`.
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiFetchAccountStateFn* =
proc(db: AristoDbRef;
): Result[Hash256,AristoError]
{.noRaise.}
## Fetch the Merkle hash of the account root.
AristoApiFetchGenericDataFn* =
proc(db: AristoDbRef;
root: VertexID;
path: openArray[byte];
): Result[Blob,AristoError]
{.noRaise.}
## For a generic sub-tree starting at `root`, fetch the data record
## indexed by `path`.
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiFetchGenericStateFn* =
proc(db: AristoDbRef;
root: VertexID;
): Result[Hash256,AristoError]
{.noRaise.}
## Fetch the Merkle hash of the argument `root`.
AristoApiFetchStorageDataFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
stoPath: openArray[byte];
): Result[Blob,AristoError]
{.noRaise.}
## For a storage tree related to account `accPath`, fetch the data
## record from the database indexed by `stoPath`.
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiFetchStorageStateFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
): Result[Hash256,AristoError]
{.noRaise.}
## Fetch the Merkle hash of the storage root related to `accPath`.
AristoApiFindTxFn* =
proc(db: AristoDbRef;
vid: VertexID;
key: HashKey;
): Result[int,AristoError]
{.noRaise.}
## Find the transaction where the vertex with ID `vid` exists and has
## the Merkle hash key `key`. If there is no transaction available,
## search in the filter and then in the backend.
##
## If the above procedure succeeds, an integer indicating the transaction
## level is returned:
##
## * `0` -- top level, current layer
## * `1`,`2`,`..` -- some transaction level further down the stack
## * `-1` -- the filter between transaction stack and database backend
## * `-2` -- the databse backend
##
## A successful return code might be used for the `forkTx()` call for
## creating a forked descriptor that provides the pair `(vid,key)`.
##
AristoApiFinishFn* =
proc(db: AristoDbRef;
eradicate = false;
) {.noRaise.}
## Backend destructor. The argument `eradicate` indicates that a full
## database deletion is requested. If set `false` the outcome might
## differ depending on the type of backend (e.g. the `BackendMemory`
## backend will always eradicate on close.)
##
## In case of distributed descriptors accessing the same backend, all
## distributed descriptors will be destroyed.
##
## This distructor may be used on already *destructed* descriptors.
AristoApiForgetFn* =
proc(db: AristoDbRef;
): Result[void,AristoError]
{.noRaise.}
## Destruct the non centre argument `db` descriptor (see comments on
## `reCentre()` for details.)
##
## A non centre descriptor should always be destructed after use (see
## also# comments on `fork()`.)
AristoApiForkTxFn* =
proc(db: AristoDbRef;
backLevel: int;
): Result[AristoDbRef,AristoError]
{.noRaise.}
## Fork a new descriptor obtained from parts of the argument database
## as described by arguments `db` and `backLevel`.
##
## If the argument `backLevel` is non-negative, the forked descriptor
## will provide the database view where the first `backLevel` transaction
## layers are stripped and the remaing layers are squashed into a single
## transaction.
##
## If `backLevel` is `-1`, a database descriptor with empty transaction
## layers will be provided where the `balancer` between database and
## transaction layers are kept in place.
##
## If `backLevel` is `-2`, a database descriptor with empty transaction
## layers will be provided without a `balancer`.
##
## The returned database descriptor will always have transaction level one.
## If there were no transactions that could be squashed, an empty
## transaction is added.
##
## Use `aristo_desc.forget()` to clean up this descriptor.
AristoApiHashifyFn* =
proc(db: AristoDbRef;
): Result[void,(VertexID,AristoError)]
{.noRaise.}
## Add keys to the `Patricia Trie` so that it becomes a `Merkle
## Patricia Tree`.
AristoApiHasPathAccountFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
): Result[bool,AristoError]
{.noRaise.}
## For an account record indexed by `accPath` query whether this record
## exists on the database.
AristoApiHasPathGenericFn* =
proc(db: AristoDbRef;
root: VertexID;
path: openArray[byte];
): Result[bool,AristoError]
{.noRaise.}
## For a generic sub-tree starting at `root` and indexed by `path`,
## mquery whether this record exists on the database.
AristoApiHasPathStorageFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
stoPath: openArray[byte];
): Result[bool,AristoError]
{.noRaise.}
## For a storage tree related to account `accPath`, query whether the
## data record indexed by `stoPath` exists on the database.
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiHasStorageDataFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
): Result[bool,AristoError]
{.noRaise.}
## For a storage tree related to account `accPath`, query whether there
## is a non-empty data storage area at all.
AristoApiIsTopFn* =
proc(tx: AristoTxRef;
): bool
{.noRaise.}
## Getter, returns `true` if the argument `tx` referes to the current
## top level transaction.
AristoApiLevelFn* =
proc(db: AristoDbRef;
): int
{.noRaise.}
## Getter, non-negative nesting level (i.e. number of pending
## transactions)
AristoApiNForkedFn* =
proc(db: AristoDbRef;
): int
{.noRaise.}
## Returns the number of non centre descriptors (see comments on
## `reCentre()` for details.) This function is a fast version of
## `db.forked.toSeq.len`.
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiMergeAccountRecordFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
accRec: AristoAccount;
): Result[bool,AristoError]
{.noRaise.}
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
## Merge the key-value-pair argument `(accKey,accRec)` as an account
## ledger value, i.e. the the sub-tree starting at `VertexID(1)`.
AristoApiMergeGenericDataFn* =
proc( db: AristoDbRef;
root: VertexID;
path: openArray[byte];
data: openArray[byte];
): Result[bool,AristoError]
{.noRaise.}
## Variant of `mergeXXX()` for generic sub-trees, i.e. for arguments
## `root` greater than `VertexID(1)` and smaller than `LEAST_FREE_VID`.
AristoApiMergeStorageDataFn* =
proc(db: AristoDbRef;
accPath: openArray[byte];
stoPath: openArray[byte];
stoData: openArray[byte];
): Result[void,AristoError]
{.noRaise.}
## Store the `stoData` data argument on the storage area addressed by
## `(accPath,stoPath)` where `accPath` is the account key (into the MPT)
## and `stoPath` is the slot path of the corresponding storage area.
AristoApiPathAsBlobFn* =
proc(tag: PathID;
): Blob
{.noRaise.}
## Converts the `tag` argument to a sequence of an even number of
## nibbles represented by a `Blob`. If the argument `tag` represents
## an odd number of nibbles, a zero nibble is appendend.
##
## This function is useful only if there is a tacit agreement that all
## paths used to index database leaf values can be represented as
## `Blob`, i.e. `PathID` type paths with an even number of nibbles.
AristoApiPersistFn* =
proc(db: AristoDbRef;
nxtSid = 0u64;
chunkedMpt = false;
): Result[void,AristoError]
{.noRaise.}
## Persistently store data onto backend database. If the system is
## running without a database backend, the function returns immediately
## with an error. The same happens if there is a pending transaction.
##
## The function merges all staged data from the top layer cache onto the
## backend stage area. After that, the top layer cache is cleared.
##
## Finally, the staged data are merged into the physical backend
## database and the staged data area is cleared.
##
## The argument `nxtSid` will be the ID for the next saved state record.
##
## Staging the top layer cache might fail with a partial MPT when it is
## set up from partial MPT chunks as it happens with `snap` sync
## processing. In this case, the `chunkedMpt` argument must be set
## `true` (see alse `fwdFilter()`.)
AristoApiReCentreFn* =
proc(db: AristoDbRef;
): Result[void,AristoError]
{.noRaise.}
## Re-focus the `db` argument descriptor so that it becomes the centre.
## Nothing is done if the `db` descriptor is the centre, already.
##
## With several descriptors accessing the same backend database there is
## a single one that has write permission for the backend (regardless
## whether there is a backend, at all.) The descriptor entity with write
## permission is called *the centre*.
##
## After invoking `reCentre()`, the argument database `db` can only be
## destructed by `finish()` which also destructs all other descriptors
## accessing the same backend database. Descriptors where `isCentre()`
## returns `false` must be single destructed with `forget()`.
AristoApiRollbackFn* =
proc(tx: AristoTxRef;
): Result[void,AristoError]
{.noRaise.}
## Given a *top level* handle, this function discards all database
## operations performed for this transactio. The previous transaction
## is returned if there was any.
AristoApiTxBeginFn* =
proc(db: AristoDbRef
): Result[AristoTxRef,AristoError]
{.noRaise.}
## Starts a new transaction.
##
## Example:
## ::
## proc doSomething(db: AristoDbRef) =
## let tx = db.begin
## defer: tx.rollback()
## ... continue using db ...
## tx.commit()
AristoApiTxTopFn* =
proc(db: AristoDbRef;
): Result[AristoTxRef,AristoError]
{.noRaise.}
## Getter, returns top level transaction if there is any.
AristoApiRef* = ref AristoApiObj
AristoApiObj* = object of RootObj
## Useful set of `Aristo` fuctions that can be filtered, stacked etc.
commit*: AristoApiCommitFn
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
deleteAccountRecord*: AristoApiDeleteAccountRecordFn
deleteGenericData*: AristoApiDeleteGenericDataFn
deleteGenericTree*: AristoApiDeleteGenericTreeFn
deleteStorageData*: AristoApiDeleteStorageDataFn
deleteStorageTree*: AristoApiDeleteStorageTreeFn
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
fetchLastSavedState*: AristoApiFetchLastSavedStateFn
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
fetchAccountRecord*: AristoApiFetchAccountRecordFn
fetchAccountState*: AristoApiFetchAccountStateFn
fetchGenericData*: AristoApiFetchGenericDataFn
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
fetchGenericState*: AristoApiFetchGenericStateFn
fetchStorageData*: AristoApiFetchStorageDataFn
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
fetchStorageState*: AristoApiFetchStorageStateFn
findTx*: AristoApiFindTxFn
finish*: AristoApiFinishFn
forget*: AristoApiForgetFn
forkTx*: AristoApiForkTxFn
hashify*: AristoApiHashifyFn
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
hasPathAccount*: AristoApiHasPathAccountFn
hasPathGeneric*: AristoApiHasPathGenericFn
hasPathStorage*: AristoApiHasPathStorageFn
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
hasStorageData*: AristoApiHasStorageDataFn
isTop*: AristoApiIsTopFn
level*: AristoApiLevelFn
nForked*: AristoApiNForkedFn
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
mergeAccountRecord*: AristoApiMergeAccountRecordFn
mergeGenericData*: AristoApiMergeGenericDataFn
mergeStorageData*: AristoApiMergeStorageDataFn
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
pathAsBlob*: AristoApiPathAsBlobFn
persist*: AristoApiPersistFn
reCentre*: AristoApiReCentreFn
rollback*: AristoApiRollbackFn
txBegin*: AristoApiTxBeginFn
txTop*: AristoApiTxTopFn
AristoApiProfNames* = enum
## Index/name mapping for profile slots
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfTotal = "total"
AristoApiProfCommitFn = "commit"
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfDeleteAccountRecordFn = "deleteAccountRecord"
AristoApiProfDeleteGenericDataFn = "deleteGnericData"
AristoApiProfDeleteGenericTreeFn = "deleteGnericTree"
AristoApiProfDeleteStorageDataFn = "deleteStorageData"
AristoApiProfDeleteStorageTreeFn = "deleteStorageTree"
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfFetchLastSavedStateFn = "fetchLastSavedState"
AristoApiProfFetchAccountRecordFn = "fetchAccountRecord"
AristoApiProfFetchAccountStateFn = "fetchAccountState"
AristoApiProfFetchGenericDataFn = "fetchGenericData"
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfFetchGenericStateFn = "fetchGenericState"
AristoApiProfFetchStorageDataFn = "fetchStorageData"
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfFetchStorageStateFn = "fetchStorageState"
AristoApiProfFindTxFn = "findTx"
AristoApiProfFinishFn = "finish"
AristoApiProfForgetFn = "forget"
AristoApiProfForkTxFn = "forkTx"
AristoApiProfHashifyFn = "hashify"
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfHasPathAccountFn = "hasPathAccount"
AristoApiProfHasPathGenericFn = "hasPathGeneric"
AristoApiProfHasPathStorageFn = "hasPathStorage"
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfHasStorageDataFn = "hasStorageData"
AristoApiProfIsTopFn = "isTop"
AristoApiProfLevelFn = "level"
AristoApiProfNForkedFn = "nForked"
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfMergeAccountRecordFn = "mergeAccountRecord"
AristoApiProfMergeGenericDataFn = "mergeGenericData"
AristoApiProfMergeStorageDataFn = "mergeStorageData"
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfPathAsBlobFn = "pathAsBlob"
AristoApiProfPersistFn = "persist"
AristoApiProfReCentreFn = "reCentre"
AristoApiProfRollbackFn = "rollback"
AristoApiProfTxBeginFn = "txBegin"
AristoApiProfTxTopFn = "txTop"
AristoApiProfBeGetVtxFn = "be/getVtx"
AristoApiProfBeGetKeyFn = "be/getKey"
AristoApiProfBeGetTuvFn = "be/getTuv"
AristoApiProfBeGetLstFn = "be/getLst"
AristoApiProfBePutVtxFn = "be/putVtx"
AristoApiProfBePutKeyFn = "be/putKey"
AristoApiProfBePutTuvFn = "be/putTuv"
AristoApiProfBePutLstFn = "be/putLst"
AristoApiProfBePutEndFn = "be/putEnd"
AristoApiProfRef* = ref object of AristoApiRef
## Profiling API extension of `AristoApiObj`
data*: AristoDbProfListRef
be*: BackendRef
# ------------------------------------------------------------------------------
# Private helpers
# ------------------------------------------------------------------------------
when AutoValidateApiHooks:
proc validate(api: AristoApiObj|AristoApiRef) =
doAssert not api.commit.isNil
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
doAssert not api.deleteAccountRecord.isNil
doAssert not api.deleteGenericData.isNil
doAssert not api.deleteGenericTree.isNil
doAssert not api.deleteStorageData.isNil
doAssert not api.deleteStorageTree.isNil
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
doAssert not api.fetchLastSavedState.isNil
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
doAssert not api.fetchAccountRecord.isNil
doAssert not api.fetchAccountState.isNil
doAssert not api.fetchGenericData.isNil
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
doAssert not api.fetchGenericState.isNil
doAssert not api.fetchStorageData.isNil
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
doAssert not api.fetchStorageState.isNil
doAssert not api.findTx.isNil
doAssert not api.finish.isNil
doAssert not api.forget.isNil
doAssert not api.forkTx.isNil
doAssert not api.hashify.isNil
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
doAssert not api.hasPathAccount.isNil
doAssert not api.hasPathGeneric.isNil
doAssert not api.hasPathStorage.isNil
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
doAssert not api.hasStorageData.isNil
doAssert not api.isTop.isNil
doAssert not api.level.isNil
doAssert not api.nForked.isNil
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
doAssert not api.mergeAccountRecord.isNil
doAssert not api.mergeGenericData.isNil
doAssert not api.mergeStorageData.isNil
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
doAssert not api.pathAsBlob.isNil
doAssert not api.persist.isNil
doAssert not api.reCentre.isNil
doAssert not api.rollback.isNil
doAssert not api.txBegin.isNil
doAssert not api.txTop.isNil
proc validate(prf: AristoApiProfRef) =
prf.AristoApiRef.validate
doAssert not prf.data.isNil
proc dup(be: BackendRef): BackendRef =
case be.kind:
of BackendMemory:
return MemBackendRef(be).dup
of BackendRocksDB, BackendRdbHosting:
when AristoPersistentBackendOk:
return RdbBackendRef(be).dup
of BackendVoid:
discard
# ------------------------------------------------------------------------------
# Public API constuctors
# ------------------------------------------------------------------------------
func init*(api: var AristoApiObj) =
## Initialise an `api` argument descriptor
##
when AutoValidateApiHooks:
api.reset
api.commit = commit
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
api.deleteAccountRecord = deleteAccountRecord
api.deleteGenericData = deleteGenericData
api.deleteGenericTree = deleteGenericTree
api.deleteStorageData = deleteStorageData
api.deleteStorageTree = deleteStorageTree
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
api.fetchLastSavedState = fetchLastSavedState
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
api.fetchAccountRecord = fetchAccountRecord
api.fetchAccountState = fetchAccountState
api.fetchGenericData = fetchGenericData
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
api.fetchGenericState = fetchGenericState
api.fetchStorageData = fetchStorageData
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
api.fetchStorageState = fetchStorageState
api.findTx = findTx
api.finish = finish
api.forget = forget
api.forkTx = forkTx
api.hashify = hashify
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
api.hasPathAccount = hasPathAccount
api.hasPathGeneric = hasPathGeneric
api.hasPathStorage = hasPathStorage
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
api.hasStorageData = hasStorageData
api.isTop = isTop
api.level = level
api.nForked = nForked
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
api.mergeAccountRecord = mergeAccountRecord
api.mergeGenericData = mergeGenericData
api.mergeStorageData = mergeStorageData
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
api.pathAsBlob = pathAsBlob
api.persist = persist
api.reCentre = reCentre
api.rollback = rollback
api.txBegin = txBegin
api.txTop = txTop
when AutoValidateApiHooks:
api.validate
func init*(T: type AristoApiRef): T =
new result
result[].init()
func dup*(api: AristoApiRef): AristoApiRef =
result = AristoApiRef(
commit: api.commit,
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
deleteAccountRecord: api.deleteAccountRecord,
deleteGenericData: api.deleteGenericData,
deleteGenericTree: api.deleteGenericTree,
deleteStorageData: api.deleteStorageData,
deleteStorageTree: api.deleteStorageTree,
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
fetchLastSavedState: api.fetchLastSavedState,
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
fetchAccountRecord: api.fetchAccountRecord,
fetchAccountState: api.fetchAccountState,
fetchGenericData: api.fetchGenericData,
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
fetchGenericState: api.fetchGenericState,
fetchStorageData: api.fetchStorageData,
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
fetchStorageState: api.fetchStorageState,
findTx: api.findTx,
finish: api.finish,
forget: api.forget,
forkTx: api.forkTx,
hashify: api.hashify,
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
hasPathAccount: api.hasPathAccount,
hasPathGeneric: api.hasPathGeneric,
hasPathStorage: api.hasPathStorage,
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
hasStorageData: api.hasStorageData,
isTop: api.isTop,
level: api.level,
nForked: api.nForked,
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
mergeAccountRecord: api.mergeAccountRecord,
mergeGenericData: api.mergeGenericData,
mergeStorageData: api.mergeStorageData,
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
pathAsBlob: api.pathAsBlob,
persist: api.persist,
reCentre: api.reCentre,
rollback: api.rollback,
txBegin: api.txBegin,
txTop: api.txTop)
when AutoValidateApiHooks:
api.validate
# ------------------------------------------------------------------------------
# Public profile API constuctor
# ------------------------------------------------------------------------------
func init*(
T: type AristoApiProfRef;
api: AristoApiRef;
be = BackendRef(nil);
): T =
## This constructor creates a profiling API descriptor to be derived from
## an initialised `api` argument descriptor. For profiling the DB backend,
## the field `.be` of the result descriptor must be assigned to the
## `.backend` field of the `AristoDbRef` descriptor.
##
## The argument desctiptors `api` and `be` will not be modified and can be
## used to restore the previous set up.
##
let
data = AristoDbProfListRef(
list: newSeq[AristoDbProfData](1 + high(AristoApiProfNames).ord))
profApi = T(data: data)
template profileRunner(n: AristoApiProfNames, code: untyped): untyped =
let start = getTime()
code
data.update(n.ord, getTime() - start)
profApi.commit =
proc(a: AristoTxRef): auto =
AristoApiProfCommitFn.profileRunner:
result = api.commit(a)
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
profApi.deleteAccountRecord =
proc(a: AristoDbRef; b: openArray[byte]): auto =
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfDeleteAccountRecordFn.profileRunner:
result = api.deleteAccountRecord(a, b)
profApi.deleteGenericData =
proc(a: AristoDbRef; b: VertexID; c: openArray[byte]): auto =
AristoApiProfDeleteGenericDataFn.profileRunner:
result = api.deleteGenericData(a, b, c)
profApi.deleteGenericTree =
proc(a: AristoDbRef; b: VertexID): auto =
AristoApiProfDeleteGenericTreeFn.profileRunner:
result = api.deleteGenericTree(a, b)
profApi.deleteStorageData =
proc(a: AristoDbRef; b, c: openArray[byte]): auto =
AristoApiProfDeleteStorageDataFn.profileRunner:
result = api.deleteStorageData(a, b, c)
profApi.deleteStorageTree =
proc(a: AristoDbRef; b: openArray[byte]): auto =
AristoApiProfDeleteStorageTreeFn.profileRunner:
result = api.deleteStorageTree(a, b)
profApi.fetchLastSavedState =
proc(a: AristoDbRef): auto =
AristoApiProfFetchLastSavedStateFn.profileRunner:
result = api.fetchLastSavedState(a)
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
profApi.fetchAccountRecord =
proc(a: AristoDbRef; b: openArray[byte]): auto =
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfFetchAccountRecordFn.profileRunner:
result = api.fetchAccountRecord(a, b)
profApi.fetchAccountState =
proc(a: AristoDbRef): auto =
AristoApiProfFetchAccountStateFn.profileRunner:
result = api.fetchAccountState(a)
profApi.fetchGenericData =
proc(a: AristoDbRef; b: VertexID; c: openArray[byte]): auto =
AristoApiProfFetchGenericDataFn.profileRunner:
result = api.fetchGenericData(a, b, c)
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
profApi.fetchGenericState =
proc(a: AristoDbRef; b: VertexID;): auto =
AristoApiProfFetchGenericStateFn.profileRunner:
result = api.fetchGenericState(a, b)
profApi.fetchStorageData =
proc(a: AristoDbRef; b, c: openArray[byte]): auto =
AristoApiProfFetchStorageDataFn.profileRunner:
result = api.fetchStorageData(a, b, c)
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
profApi.fetchStorageState =
proc(a: AristoDbRef; b: openArray[byte]): auto =
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfFetchStorageStateFn.profileRunner:
result = api.fetchStorageState(a, b)
profApi.findTx =
proc(a: AristoDbRef; b: VertexID; c: HashKey): auto =
AristoApiProfFindTxFn.profileRunner:
result = api.findTx(a, b, c)
profApi.finish =
proc(a: AristoDbRef; b = false) =
AristoApiProfFinishFn.profileRunner:
api.finish(a, b)
profApi.forget =
proc(a: AristoDbRef): auto =
AristoApiProfForgetFn.profileRunner:
result = api.forget(a)
profApi.forkTx =
proc(a: AristoDbRef; b: int): auto =
AristoApiProfForkTxFn.profileRunner:
result = api.forkTx(a, b)
profApi.hashify =
proc(a: AristoDbRef): auto =
AristoApiProfHashifyFn.profileRunner:
result = api.hashify(a)
profApi.hasPathAccount =
proc(a: AristoDbRef; b: openArray[byte]): auto =
AristoApiProfHasPathAccountFn.profileRunner:
result = api.hasPathAccount(a, b)
profApi.hasPathGeneric =
proc(a: AristoDbRef; b: VertexID; c: openArray[byte]): auto =
AristoApiProfHasPathGenericFn.profileRunner:
result = api.hasPathGeneric(a, b, c)
profApi.hasPathStorage =
proc(a: AristoDbRef; b, c: openArray[byte]): auto =
AristoApiProfHasPathStorageFn.profileRunner:
result = api.hasPathStorage(a, b, c)
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
profApi.hasStorageData =
proc(a: AristoDbRef; b: openArray[byte]): auto =
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfHasStorageDataFn.profileRunner:
result = api.hasStorageData(a, b)
profApi.isTop =
proc(a: AristoTxRef): auto =
AristoApiProfIsTopFn.profileRunner:
result = api.isTop(a)
profApi.level =
proc(a: AristoDbRef): auto =
AristoApiProfLevelFn.profileRunner:
result = api.level(a)
profApi.nForked =
proc(a: AristoDbRef): auto =
AristoApiProfNForkedFn.profileRunner:
result = api.nForked(a)
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
profApi.mergeAccountRecord =
proc(a: AristoDbRef; b: openArray[byte]; c: AristoAccount): auto =
Update storage tree admin (#2419) * Tighten `CoreDb` API for accounts why: Apart from cruft, the way to fetch the accounts state root via a `CoreDbColRef` record was unnecessarily complicated. * Extend `CoreDb` API for accounts to cover storage tries why: In future, this will make the notion of column objects obsolete. Storage trees will then be indexed by the account address rather than the vertex ID equivalent like a `CoreDbColRef`. * Apply new/extended accounts API to ledger and tests details: This makes the `distinct_ledger` module obsolete * Remove column object constructors why: They were needed as an abstraction of MPT sub-trees including storage trees. Now, storage trees are handled by the account (e.g. via address) they belong to and all other trees can be identified by a constant well known vertex ID. So there is no need for column objects anymore. Still there are some left-over column object methods wnich will be removed next. * Remove `serialise()` and `PayloadRef` from default Aristo API why: Not needed. `PayloadRef` was used for unstructured/unknown payload formats (account or blob) and `serialise()` was used for decodng `PayloadRef`. Now it is known in advance what the payload looks like. * Added query function `hasStorageData()` whether a storage area exists why: Useful for supporting `slotStateEmpty()` of the `CoreDb` API * In the `Ledger` replace `storage.stateEmpty()` by `slotStateEmpty()` * On Aristo, hide the storage root/vertex ID in the `PayloadRef` why: The storage vertex ID is fully controlled by Aristo while the `AristoAccount` object is controlled by the application. With the storage root part of the `AristoAccount` object, there was a useless administrative burden to keep that storage root field up to date. * Remove cruft, update comments etc. * Update changed MPT access paradigms why: Fixes verified proxy tests * Fluffy cosmetics
2024-06-27 09:01:26 +00:00
AristoApiProfMergeAccountRecordFn.profileRunner:
result = api.mergeAccountRecord(a, b, c)
profApi.mergeGenericData =
proc(a: AristoDbRef; b: VertexID, c, d: openArray[byte]): auto =
AristoApiProfMergeGenericDataFn.profileRunner:
result = api.mergeGenericData(a, b, c, d)
profApi.mergeStorageData =
proc(a: AristoDbRef; b, c, d: openArray[byte]): auto =
AristoApiProfMergeStorageDataFn.profileRunner:
result = api.mergeStorageData(a, b, c, d)
profApi.pathAsBlob =
proc(a: PathID): auto =
AristoApiProfPathAsBlobFn.profileRunner:
result = api.pathAsBlob(a)
profApi.persist =
proc(a: AristoDbRef; b = 0u64; c = false): auto =
AristoApiProfPersistFn.profileRunner:
result = api.persist(a, b, c)
profApi.reCentre =
proc(a: AristoDbRef): auto =
AristoApiProfReCentreFn.profileRunner:
result = api.reCentre(a)
profApi.rollback =
proc(a: AristoTxRef): auto =
AristoApiProfRollbackFn.profileRunner:
result = api.rollback(a)
profApi.txBegin =
proc(a: AristoDbRef): auto =
AristoApiProfTxBeginFn.profileRunner:
result = api.txBegin(a)
profApi.txTop =
proc(a: AristoDbRef): auto =
AristoApiProfTxTopFn.profileRunner:
result = api.txTop(a)
let beDup = be.dup()
if beDup.isNil:
profApi.be = be
else:
beDup.getVtxFn =
proc(a: VertexID): auto =
AristoApiProfBeGetVtxFn.profileRunner:
result = be.getVtxFn(a)
data.list[AristoApiProfBeGetVtxFn.ord].masked = true
beDup.getKeyFn =
proc(a: VertexID): auto =
AristoApiProfBeGetKeyFn.profileRunner:
result = be.getKeyFn(a)
data.list[AristoApiProfBeGetKeyFn.ord].masked = true
beDup.getTuvFn =
proc(): auto =
AristoApiProfBeGetTuvFn.profileRunner:
result = be.getTuvFn()
data.list[AristoApiProfBeGetTuvFn.ord].masked = true
beDup.getLstFn =
proc(): auto =
AristoApiProfBeGetLstFn.profileRunner:
result = be.getLstFn()
data.list[AristoApiProfBeGetLstFn.ord].masked = true
beDup.putVtxFn =
proc(a: PutHdlRef; b: VertexID, c: VertexRef) =
AristoApiProfBePutVtxFn.profileRunner:
be.putVtxFn(a, b, c)
data.list[AristoApiProfBePutVtxFn.ord].masked = true
beDup.putKeyFn =
proc(a: PutHdlRef; b: VertexID, c: HashKey) =
AristoApiProfBePutKeyFn.profileRunner:
be.putKeyFn(a, b, c)
data.list[AristoApiProfBePutKeyFn.ord].masked = true
beDup.putTuvFn =
proc(a: PutHdlRef; b: VertexID) =
AristoApiProfBePutTuvFn.profileRunner:
be.putTuvFn(a,b)
data.list[AristoApiProfBePutTuvFn.ord].masked = true
beDup.putLstFn =
proc(a: PutHdlRef; b: SavedState) =
AristoApiProfBePutLstFn.profileRunner:
be.putLstFn(a,b)
data.list[AristoApiProfBePutLstFn.ord].masked = true
beDup.putEndFn =
proc(a: PutHdlRef): auto =
AristoApiProfBePutEndFn.profileRunner:
result = be.putEndFn(a)
data.list[AristoApiProfBePutEndFn.ord].masked = true
profApi.be = beDup
when AutoValidateApiHooks:
profApi.validate
profApi
# ------------------------------------------------------------------------------
# End
# ------------------------------------------------------------------------------