nimbus-eth1/nimbus/db/kvt/kvt_debug.nim

178 lines
5.0 KiB
Nim
Raw Normal View History

Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
# nimbus-eth1
# Copyright (c) 2023-2024 Status Research & Development GmbH
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
# 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.
{.push raises: [].}
import
std/[algorithm, sequtils, strutils, tables],
eth/common,
results,
stew/byteutils,
./kvt_desc/desc_backend,
./kvt_init/[memory_db, memory_only, rocks_db],
"."/[kvt_desc, kvt_layers]
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
# ------------------------------------------------------------------------------
# Private functions
# ------------------------------------------------------------------------------
proc squeeze(s: string; hex = false; ignLen = false): string =
## For long strings print `begin..end` only
if hex:
let n = (s.len + 1) div 2
result = if s.len < 20: s else: s[0 .. 5] & ".." & s[s.len-8 .. ^1]
if not ignLen:
result &= "[" & (if 0 < n: "#" & $n else: "") & "]"
elif s.len <= 30:
result = s
else:
result = if (s.len and 1) == 0: s[0 ..< 8] else: "0" & s[0 ..< 7]
if not ignLen:
result &= "..(" & $s.len & ")"
result &= ".." & s[s.len-16 .. ^1]
#proc stripZeros(a: string): string =
# a.strip(leading=true, trailing=false, chars={'0'})
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
proc toPfx(indent: int; offset = 0): string =
if 0 < indent+offset: "\n" & " ".repeat(indent+offset) else: ""
func getOrVoid*(tab: Table[Blob,uint64]; w: Blob): uint64 =
tab.getOrDefault(w, 0u64)
func getOrVoid*(tab: Table[uint64,Blob]; w: uint64): Blob =
tab.getOrDefault(w, EmptyBlob)
func isValid*(id: uint64): bool =
0 < id
proc keyID(key: Blob; db = KvtDbRef(nil)): uint64 =
if key.len == 0:
return 0
elif db.isNil:
return high(uint64)
else:
let
ctr = db.getCentre
id = ctr.xMap.getOrVoid key
if id.isValid:
id
else:
# Save new ID
ctr.xIdGen.inc
ctr.xMap[key] = db.xIdGen
ctr.pAmx[db.xIdGen] = key
ctr.xIdGen
#proc keyBlob(id: uint64; db = KvtDbRef(nil)): Blob =
# if 0 < id and not db.isNil:
# result = db.getCentre.pAmx.getOrVoid id
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
proc ppID(id: uint64): string =
"$" & (if id == 0: "ø" else: $id)
proc ppKey(key: Blob; db = KvtDbRef(nil)): string =
if key.len == 0:
0.ppID
elif db.isNil:
key[0 .. 0].toHex & "-" &
key[1 ..< key.len].toHex.squeeze(hex=true,ignLen=(key.len==33))
else:
key.keyID(db).ppID
proc ppValue(data: Blob): string =
data.toHex.squeeze(hex=true)
proc ppTab(tab: Table[Blob,Blob]; db = KvtDbRef(nil); indent = 4): string =
result = "{"
if db.isNil:
let keys = tab.keys.toSeq.sorted
result &= keys.mapIt((it, tab.getOrVoid it))
.mapIt("(" & it[0].ppKey & "," & it[1].ppValue & ")")
.join(indent.toPfx(1))
else:
let keys = tab.keys.toSeq.mapIt(it.keyID db).sorted
result &= keys.mapIt((it, tab.getOrVoid db.pAmx.getOrVoid(it)))
.mapIt("(" & it[0].ppID & "," & it[1].ppValue & ")")
.join(indent.toPfx(1))
result &= "}"
proc ppMap(tab: Table[uint64,Blob]; indent = 4): string =
let keys = tab.keys.toSeq.sorted
"{" &
keys.mapIt((it, tab.getOrVoid it))
.mapIt("(" & it[0].ppID & "," & it[1].ppKey & ")")
.join(indent.toPfx(1)) &
"}"
proc ppBe[T](be: T; db: KvtDbRef; indent: int): string =
## Walk over backend table
let
pfx1 = indent.toPfx(1)
pfx2 = indent.toPfx(2)
pfx3 = indent.toPfx(3)
var
data = ""
n = 0
for (key,val) in be.walk:
if 0 < n: data &= pfx3
n.inc
data &= $n & "(" & key.ppKey(db) & "," & val.ppValue & ")"
var
spc = if 0 < n: pfx2 else: " "
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
"<" & $be.kind & ">" & pfx1 & "tab" & spc & "{" & data & "}"
proc ppLayer(layer: LayerRef; db: KvtDbRef; indent = 4): string =
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
let
tLen = layer.sTab.len
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
info = "tab(" & $tLen & ")"
pfx1 = indent.toPfx(1)
pfx2 = if 0 < tLen: indent.toPfx(2) else: " "
"<layer>" & pfx1 & info & pfx2 & layer.sTab.ppTab(db,indent+2)
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
# ------------------------------------------------------------------------------
# Public functions
# ------------------------------------------------------------------------------
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
proc pp*(
be: BackendRef;
db: KvtDbRef;
indent = 4;
): string =
case be.kind:
of BackendMemory:
result &= be.MemBackendRef.ppBe(db, indent)
of BackendRocksDB,BackendRdbTriggered:
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
result &= be.RdbBackendRef.ppBe(db, indent)
of BackendVoid:
result &= "<NoBackend>"
proc pp*(
db: KvtDbRef;
backendOk = false;
keysOk = false;
indent = 4;
): string =
let
pfx = indent.toPfx
pfx1 = indent.toPfx(1)
result = db.layersCc.ppLayer(db, indent=indent)
Core db and aristo updates for destructor and tx logic (#1894) * Disable `TransactionID` related functions from `state_db.nim` why: Functions `getCommittedStorage()` and `updateOriginalRoot()` from the `state_db` module are nowhere used. The emulation of a legacy `TransactionID` type functionality is administratively expensive to provide by `Aristo` (the legacy DB version is only partially implemented, anyway). As there is no other place where `TransactionID`s are used, they will not be provided by the `Aristo` variant of the `CoreDb`. For the legacy DB API, nothing will change. * Fix copyright headers in source code * Get rid of compiler warning * Update Aristo code, remove unused `merge()` variant, export `hashify()` why: Adapt to upcoming `CoreDb` wrapper * Remove synced tx feature from `Aristo` why: + This feature allowed to synchronise transaction methods like begin, commit, and rollback for a group of descriptors. + The feature is over engineered and not needed for `CoreDb`, neither is it complete (some convergence features missing.) * Add debugging helpers to `Kvt` also: Update database iterator, add count variable yield argument similar to `Aristo`. * Provide optional destructors for `CoreDb` API why; For the upcoming Aristo wrapper, this allows to control when certain smart destruction and update can take place. The auto destructor works fine in general when the storage/cache strategy is known and acceptable when creating descriptors. * Add update option for `CoreDb` API function `hash()` why; The hash function is typically used to get the state root of the MPT. Due to lazy hashing, this might be not available on the `Aristo` DB. So the `update` function asks for re-hashing the gurrent state changes if needed. * Update API tracking log mode: `info` => `debug * Use shared `Kvt` descriptor in new Ledger API why: No need to create a new descriptor all the time
2023-11-16 19:35:03 +00:00
if backendOk:
result &= pfx & db.backend.pp(db, indent=indent)
if keysOk:
result &= pfx & "<keys>" & pfx1 & db.getCentre.pAmx.ppMap(indent=indent+1)
# ------------------------------------------------------------------------------
# End
# ------------------------------------------------------------------------------