[pallet_balances] Align FreezeIdentifier, RuntimeFreezeReason and MaxFreezes
#2997
Assignees
Comments
github-merge-queue bot
pushed a commit
that referenced
this issue
Jan 31, 2024
… removed `pallet_balances`'s `MaxHolds` config (#2657) I started this investigation/issue based on @liamaharon question [here](#1801 (comment)). ## Problem The `pallet_balances` integrity test should correctly detect that the runtime has correct distinct `HoldReasons` variant count. I assume the same situation exists for RuntimeFreezeReason. It is not a critical problem, if we set `MaxHolds` with a sufficiently large value, everything should be ok. However, in this case, the integrity_test check becomes less useful. **Situation for "any" runtime:** - `HoldReason` enums from different pallets: ```rust /// from pallet_nis #[pallet::composite_enum] pub enum HoldReason { NftReceipt, } /// from pallet_preimage #[pallet::composite_enum] pub enum HoldReason { Preimage, } // from pallet_state-trie-migration #[pallet::composite_enum] pub enum HoldReason { SlashForContinueMigrate, SlashForMigrateCustomTop, SlashForMigrateCustomChild, } ``` - generated `RuntimeHoldReason` enum looks like: ```rust pub enum RuntimeHoldReason { #[codec(index = 32u8)] Preimage(pallet_preimage::HoldReason), #[codec(index = 38u8)] Nis(pallet_nis::HoldReason), #[codec(index = 42u8)] StateTrieMigration(pallet_state_trie_migration::HoldReason), } ``` - composite enum `RuntimeHoldReason` variant count is detected as `3` - we set `type MaxHolds = ConstU32<3>` - `pallet_balances::integrity_test` is ok with `3`(at least 3) However, the real problem can occur in a live runtime where some functionality might stop working. This is due to a total of 5 distinct hold reasons (for pallets with multi-instance support, it is even more), and not all of them can be used because of an incorrect `MaxHolds`, which is deemed acceptable according to the `integrity_test`: ``` // pseudo-code - if we try to call all of these: T::Currency::hold(&pallet_nis::HoldReason::NftReceipt.into(), &nft_owner, deposit)?; T::Currency::hold(&pallet_preimage::HoldReason::Preimage.into(), &nft_owner, deposit)?; T::Currency::hold(&pallet_state_trie_migration::HoldReason::SlashForContinueMigrate.into(), &nft_owner, deposit)?; // With `type MaxHolds = ConstU32<3>` these two will fail T::Currency::hold(&pallet_state_trie_migration::HoldReason::SlashForMigrateCustomTop.into(), &nft_owner, deposit)?; T::Currency::hold(&pallet_state_trie_migration::HoldReason::SlashForMigrateCustomChild.into(), &nft_owner, deposit)?; ``` ## Solutions A macro `#[pallet::*]` expansion is extended of `VariantCount` implementation for the `#[pallet::composite_enum]` enum type. This expansion generates the `VariantCount` implementation for pallets' `HoldReason`, `FreezeReason`, `LockId`, and `SlashReason`. Enum variants must be plain enum values without fields to ensure a deterministic count. The composite runtime enum, `RuntimeHoldReason` and `RuntimeFreezeReason`, now sets `VariantCount::VARIANT_COUNT` as the sum of pallets' enum `VariantCount::VARIANT_COUNT`: ```rust #[frame_support::pallet(dev_mode)] mod module_single_instance { #[pallet::composite_enum] pub enum HoldReason { ModuleSingleInstanceReason1, ModuleSingleInstanceReason2, } ... } #[frame_support::pallet(dev_mode)] mod module_multi_instance { #[pallet::composite_enum] pub enum HoldReason<I: 'static = ()> { ModuleMultiInstanceReason1, ModuleMultiInstanceReason2, ModuleMultiInstanceReason3, } ... } impl self::sp_api_hidden_includes_construct_runtime::hidden_include::traits::VariantCount for RuntimeHoldReason { const VARIANT_COUNT: u32 = 0 + module_single_instance::HoldReason::VARIANT_COUNT + module_multi_instance::HoldReason::<module_multi_instance::Instance1>::VARIANT_COUNT + module_multi_instance::HoldReason::<module_multi_instance::Instance2>::VARIANT_COUNT + module_multi_instance::HoldReason::<module_multi_instance::Instance3>::VARIANT_COUNT; } ``` In addition, `MaxHolds` is removed (as suggested [here](#2657 (comment))) from `pallet_balances`, and its `Holds` are now bounded to `RuntimeHoldReason::VARIANT_COUNT`. Therefore, there is no need to let the runtime specify `MaxHolds`. ## For reviewers Relevant changes can be found here: - `substrate/frame/support/procedural/src/lib.rs` - `substrate/frame/support/procedural/src/pallet/parse/composite.rs` - `substrate/frame/support/procedural/src/pallet/expand/composite.rs` - `substrate/frame/support/procedural/src/construct_runtime/expand/composite_helper.rs` - `substrate/frame/support/procedural/src/construct_runtime/expand/hold_reason.rs` - `substrate/frame/support/procedural/src/construct_runtime/expand/freeze_reason.rs` - `substrate/frame/support/src/traits/misc.rs` And the rest of the files is just about removed `MaxHolds` from `pallet_balances` ## Next steps Do the same for `MaxFreezes` #2997. --------- Co-authored-by: command-bot <> Co-authored-by: Bastian Köcher <git@kchr.de> Co-authored-by: Dónal Murray <donal.murray@parity.io> Co-authored-by: gupnik <nikhilgupta.iitk@gmail.com>
This was referenced Mar 11, 2024
bgallois
pushed a commit
to duniter/duniter-polkadot-sdk
that referenced
this issue
Mar 25, 2024
… removed `pallet_balances`'s `MaxHolds` config (paritytech#2657) I started this investigation/issue based on @liamaharon question [here](paritytech#1801 (comment)). ## Problem The `pallet_balances` integrity test should correctly detect that the runtime has correct distinct `HoldReasons` variant count. I assume the same situation exists for RuntimeFreezeReason. It is not a critical problem, if we set `MaxHolds` with a sufficiently large value, everything should be ok. However, in this case, the integrity_test check becomes less useful. **Situation for "any" runtime:** - `HoldReason` enums from different pallets: ```rust /// from pallet_nis #[pallet::composite_enum] pub enum HoldReason { NftReceipt, } /// from pallet_preimage #[pallet::composite_enum] pub enum HoldReason { Preimage, } // from pallet_state-trie-migration #[pallet::composite_enum] pub enum HoldReason { SlashForContinueMigrate, SlashForMigrateCustomTop, SlashForMigrateCustomChild, } ``` - generated `RuntimeHoldReason` enum looks like: ```rust pub enum RuntimeHoldReason { #[codec(index = 32u8)] Preimage(pallet_preimage::HoldReason), #[codec(index = 38u8)] Nis(pallet_nis::HoldReason), #[codec(index = 42u8)] StateTrieMigration(pallet_state_trie_migration::HoldReason), } ``` - composite enum `RuntimeHoldReason` variant count is detected as `3` - we set `type MaxHolds = ConstU32<3>` - `pallet_balances::integrity_test` is ok with `3`(at least 3) However, the real problem can occur in a live runtime where some functionality might stop working. This is due to a total of 5 distinct hold reasons (for pallets with multi-instance support, it is even more), and not all of them can be used because of an incorrect `MaxHolds`, which is deemed acceptable according to the `integrity_test`: ``` // pseudo-code - if we try to call all of these: T::Currency::hold(&pallet_nis::HoldReason::NftReceipt.into(), &nft_owner, deposit)?; T::Currency::hold(&pallet_preimage::HoldReason::Preimage.into(), &nft_owner, deposit)?; T::Currency::hold(&pallet_state_trie_migration::HoldReason::SlashForContinueMigrate.into(), &nft_owner, deposit)?; // With `type MaxHolds = ConstU32<3>` these two will fail T::Currency::hold(&pallet_state_trie_migration::HoldReason::SlashForMigrateCustomTop.into(), &nft_owner, deposit)?; T::Currency::hold(&pallet_state_trie_migration::HoldReason::SlashForMigrateCustomChild.into(), &nft_owner, deposit)?; ``` ## Solutions A macro `#[pallet::*]` expansion is extended of `VariantCount` implementation for the `#[pallet::composite_enum]` enum type. This expansion generates the `VariantCount` implementation for pallets' `HoldReason`, `FreezeReason`, `LockId`, and `SlashReason`. Enum variants must be plain enum values without fields to ensure a deterministic count. The composite runtime enum, `RuntimeHoldReason` and `RuntimeFreezeReason`, now sets `VariantCount::VARIANT_COUNT` as the sum of pallets' enum `VariantCount::VARIANT_COUNT`: ```rust #[frame_support::pallet(dev_mode)] mod module_single_instance { #[pallet::composite_enum] pub enum HoldReason { ModuleSingleInstanceReason1, ModuleSingleInstanceReason2, } ... } #[frame_support::pallet(dev_mode)] mod module_multi_instance { #[pallet::composite_enum] pub enum HoldReason<I: 'static = ()> { ModuleMultiInstanceReason1, ModuleMultiInstanceReason2, ModuleMultiInstanceReason3, } ... } impl self::sp_api_hidden_includes_construct_runtime::hidden_include::traits::VariantCount for RuntimeHoldReason { const VARIANT_COUNT: u32 = 0 + module_single_instance::HoldReason::VARIANT_COUNT + module_multi_instance::HoldReason::<module_multi_instance::Instance1>::VARIANT_COUNT + module_multi_instance::HoldReason::<module_multi_instance::Instance2>::VARIANT_COUNT + module_multi_instance::HoldReason::<module_multi_instance::Instance3>::VARIANT_COUNT; } ``` In addition, `MaxHolds` is removed (as suggested [here](paritytech#2657 (comment))) from `pallet_balances`, and its `Holds` are now bounded to `RuntimeHoldReason::VARIANT_COUNT`. Therefore, there is no need to let the runtime specify `MaxHolds`. ## For reviewers Relevant changes can be found here: - `substrate/frame/support/procedural/src/lib.rs` - `substrate/frame/support/procedural/src/pallet/parse/composite.rs` - `substrate/frame/support/procedural/src/pallet/expand/composite.rs` - `substrate/frame/support/procedural/src/construct_runtime/expand/composite_helper.rs` - `substrate/frame/support/procedural/src/construct_runtime/expand/hold_reason.rs` - `substrate/frame/support/procedural/src/construct_runtime/expand/freeze_reason.rs` - `substrate/frame/support/src/traits/misc.rs` And the rest of the files is just about removed `MaxHolds` from `pallet_balances` ## Next steps Do the same for `MaxFreezes` paritytech#2997. --------- Co-authored-by: command-bot <> Co-authored-by: Bastian Köcher <git@kchr.de> Co-authored-by: Dónal Murray <donal.murray@parity.io> Co-authored-by: gupnik <nikhilgupta.iitk@gmail.com>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Continuation of #2657 for
RuntimeHoldReasonwhereMaxHoldsare removed.Summary
The
pallet_balanceshas storage itemFreezes. They useT::FreezeIdentifieras identifier, which is not meant to be a composite_enum. There is one usage ofFreezeIdentifierfor implementingInspectFreezehere. However, we also haveRuntimeFreezeReasonhere, which is not used anywhere.For example, I see one usage of freezes in
pallet_nomination_poolshere, where the composite_enumFreezeReason(evaluated asRuntimeFreezeReason) is used as aFreezeIdentifier.In the Westend runtime, it is configured as:
There is also an
integrity_testthat verifiesRuntimeFreezeReason::VARIANT_COUNTagainstT::MaxFreezes.And it seems that
RuntimeFreezeReason::VARIANT_COUNTis kind of disconnected fromFreezeIdentifier.According to the original implementation:
Questions:
RuntimeFreezeReasonandFreezeIdentifier?FreezeIdentifierused as an identifier forFreezesinstead ofRuntimeFreezeReason?Proposed solution
We could simplify it as follows:
MaxFreezesfrompallet_balancesFreezeIdentifierfrompallet_balances(add a migration to theRuntimeFreezeReasonwith sometry-statecheck)RuntimeFreezeReasonas an id and also bound vec with itsVARIANT_COUNT(as we did forRuntimeHoldReason)I found out, that @bkchr did the same thing for
HoldIdentifierhere: paritytech/substrate#13869The text was updated successfully, but these errors were encountered: