This repository has been archived by the owner on Nov 15, 2023. It is now read-only.
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
fixes paritytech/polkadot-sdk#257
In the quest of getting a clear picture of the problem,
I have included the final_scaled_tip for normal txn to include their inclusion fees. But I have a question on regarding the normal txn will be competing with operational txn and I think we have to make sure the normal txn scaled_tip wont be greater than operational txn scaled tip
Also concerning Gav issue on standard scale txn priority paritytech/polkadot-sdk#277 , I think we should change the how we operate on txn priority and maybe have a trait so that substrate chains can define how they want the priority to be and provide maybe a default implementation.