Look into adding option for client software to specify using a non-inlined payload

[fxamacker]

For some cases, I think it might be useful for client software to be able to specify using a non-inlined payload (aka register) to simplify concurrency.

Suggested Solution

Look into adding an option in Atree's API that allows client software to specify using a non-inlined payload. For example, new Atree API can allow Cadence to pass a flag (or callback function) to indicate if the given data structure can be inlined or not.

This shouldn't require migration and it should be backward compatible with existing data.

Scope

Look into the feasibility of adding support for client software to specify to Atree if a given data structure can or cannot be inlinded (e.g., by using new Atree API to pass a flag or callback function from Cadence to Atree).

NOTE: It is outside the scope of this issue to evaluate if some use case(s) that would utilize this new feature is better/worse than alternative solutions.

Results and Caveats

As mentioned on Dec 8, 2025, it appears to be feasible to add support for this new feature to Atree.

However, it would be safer (and easier for future optimizations) to not expose the option of using not-inlinable payloads for the general public to use directly on Flow mainnet, because not inlining payloads can increase bloat and reduce efficiency of:

• payload databases
• payload indexers
• payload caches (e.g., LRU caches)
• MTrie (execution state)

My preference (even for internal use) is to limit the use of this new feature as much as possible (e.g., only allow it for system account vaults if appropriate to do so and carefully approve other uses in the future after evaluation). Adding a feature or making it more widely available is easier than taking that feature away.

As noted in "Scope" section, it is outside the scope of this issue to evaluate if some use case(s) that would utilize this new feature is better/worse than alternative solutions.

[fxamacker]

On Friday and Sunday, I looked into feasibility of adding a feature to Atree so that the caller (Cadence) can specify to Atree whether a container is inlinable or not-inlinable.

I think we can add a feature in Atree that lets us store arrays and maps without inlining, by allowing the caller (Cadence) to provide a flag (or callback function) to Atree to specify whether or not an array or map should be inlined.

This feature can be useful because Janez mentioned that a blocker for using multiple vaults within the same account (for the concurrent execution project) is that each vault isn't guaranteed to be stored in separate payloads, so he might use a workaround that involves using child accounts (instead of using multiple vaults per account).

Caveats

It would be safer (and easier for future optimizations) to not expose the option of using not-inlinable payloads directly to the general public because not inlining payloads can increase bloat in:

• payload databases
• payload indexers
• execution state

My preference (even for internal use) is to limit the use of this new feature as much as possible (e.g., only allow it for vaults and carefully approve other uses in the future). Adding a feature or making it more widely available is easier than taking that feature away.

Today, we discussed the possibility of using this feature during the Execution Sync.

[fxamacker]

I updated the issue text by adding "Results and Caveats" section and "Scope" section.

As mentioned in "Scope" section, it is outside the scope of this issue to evaluate if some use case(s) that would utilize this new feature is better/worse than alternative solutions.