Hello all, today I am introducing a methodology I have been personally using to enhance post-deployment upgradeability, and am excited to hear the community’s feedback regarding the concept.
The core of the idea can be understood from the name –
Consolidated: A single contract centrally manages storage for the rest of the contract system.
Dynamic: Storage structures can be adapted post-deployment at the contract level.
Storage: The focus is on making storage evolvable and extendable in itself.
CDS aims to introduce a new methodology to upgradeability – that (admittedly) sacrifices ease-of-implementation for post-deployment versatility. Two features make up the core of the model:
Dynamic storage spaces: Segregated storage areas (defined by a hash structure) that can be created through a function invocation in a post-deployment setting. They are populated by a mapping-of-extendable-struct.
Extendable structs: A custom struct type that can have new members appended to it through function invocation.
Together, these features allow storage to evolve and adapt as system requirements change, without needing state migrations or rigid layouts.
It does come with tradeoffs—ease of implementation is sacrificed for flexibility—but I’ve found this model useful for systems where upgradeability requirements can grow over time. Additionally, the high applicability of a complete model leaves a lot of room for the complex implementation and obvious vulnerability risks to be abstracted away as working models are iterated upon. It’s certainly an unorthodox approach, but I believe that this model does provide clear advantages.
Really looking forward to hearing the community’s thoughts about this!
@wisecameron your methodology is fascinating and tackles one of the more challenging aspects of blockchain systems: achieving true upgradeability without introducing major disruptions. The Consolidated Dynamic Storage (CDS) model, as you’ve outlined, strikes a bold balance between adaptability and complexity, and it opens the door to a much-needed evolution in smart contract design.
The idea of dynamic storage spaces defined by hash structures is particularly compelling. It introduces a modularity that not only future-proofs the contract but also allows developers to adapt to changing requirements without cumbersome state migrations. This is especially critical for long-lived systems that need to scale and evolve in unpredictable ways.
Your concept of extendable structs is equally intriguing. The ability to append new members dynamically brings a level of flexibility rarely seen in existing models. It effectively transforms storage into a living entity, capable of growing alongside the system’s needs. The tradeoff—sacrificing ease of implementation—feels justified for systems where adaptability outweighs simplicity, especially given the potential to abstract away these complexities through reusable patterns and libraries.
However, the unorthodox nature of CDS does highlight areas where the community might need to collaborate further:
Auditing and security: Dynamically evolving storage structures could introduce vulnerabilities, particularly in managing boundaries and ensuring that state modifications are well-controlled.
Standardization and tooling: For CDS to gain widespread adoption, robust developer tools and clear patterns will be essential to manage complexity and mitigate implementation risks.
Your willingness to embrace tradeoffs for versatility demonstrates a forward-thinking approach, and I can see this methodology being especially impactful in environments like DAO frameworks, modular dApps, and systems with high uncertainty about future requirements. Looking forward to seeing this concept evolve and how the community engages with it!
Great callout on EIP-6224 – the Dependency Registry definitely shares conceptual overlap with CDS, but I agree that the two systems fundamentally diverge in their approach – and I believe it’s actually to quite a great extent under the hood:
• Dependency Registry essentially consolidates the process of linking upgrades between dependent contracts in a traditional proxy-delegate model. That structural similarity is clear – there’s a central control layer that’s interacting with the rest of the system.
• CDS manages all storage itself, and leverages the extendable structs / storage spaces to scale infinitely in-place. Similar to Dependency Registry, it’s a ‘central’ contract that has calls propagated to it externally. Effectively, there is no longer any need to use proxy-delegate for the linked contracts, which are typically pure + globals (although globals can be stored in CDS as well) – their mapped storage is all delegated to the CDS layer.
One cool thing about this is you effectively get implicit delegate access between linked contracts. This access can be secured through basic permission management systems—either at a storage-space level or even down to specific struct members. This plus the ease of redeploying what are effectively pure contracts makes it really easy to integrate new systems with longstanding ones.
As a footnote, CDS effectively favors making capabilities exist and puts the onus on developers to make it secure – but critically, it can have full decentralization with permission management + a governance system. The centralization of the layer makes that process much more digestible.
Thanks for checking it out and for your reply :). Big fan of your work by the way, it’s definitely some of the most interesting stuff that pops up in my Linkedin feed. That ZK resources post has been particularly helpful.
Just took a deeper dive into the ERC. Several questions:
Is it correct that the entryIndex from the spec is safeIndex in the reference implementation?
It would be great to see an implementation of the put() function. Also, I think it would be great to remember the names of variables put in order to retrieve them by name. Can mapping be used?
It is quite hard to understand the overall storage layout after initialization and push steps. A low-level diagram would be highly appreciated.
What is the intended use of the system? Is it a standalone contract? Is it a library that manages the storage of a specific smart contact?
entryIndex actually refers to the sequential index (ie; userData[0] or userData[1]). The safeIndex stores the last fixed-sized member inserted into the system for bitCount calculations, which is necessary because strings (equivalent to arrays, functionally) have dynamic sizing.
Yeah, I think this is a great point regarding naming variables. My big idea for making the system more palatable in a syntactical sense has just been to build a dedicated VS code extension that provides alias visualization but that could be better. It’s true that the functionality you just described could be implemented with mappings relatively easily, with a moderate gas cost increase. I mainly opted not to include this in the base system because the core implementation is already “hard” compared to competing solutions like proxy-delegate.
On it, great feedback and agreed.
Yeah, the CDS layer is a standalone contract. Essentially, you link logic contracts to your CDS layer, and let it handle all of your (particularly mapped since they are more cumbersome, although globals are fine too) storage.
The centrality of the CDS layer seems like a big drawback at face value, since it’s a single point of failure AND it’s rather involved in terms of low-level logic and bitwise operations. However, I think it’s actually the ideal structure fundamentally because you can fully prevent non-authorized invocations with basic permission management, and it’s structurally more palatable and efficient than using a complex array of linked storage dependencies like you see in proxy-delegate.
Given that a CDS layer has fully-malleable storage (ie; you can introduce new storage spaces and custom extendible structs through simple function invocations) a single layer can easily support an arbitrary-sized linked system. Additionally, the “shared” storage space makes deep integrations with legacy logic trivial, which is a big plus.
The main premise behind CDS is to front-load complexity (which is nullified in practice because I will be open-sourcing a highly-optimized CDS implementation called HoneyBadger in a month or two) in exchange for making storage fully fluid and controllable. The key reasons why it’s a step up from competing solutions like proxy-delegate are because it’s:
A) More practical to extend storage structures with a function invocation rather than re-deployment. I don’t claim it’s impossible to add new struct members to a deployed system given proxy-delegate is leveraged, but it’s certainly more cumbersome and error-prone.
B) Structurally, it manages storage in a single layer, which is significantly less complex than leveraging a collection of proxy contracts or diamonds.
C) Equivalent storage space access between linked contracts makes close interoperability relatively trivial. It’s almost similar to 7208, but I think in a more complete package (at the cost of being harder to implement up-front).
You’re the man for taking the time to read my proposal and reply, those are some great points and it’s obvious that I need to include better clarifications, I sincerely appreciate it .
.