Upgradeable proxy contract from scratch

Photo by Rafał Naczyński on Unsplash

Immutability gives Ethereum a big benefit: once a smart contract is deployed, there’s no way to change it. This means that smart contract logic won’t change over time and it’ll remain as reliable as it was from the beginning, e.g. you won’t suddenly see a contract doing something no one thought it could do.

However, immutability comes with drawbacks that can be significant in some cases. For example, due to immutability you won’t be able to fix a bug in a contract or upgrade it to optimize gas consumption or to improve existing functionality. Thus, sometimes, mutability is necessary.

Ditching immutability or making it optional would be a bad solution that would rob Ethereum of one of its core features. Instead, an elegant and tricky solution was found.

You can find full code of this blog post on GitHub.

Proxy smart contracts

Proxy contract is a contract that relays all calls to another smart contract, that is called implementation contract. Schematically, this looks like that:

Implementation contract implements some business logic users wants to interact with, but instead of interacting directly with the contract, they call functions from the proxy contract. As a result of such chaining, it becomes possible to swap the implementation contract with a different one: proxy contract knows implementation contract address.

Let’s write a simple contract and then see how we can use a proxy to call it. We don’t need something complicated, just a basic contract that allows to read and write some state variable:

Now, let’s build a proxy contract step-by-step:

As you remember, proxy contract knows implementation contract address. Let’s add that:

That was the easy part. Now, how do we relay all calls to the proxy contract to the logic contract?

Solidity has a special fallback function which is called whenever a function unsupported by a contract is called. This function is called…fallback! Unfortunately, this is all Solidity can give us here since Solidity is a higher-level abstraction. To achieve our goal we have to go deeper!

How contract functions are called

Ethereum transactions contain a field called data. This field is optional and must be empty when sending ethers, but, when interacting with a contract, it must contain something. It contains call data, which is information required to call a specific contract function. This information includes:

  1. Function identifier, which is defined as the first 4 bytes of hashed function signature, e.g.:
  2. Function arguments that follow function identifier and are encoded according to the ABI specification.

Every smart contract compiled by Solidity has a branching logic that parses call data and decides which function to call depending on function identifier extracted from call data. Solidity won’t allow us to make decisions on that level of deepness, so we’ll have to use assembly to write relaying logic in the Yul language.

What we want is, in the fallback function, to get call data and pass it to the implementation contract as is, without parsing or modifying it. Here’s how to do that:

Let’s break it down because it looks too complicated:

  1. First, we need to load call data into memory. We do that in the first two lines:

Memory in EVM is organized in slots, with every slot having an index and occupying 32 bytes. We’re using calldatasize function to get call data size and are using calldatacopy to copy call data of specific size to a slot located at index ptr (it’ll occupy other slots if it doesn’t fit). mload is a function that reads 32 bytes from specified index, and index 0x40 points a special slot that contains the index of the next free memory slot. So we’re basically saving call data to a free memory slot.

2. Next, we relay the call:

delegatecall is a tricky way to call another contract, and we’ll soon learn why it’s tricky.

First argument in delegatecall is how much gas is remaining in the current call – this tells the other contract how much of gas it’s allowed to spend.

Next argument is the address of the contract we’re calling, and it looks somewhat complicated here. implementation is a state variable, it’s value is stored in memory when the code is executing. .slot postfix is a feature of Solidity: it allows to easily get the slot address of a state variable. Having that address, we’re calling sload function to read the value at that address. We need to do all that because Yul doesn’t do anything with state variables, since state variables is syntactic sugar of Solidity.

Last two arguments, which are zeroes, are out and outsize respectively: they allow to define where in memory to store return data. But since we’re relaying all calls and we’re not aware of what data each call returns, we won’t use these variables. Instead…

3. We’ll use functions similar to the two call data functions we used earlier:

Here, we’re getting the size of the data returned by the relayed call and save it in a slot at index ptr – we don’t need its previous value (call data) anymore.

4. Finally, we’re checking if the relayed call was successful and returning; otherwise, we’re reverting. Notice that we’re using the data returned by the relayed call in both of the cases: we want to return what was returned by the call and we want to revert with the same message if the call has reverted.

In fact, Solidity also has delegatecall function, but we’re not using it here for a reason: we want the Proxy contract to return whatever was returned from the callee and we don’t know return data type in advance. Since Solidity is a statically-typed language, it requires us to define function return type before compilation.

Let’s see if this works! We’ll do that by testing.

First, let’s set up our tests: we need to deploy both contracts:

Then, let’s ensure that implementation contract address was set correctly:

And then, let’s ensure that proxying works:

Ethers.js is quite a helpful library: it parses ABI and generates contract methods in JS to make our life easier. However, in this situation, this is not what we want. We want to call our proxy logic using ABI from the implementation contract. That’s why we need to create a different contract instance, using the address of the proxy contract and the ABI of the implementation contract.

Let’s run the test:

Uh-oh! This is not something we expected.

The time has come to learn about the trickiness of delegatecall.


Every program consists of two components:

  1. State, which is the data that’s stored in memory, or in a persistent storage, and that’s used by the program.
  2. Logic, which is the functionality of the program.

Likewise, smart contracts also have state and logic. Smart contract’s persistent state is stored on blockchain and is accessible via state variables. There’s also in-memeory state, but it’s not relevant for our discussion.

Both call and delegatecall are used to call another contract but they differ in how they handle callee contract’s state:

  1. When using call, caller and callee have their own, separated, states. This makes sense and this is what we expect by default.
  2. When delegatecall is used, callee uses caller’s state. That’s it: the contract you’re calling with delegatecall uses the state of the caller contract.

We could’ve used call instead of delegatecall but that’d have broken the upgradeability!

Imagine that we’re running a smart contract for long time and there’s a lot of data stored on chain but then we decide to upgrade the contract and deploy a different implementation. That would mean that we would have to migrate all the data from the smart contract to the new implementation! That can be quite pricey depending on network congestion. With delegatecall, we’re storing state in the proxy contract, which allows us to swap implementations without the need of migrating data!

Now, let’s return to the error in tests.

What has happened? Let’s look at state variables of our contracts:

  1. In the proxy contract, we’re storing implementation contract address in the first state variable:

2. In the logic contract, we’re storing magic number in the first state variable:

As I said earlier, state variables is syntactic sugar in Solidity: they simplify reading and writing to/from memory. What happens under the hood is:

  1. Every state variable has an index that starts from 0. First state variable of a contract has index 0, second one — 1, third one — 2, and so on.
  2. Every state variable is mapped to a slot in memory. Its value is stored in memory.
  3. Slot address is defined as hashed state variable index, literally:


And this is what causes the problem: when Logic contract is called via delegatecall from Proxy contract, variable magicNumber points at the same memory slot as variable implementation in Proxy contract! Since both of them have the same index 0. And that big number that we saw in the failed test is the implementation contract address (in the decimal system)!

To better understand the idea, try changing the proxy contract like that and run the tests:

You can learn more about the layout of state variables here.

Resolving the state collision

The collision we saw earlier happened because the implementation state variable in Proxy has the same slot index (and, thus, address) as the magicNumber state variable. And since we’re using delegatecall, Proxy’s state is used when Logic is called.

There are multiple ways of resolving such collisions, I’ll show you one of them, the one that’s easier to implement and that still provides a reliable way to avoid collisions.

The idea is to store the implementation variable at a unique address, such that it’s very unlikely that anything else would also use it. Since addresses are simply byte32 values, we can use whatever address we want!

So, we need to choose an address to store the implementation variable value. Luckily, there’s already a (draft) EIP – EIP-1967 – that proposes a scheme to find such address. The idea is simple:

We’re hashing the string “eip1967.proxy.implementation” and then subtracting 1 from the hash. Hashing is needed to define the address in a unique way (it’s unlikely someone is using this string for other purposes) and subtraction is needed to reduce the probability of getting the hash even further.

We have a solution, let’s implement it!

Low-level storage management

As we’ve already learned, Solidity is a higher level language and it doesn’t allow us to do low-level manipulations — we have to use Yul instead.

This time, we’ll build a library to help us save an Ethereum address at a custom slot and read it from there — that’s basically it, the library is simple and has only two functions:

getAddressAt wraps low-level sload function and setAddressAt wraps sstore function, which allows to read or write data in a specific slot respectively.

Next step, we need to define the slot address:

Notice that this is a constant so it won’t occupy a memory slot — the compiler will replace all its occurrences with its value. Also, notice that the implementation state variable has gone now – we won’t use it!

Next, rewrite the setter and getter functions:

Finally, rework the fallback function to extract Yul code into _delegate function so we don’t mix Solidity and Yul in one function:

And slightly rework the delegatecall for it to take implementation address as is, without calling sload:

That’s it for the proxy contract! However, if you run the tests you’ll see this:

Callee contract initialization

We forgot one thing: we cannot use constructor to initialize callee contract when it’s used via a Proxy. When constructor is used to initialize state variables, they’re initialized within the state of the contract. But we want them to be initialized within the state of the proxy contract.

There’s a simple solution: use a custom initialization function instead of constructor. It looks like that:

This function must be called after an implementation contract was set in the proxy contract. In our tests, it’ll look like that:

If you run the tests now, they will pass.

Changing implementations

Now, let’s add a new version of the logic contract and see if we can upgrade our older implementation.

Copy Logic.sol, rename it to LogicV2, and add a new function:

Everything else remains the same.

Now, let’s check if it works. First, deploy the new contract:

Then, change implementation in the proxy contract:

Since the logic contract has already been initialized, we don’t need to initialize it again.

Next, initialize a proxy contract instance with the ABI from the new implementation:

Finally, do something to ensure version 2 is working correctly:

Run the tests and see that they pass!


Studying upgradeable proxy contracts is a good way to start learning about some deeper concepts of EVM, like storage layout, storage management, and contract calls. If some parts are not clear to you, feel free returning and giving them another try. They’re not as hard as they might look like. If everything was clear to you, I’d recommend checking out the proxy contract implementation by OpenZeppelin — it’ll give you a deeper understanding.


  1. Source code
  2. Yul language documentation
  3. Contract ABI specification
  4. Proxy patterns, a detailed post in OpenZeppelin Blog about different patterns used in proxy contracts.

Check out my blog to learn more about blockchains, smart contracts, and DeFi.

Join Coinmonks Telegram Channel and learn about crypto trading and investing

Also Read

Upgradeable proxy contract from scratch was originally published in Coinmonks on Medium, where people are continuing the conversation by highlighting and responding to this story.