Matter Labs: Why We Failed to Launch zkSync 2.0 in August?

Source: Matter Labs Blog

Compiled by: Chain Catcher

On March 27, 2021, we announced the plans for zkSync 1.x and 2.0.
We successfully upgraded zkSync 1.x and deployed it to the mainnet, but we failed to meet our prediction for the August release of zkSync 2.0. In this article, we will discuss the delays, the gradual rollout of the testnet, and the fair launch of the mainnet.
Why the Delay?
As early as March, we completed the design of zkSync 2.0 and estimated the time required for construction. Due to consistently high gas fees, our design prioritized security and time, making some trade-offs in efficiency, optimization, and compatibility with Ethereum—adapting the line environment to EVM is not straightforward due to fundamental limitations of the line.
However, there was one key decision that did not prioritize the mainnet launch time: the choice of LLVM. While implementing a custom compiler from scratch would have been faster, there was no choice but to use LLVM in the long run. LLVM is built by engineers working on industrial-grade products (LLVM is an integral part of macOS and iOS) and is the state-of-the-art compiler framework for producing industrial-grade products, forcing us to consider debuggers, linkers, assemblers, disassemblers, and binary utilities, even when we just wanted to quickly release a compiler. By leveraging LLVM, our compiler has all classic optimizations, over 20,000 regression tests, and 3,000 integration/executable tests, with a low maintenance burden.
In May, although our nodes and VM were ready, there were some unforeseen incompatibilities with our architecture and LLVM that required additional time to integrate into the framework. We did not want to launch a testnet missing one of three core components, but even with the initial overhead, we still stand by our decision to adopt LLVM from the beginning. Matter Labs will never compromise on security or code quality. Following best industrial practices is slow, but the alternative is programming with technical debt. Debt must be repaid someday.
Building zkSync 2.0 has been a tense R&D process:

• A snark-friendly EVM and different data availability strategies for each account in the same address space have never been done before;
• It requires addressing the needs of the compiler, zkEVM, and nodes simultaneously.

Since implementation is highly related to research, many times we found better solutions that led to lower costs, better compatibility, or more convenient interfaces:

• We went through several iterations to improve the efficiency of the compiler, which gave us ideas on how to make our VM more efficient (more details in the technical insights post);
• Our API and SDK are very similar to Web3 API and ethers, so we decided to support both with additional zkSync L2 specific features;
• We found a way to lift the limit on transaction execution tracking length, allowing for arbitrarily large transactions.

With gas prices averaging 20 Gwei in June and July, we felt the time pressure eased, and when we saw clear ways to do better, we did not want to launch. Now, we will integrate all improvements before releasing any version to avoid any disruptive upgrades as much as possible. The version we are currently developing has more features, lower costs, and is more compatible and convenient than the version we released in March.
However, we do want to be straightforward with you. Aiming for August was a very ambitious goal. We are now clear that even if we abandon improvements in efficiency and Ethereum compatibility, delivering without compromising on quality is also impossible. This is the nature of cutting-edge R&D in technological innovation: any prediction is always a guess around many unknown factors.
Status Update
In the past 6 months, we have achieved and tested:

• Execution nodes
• zkEVM (line and execution environment)
• Solidity and Zinc compilers (compiled and successfully tested Uniswap v2)

We are finalizing:

• Web3+ API (fully supports out-of-the-box Web3 API + zkSync L2 specific features)
• Ethereum+ SDK
• L1 to L2 communication (important for anti-censorship: funds from zkSync can be withdrawn via L1 transactions, even if you have to transfer them from smart contracts first.)

In progress:

• Integrating line/prover and data availability protocols into the executor nodes
• L2 to L1 communication (e.g., triggering L1 contracts from L2)
• Vyper developers: The Vyper team is currently building a Vyper-to-LLVM frontend, and our compiler team is working closely with them to ensure you can seamlessly deploy smart contracts in Vyper on zkSync 2.0.

Testnet Plan
We will open the testnet in three phases:

1. North Star
   In a few weeks, we will begin releasing the first version of a set of protocols. We hope to work closely with these teams to facilitate rapid iteration of feedback and fixes when we launch the open testnet, minimizing bugs and challenges. If you are interested in this phase, please fill out this form.
2. Spotlight Search
   Then, we hope to open it to those willing to create content. This can range from tweet threads to blog posts to complete developer demo videos. Our goal is to have a variety of useful resources available whenever we fully open our testnet, serving as another round of testing for our system. If you enjoy writing or creating technical content, please fill out this form.
3. Kindergarten
   The testnet will be open to everyone.
   Mainnet Launch
   Matter Labs is committed to building a neutral second layer where all protocols and infrastructure can be deployed and operated in a fair and just manner.
   Therefore, our mainnet release is a fair launch that is crucial for us: we will provide everyone with time to prepare and launch projects on zkSync under equal conditions.
   We are nearing the finish line! No major changes are expected, and we are working hard to integrate all active components, implement a convenient SDK, and write documentation.