rust - 为什么Serde不能用于Libra的加密目的序列化和反序列化？

Question

In Libra, participants pass around messages or data structures that often times need to be signed by a prover and verified by one or more verifiers. Serialization in this context refers to the process of converting a message into a byte array. Many serialization approaches support loose standards such that two implementations can produce two different byte streams that would represent the same, identical message. While for many applications, non-deterministic serialization causes no issues, it does so for applications using serialization for cryptographic purposes. For example, given a signature and a message, a verifier may not unable to produce the same serialized byte array constructed by the prover when the prover signed the message resulting in a non-verifiable message. In other words, to ensure message verifiability when using non-deterministic serialization, participants must either retain the original serialized bytes or risk losing the ability to verify messages. This creates a burden requiring participants to maintain both a copy of the serialized bytes and the deserialized message often leading to confusion about safety and correctness. While there exist a handful of existing deterministic serialization formats, there is no obvious choice. To address this, we propose Libra Canonical Serialization that defines a deterministic means for translating a message into bytes.

那是 what the Libra project says。什么是真正的确定性序列化？如果serde不是确定性的，在序列化数据结构后如何反序列化？

Answer 1

HashSet和HashMap可以根据插入，更新和删除的确切顺序将项目插入不同的插槽。这些数据结构的序列化器以插槽顺序运行，发出一系列(键，值)对。由于插槽顺序是不确定的，因此序列化的字节将是不确定的。