HackerNews中文版

大家好，我构建了一个名为 SJT（结构化 JSON 表格）的小型实验性格式，用于优化 API 中的数据传输。这个想法很简单：SJT 将结构（标头）与值分开，而不是为每一行重复对象键。这使得它更紧凑，也更容易进行流式传输。例如，使用 Discord 的 /messages 端点：原始 JSON 负载：约 50,110 字节使用 SJT 编码的相同数据：约 26,494 字节因此，您大约可以减少 50% 的大小，同时仍然能够增量解码（逐条记录）。令人惊讶的是，解码甚至可以比纯 JSON 更快，因为字符串解析的开销更少。快速基准测试： | 格式 | 大小 (KB) | 编码时间 | 解码时间 | | ----------- | --------- | ----------- | ----------- | | JSON | 3849.34 | 41.81 毫秒 | 51.86 毫秒 | | JSON + Gzip | 379.67 | 55.66 毫秒 | 39.61 毫秒 | | MessagePack | 2858.83 | 51.66 毫秒 | 74.53 毫秒 | | SJT (json) | 2433.38 | 36.76 毫秒 | 42.13 毫秒 | | SJT + Gzip | 359.00 | 69.59 毫秒 | 46.82 毫秒 | 测试条件：数据集：合成表格数据集，包含 50,000 条记录，包含混合基本字段、嵌套数组和嵌套对象（代表大型 REST API 负载）。运行时：Node.js 20 (V8 引擎)。实现：JavaScript (通过 sjt.js)。大小 (KB)：未压缩大小，以千字节为单位（针对二进制格式进行估算）。编码 / 解码 (ms)：序列化/反序列化整个数据集的平均时间，以毫秒为单位。规范：[https://github.com/SJTF/SJT](https://github.com/SJTF/SJT) JS 实现：[https://github.com/yukiakai212/SJT.js](https://github.com/yukiakai212/SJT.js) 很想听听那些使用过 JSON 密集型 API、流式传输或紧凑数据格式（CSV、Parquet 等）的人的反馈。

查看原文

Hi HN, I built a small experimental format called SJT (Structured JSON Table) to optimize data transport in APIs. The idea is simple: instead of repeating object keys for every row, SJT separates the structure (headers) from the values. This makes it both more compact and easier to stream.For example, with Discord’s /messages endpoint:Raw JSON payload: ~50,110 bytesSame data encoded with SJT: ~26,494 bytesSo you get about a 50% reduction in size, while still being able to decode incrementally (record by record). Surprisingly, decoding can even be faster than plain JSON, because there’s less string parsing overhead.Quick benchmark:| Format | Size (KB) | Encode Time | Decode Time || ----------- | --------- | ----------- | ----------- | | JSON | 3849.34 | 41.81 ms | 51.86 ms || JSON + Gzip | 379.67 | 55.66 ms | 39.61 ms || MessagePack | 2858.83 | 51.66 ms | 74.53 ms || SJT (json) | 2433.38 | 36.76 ms | 42.13 ms || SJT + Gzip | 359.00 | 69.59 ms | 46.82 ms |Test conditions:Dataset: Synthetic tabular dataset containing 50,000 records with mixed primitive fields, nested arrays, and nested objects (representative of large REST API payloads).Runtime: Node.js 20 (V8 engine).Implementation: JavaScript (via sjt.js).Size (KB): Uncompressed size in kilobytes (estimated for binary formats).Encode / Decode (ms): Average time in milliseconds to serialize/deserialize the entire dataset.Spec: <a href="https://github.com/SJTF/SJT" rel="nofollow">https://github.com/SJTF/SJT</a>JS implementation: <a href="https://github.com/yukiakai212/SJT.js" rel="nofollow">https://github.com/yukiakai212/SJT.js</a>Curious to hear feedback from people who have worked with JSON-heavy APIs, streaming, or compact data formats (CSV, Parquet, etc.).

Show HN: SJT - 一种轻量级的结构化 JSON 表格格式，适用于 API