离线 · 无上传 · 浏览器本地完成Offline · no uploads · runs in the browser

TXT 转 Excel——离线、文本得是 JSON 数组、不上传TXT to Excel offline — the text must be a JSON array

这条链是四步，前两步决定了它能不能成立：`txt>html` 把整个文件按字符包进一个 `<pre>`（只转义 `&`、`<`、`>`，换行写成 `<br>`，文档标题取文件名的主干（去掉扩展名），没有名字时是 `converted`），**不解释任何记号**——`#` 就是 `#`，粘进去的 HTML 只会是显示出来的字符；`html>json` 发现文档里没有表格，就退一步把**第一个** `<pre>` 的文字当 JSON 解析。所以这一步真正要的条件是「文本本身是一段合法 JSON」，而且是**数组**：中间那步是对象的话，第三步 `json>csv` 会报「JSON 内容必须是包含数据的数组才能转换为 CSV」。两个细节值得知道：`<br>` 对 `textContent` 不贡献字符，所以你文件里的换行到取回时已经没了——JSON 语法不依赖换行，照常解析，但你手工排的版式不会带着走；回退只看第一个 `<pre>`，如果它排在真正的 JSON 块之前，就会解析失败而不是跳过去找。第三步摊成表：列是所有记录键名的并集，某条记录缺的键留空格，没有键可挂的元素写进合成的 `_value` 列而不是丢掉，产物开头带 UTF-8 BOM。第四步 `csv>xlsx` 才是决定工作簿形态的地方：读 CSV 用 `raw: true`，SheetJS 的推断器碰不到数据，随后只有一道规则把该是数字的格子还回去——十进制形状、`String(Number(x)) === x` 逐字节回得来、有效数字不超过 15 位；16 位以上刻意留在文本：那条规则的上限就是 15 位有效数字（仓库里按位数量过 30 万个整数的回写命中率，16 位起只剩 94.5%）。仍是文本的格子再加公式护栏，并把数字格式钉成文本（写出的那一格是内置格式号 49）。还有一条得说在前头：如果你的文本里存的是超长整数（比如 26 位账号），`JSON.parse` 那一刻它就已经是双精度值了，这条链保住的是**解析之后**的样子，超过 15 位有效数字就落在文本里——所以你看到的是那个双精度值自己的最短十进制写法，并且被当作文本存下——不是悄悄四舍五入过的数字。编码判定在第一步做：先按 BOM 与 NUL 的奇偶形状认 UTF-16，再依次试严格的 UTF-8、GB18030、GBK，四种都不像文本就明确报错。全程离线，只有 storage 权限。Four steps, and the first two decide whether the route can run at all: `txt>html` wraps the whole file character by character into a `<pre>` (only `&`, `<` and `>` are escaped, newlines become `<br>`, and the document title is the file’s basename with the extension dropped, or `converted` when there is none), **interpreting no markup** — a `#` stays a `#`, and HTML pasted into the text shows up as characters; `html>json`, finding no table in that document, falls back to parsing the text of the **first** `<pre>` as JSON. So the real condition is “the text itself is valid JSON”, and an **array** at that: if the middle step yields an object, `json>csv` reports “JSON must be an array containing data to convert to CSV”. Two details worth knowing: `<br>` contributes nothing to `textContent`, so the line breaks in your file are gone by the time the text is read back — JSON does not depend on them, so it still parses, but your own layout does not travel; and the fallback reads only the first `<pre>`, so a non-JSON block ahead of the real one fails the parse instead of being skipped past. The third step spreads the array into a table: columns are the union of every record’s keys, a key a record lacks becomes a blank cell, elements with no keys go into a synthetic `_value` column rather than being dropped, and the file starts with a UTF-8 BOM. The fourth, `csv>xlsx`, is what shapes the workbook: the CSV is read with `raw: true`, so SheetJS’s inferrer never touches the data, and one rule alone returns a number type — decimal shape, `String(Number(x)) === x` giving the exact characters back, at most 15 significant digits. Sixteen and above stay text deliberately: the rule itself stops at 15 significant digits, measured in this repo over 300 000 sampled integers per digit length (94.5% survive at 16). Cells still holding text get the formula guard and a pinned text format (that cell carries the built-in format id 49). One caveat stated up front: if your text stores very long integers (a 26-digit account, say), `JSON.parse` has already turned it into a double, and this chain preserves what came **after** the parse — past 15 significant digits it lands in text, so what you see is that double’s own shortest decimal text, kept as a text cell, rather than a silently rounded number. Encoding is settled in the first step: UTF-16 from its BOM or NUL parity, then strict UTF-8, GB18030 and GBK in order, with an outright error when the bytes are not text at all. Everything runs offline on the storage permission.