import { textDirection } from "~/textDirection";
type HastNode = {
type?: string;
tagName?: string;
value?: string;
properties?: Record<string, unknown>;
children?: HastNode[];
};
// Raw HTML such as `<li>a</li><li>b</li>` has no whitespace between its
// blocks, so they add their own to keep the words apart.
const BLOCK_TAGS = new Set([
"blockquote",
"br",
"details",
"div",
"h1",
"h2",
"h3",
"h4",
"h5",
"h6",
"hr",
"li",
"ol",
"p",
"summary",
"table",
"td",
"th",
"tr",
"ul",
]);
function proseText(node: HastNode): string {
if (node.type === "text") return node.value ?? "";
if (node.tagName === "code") return " ";
const text = (node.children ?? []).map(proseText).join("");
return node.tagName && BLOCK_TAGS.has(node.tagName) ? ` ${text} ` : text;
}
function isolateInlineCode(node: HastNode): void {
if (node.tagName === "pre") return;
if (node.tagName === "code") {
node.properties = { ...node.properties, dir: "ltr" };
return;
}
node.children?.forEach(isolateInlineCode);
}
/**
* Gives each top-level block the direction of its own words, so an Arabic or
* Hebrew reply reads right to left beside English ones. Lines inside a list or
* quote follow it, keeping markers and rules on one side. Inline code is left
* out of the count and isolated left to right, so its punctuation stays put.
*/
export function rehypeTextDirection() {
return (tree: HastNode) => {
for (const block of tree.children ?? []) {
if (block.type !== "element" || block.tagName === "pre") continue;
const dir = textDirection(proseText(block));
if (dir) block.properties = { ...block.properties, dir };
}
isolateInlineCode(tree);
};
}