# htmlentities.zig
The bundled [`entities.json`](/entities.json) is sourced from .
Modelled on [Philip Jackson's `entities` crate](https://github.com/p-jackson/entities) for Rust.
## Overview
The core datatypes are:
```zig
pub const Entity = struct {
entity: []u8,
codepoints: Codepoints,
characters: []u8,
};
pub const Codepoints = union(enum) {
Single: u32,
Double: [2]u32,
};
```
The list of entities is directly exposed, as well as a binary search function:
```zig
pub const ENTITIES: [_]Entity
pub fn lookup(entity: []const u8) ?Entity
```
## Serving suggestion
Add it to your `build.zig.zon`:
```
zig fetch --save git+https://nossa.ee/~talya/htmlentities.zig
```
In your `build.zig`:
```zig
const htmlentities_dep = b.dependency("htmlentities.zig", .{ .target = target, .optimize = optimize });
exe.root_module.addImport("htmlentities", htmlentities_dep.module("htmlentities"));
```
In your `main.zig`:
```zig
const std = @import("std");
const htmlentities = @import("htmlentities");
pub fn main() !void {
var eacute = htmlentities.lookup("é").?;
std.debug.print("eacute: {}\n", .{eacute});
}
```
Output:
```
eacute: Entity{ .entity = é, .codepoints = Codepoints{ .Single = 233 }, .characters = é }
```
## Help wanted
Ideally we'd do the JSON parsing and struct creation at comptime. The std JSON
tokeniser uses ~80GB of RAM and millions of backtracks to handle the whole
`entities.json` at comptime, so it's not gonna happen yet. Maybe once we get a
comptime allocator we can use the regular parser.
As it is, we do codegen. Ideally we'd piece together an AST and render
that instead of just writing Zig directly -- I did try it with a 'template'
input string (see some broken wip at `63b9393`), but it's hard to do since
`std.zig.render` expects all tokens, including string literal, to be available
in the originally parsed source. At the moment we parse our generated source
and format it so we can at least validate it syntactically in the build step.