Zigshape: a sample-to-struct generator for serde.zig (and plain std.json)

Showcase
,
1

Hi all,

I built a small web tool that generates Zig structs from JSON/YAML/TOML/XML samples. It’s a companion to serde.zig, but it can also emit plain structs for std.json users.

Two output targets:

  • serde.zig: a struct with a pub const serde = .{ ... } block, ready to drop in.
  • plain: a regular struct usable with std.json.parseFromSlice.

Example input:

​json {"userId":1,"firstName":"Alice","email":"alice@example.com","lastSeenAt":"2026-01-01T10:00:00Z"} ​

Output, serde.zig target:

const serde = @import("serde");

pub const User = struct {
    id: u8,
    name: []const u8,
    roles: []const []const u8,
    profile: Profile,
};

pub const Profile = struct {
    created_at: []const u8,
    active: bool,

    pub const serde = .{
        .rename_all = serde.NamingConvention.camel_case,
    };
};

What it infers from the samples:

  • smallest int width that covers the observed values (u8 here),
  • optional fields when a key is missing or null in any sample,
  • rename_all when one naming convention round-trips every field, otherwise per-field rename,
  • given several samples, sibling fields that share a shape and never co-occur get folded under alias,
  • deny_unknown_fields is one toggle away in the advanced panel.

Everything runs in the browser (cloudflare pages hosted), samples are not uploaded anywhere.

Would appreciate feedback on the inference defaults and anything that produces wrong output. XML in particular still has rough corners around attribute vs. element ambiguity, so real-world payloads that break it would help.

2 Likes
2

This seems pretty cool! Is the source for the site available?

3

Sure GitHub - OrlovEvgeny/zigshape: Zig schema inference and struct generator for real-world data formats · GitHub

1 Like