Manifest, file header, page (guides, flows), shape (geometry, fills with gradients and images, strokes, shadows, blurs, radii, constraints, flex and grid layout fields, component links, applied tokens) and the library records (colours, typographies, components, media), named as the files name their fields. Every record keeps the fields it has no name for, so a member read and written loses nothing but nulls. Archive::typed reads a member as any of them within the archive's budgets.
craft-penpot
A clean-room reader and writer for the Penpot 2.x export container: the .penpot archive
(penpot/export-files, version 1). Offline: no network, no converter, no unsafe.
This crate is the container layer. It validates the ZIP, checks the manifest, and hands out the
members as the JSON the file holds (serde_json::Value) and media objects as bytes; it writes
members back into an archive. What the shapes, components and tokens mean, and how they map onto
an application's document, stays with the caller.
Provenance
Written by FigCraft's contributors (the offline Penpot import and export, first by Guillermo
Bueno) under FigCraft's clean-room rule: behaviour comes from public documentation and from
observing files, no code of another product is copied, and no GPL or AGPL code is used.
FigCraft's own notes say its tests do not establish compatibility with a live Penpot server;
that holds here too. Moved here from figcraft/crates/format/src/penpot.rs (the archive half)
and penpot/embedded_fonts.rs.
The format
A ZIP of JSON members and blobs:
| Member | Holds |
|---|---|
manifest.json |
type (penpot/export-files), version (1), files: each with its id (a UUID) and name |
files/<file>.json |
the file's header: features, version, migrations, options |
files/<file>/pages/<page>.json |
a page: id, name, index, background, guides, flows |
files/<file>/pages/<page>/<shape>.json |
one shape; the root frame has the nil UUID (ROOT_ID) and lists its children in shapes |
files/<file>/components/<id>.json, colors/, typographies/ |
library entries |
files/<file>/tokens.json |
design tokens |
files/<file>/media/<id>.json |
a media record: mediaId names the stored object, mtype its type |
objects/<id>.json, objects/<id>.<ext> |
a stored object's record and its bytes |
Penpot 1.x binary exports are a different format and are not read.
Reading
let mut archive = craft_penpot::Archive::open(&bytes)?;
for file in archive.files.clone() {
let header = archive.json(&format!("files/{}.json", file.id)).ok();
for page in archive.pages(&file.id)? { // sorted by index
let Some(page_id) = page.get("id").and_then(|v| v.as_str()) else { continue };
let shapes = archive.page_shapes(&file.id, page_id)?; // id -> shape JSON
let root = shapes.get(craft_penpot::ROOT_ID);
}
let colors = archive.uuid_members(&format!("files/{}/colors", file.id), 2048);
let image = archive.media(&file.id, media_id, max_bytes); // Option<Vec<u8>>
}
let fonts = archive.embedded_fonts(); // counted by signature
archive.check()?; // no budget was overrun
Members a caller treats as optional return errors it may drop; check then says whether a
budget ran out along the way, so exhaustion is never swallowed.
Writing
let manifest = craft_penpot::manifest("my-app/1.0", &[(file_id, "Name", features)]);
let bytes = craft_penpot::write_archive([("manifest.json", manifest_bytes), (path, member_bytes)])?;
write_archive refuses paths Archive::open would refuse.
Limits
Input is hostile. Nothing panics; every limit has a test.
| Limit | Value |
|---|---|
The archive (MAX_ARCHIVE) |
512 MiB |
ZIP members (MAX_ENTRIES) |
100 000 |
Bytes unpacked, by declared sizes at open and by bytes actually read after (MAX_DECOMPRESSED) |
1 GiB |
A JSON member (MAX_JSON) |
16 MiB |
JSON values and keys, every member together, counted before parsing (MAX_JSON_VALUES) |
4 000 000 |
Shape JSON for one page (MAX_PAGE_SHAPES_BYTES) |
128 MiB |
Shapes on one page (MAX_PAGE_SHAPES) |
50 000 |
Pages in one file (MAX_PAGES) |
500 |
Files in the manifest (MAX_FILES) |
50 |
A media object (MAX_IMAGE; media also takes the caller's own cap) |
128 MiB |
| JSON nesting | serde_json's 128 levels |
Duplicate member names (told by the directory's own member count, since the ZIP reader keeps one
member per name), absolute paths, . and .. parts, backslashes and control characters
fail the archive. Ids are checked to be UUIDs before they are used in a member path.
ZIP
The archive is read with the zip crate directly, with the bounds above applied to each member.
A shared bounded ZIP reader is planned (craft-libs #126); this crate moves to it when it lands.
Typed members (model)
model types the members as the files name their fields: Manifest, FileHeader, Page (with
its Guides and Flows), Shape (geometry, SelRect, Matrix, points, Fills with colours,
Gradients and ImageRefs, Strokes, Shadows, Blurs, corner radii, constraints, flex and
grid layout fields, component links, applied tokens), and the library records LibraryColor,
Typography, Component and Media. ShapeType, GradientType and BlendMode keep values
they don't know (Other).
let page: craft_penpot::model::Page = archive.typed(&format!("files/{file}/pages/{page_id}.json"))?;
let shape: craft_penpot::model::Shape = archive.typed(&shape_path)?;
let bytes = serde_json::to_vec(&shape)?; // written back as the file had it
Every record keeps the fields it has no name for in extra and writes them back: a member read
and written loses only null fields, and whole numbers come back as floats. Kept as JSON: text
and path content, SVG attributes, interactions, display grids, grid tracks and cells, tokens
(tokens.json), and typography sizes (numbers or numeric strings in the files). A field of the
wrong type fails typed, naming the member; a reader that must skip bad fields one by one reads
json instead. typed reads within the same byte and value budgets as json. Shapes are flat
(children by id), so no record nests deeper than the JSON itself.
Not here yet
Typed text and path content, interactions and tokens.