fif/src/mime_db.rs

// SPDX-FileCopyrightText: 2021-2022 Lynnesbian
// SPDX-License-Identifier: GPL-3.0-or-later

//! Backend-neutral Mime database abstraction.

use cfg_if::cfg_if;
use mime::Mime;

/// A thin wrapper around either [`Infer`] or [`xdg-mime::SharedMimeInfo`], depending on which [cargo features]
///  fif was compiled with. By default, fif uses an [`Infer`]-based implementation on Windows, and an
/// [`xdg-mime`]-based one everywhere else. This behaviour can be changed at compile time by using the aforementioned
/// [cargo features].
///
/// [cargo features]: https://gitlab.com/Lynnesbian/fif/-/wikis/Cargo-Features
/// [`Infer`]: https://docs.rs/infer/
/// [`xdg-mime::SharedMimeInfo`]: https://docs.rs/xdg-mime/0/xdg_mime/struct.SharedMimeInfo.html
/// [`xdg-mime`]: https://docs.rs/xdg-mime/
pub trait MimeDb {
	/// Initialise the database.
	fn init() -> Self;
	/// Given a slice of bytes, returns the inferred MIME type, if any.
	fn get_type(&self, data: &[u8]) -> Option<Mime>;
}

cfg_if! {
	if #[cfg(any(all(unix, feature = "infer-backend"), all(not(unix), not(feature = "xdg-mime-backend"))))] {
		use std::str::FromStr;

		/// The [`Infer`](https://docs.rs/infer/)-based implementation of [`MimeDb`].
		pub struct InferDb {
			db: infer::Infer,
		}

		fn open_document_check(buf: &[u8], kind: &str) -> bool {
			let mime = format!("application/vnd.oasis.opendocument.{}", kind);
			let mime = mime.as_bytes();

			buf.len() > 38 + mime.len() && buf.starts_with(b"PK\x03\x04") && buf[38..mime.len() + 38] == mime[..]
		}

		impl MimeDb for InferDb {
			fn init() -> Self {
				let mut info = infer::Infer::new();

				// In addition to the file inferences provided by Infer, I've also added a few of my own below. Some of them
				// replace Infer's existing ones, some of them are less than perfect, and still others are for relatively
				// obscure formats, so I'm not really sure whether or not they should be contributed upstream.

				// OpenDocument Text (used by e.g. LibreOffice Writer)
				info.add("application/vnd.oasis.opendocument.text", "odt", |buf| {
					open_document_check(buf, "text")
				});

				// OpenDocument Spreadsheet (LibreOffice Calc)
				info.add("application/vnd.oasis.opendocument.spreadsheet", "ods", |buf| {
					open_document_check(buf, "spreadsheet")
				});

				// OpenOffice Presentation (LibreOffice Impress)
				info.add("application/vnd.oasis.opendocument.presentation", "odp", |buf| {
					open_document_check(buf, "presentation")
				});

				// Ren'Py Archive (Ren'Py: https://www.renpy.org/)
				info.add("application/x-rpa", "rpa", |buf| {
					buf.len() >= 34 && buf.starts_with(b"RPA-") && buf[7] == b' ' && buf[24] ==b' '
				});

				// Mach-O Binaries (The executable format used by macOS)
				// my source for most of this info is this article: https://h3adsh0tzz.com/2020/01/macho-file-format/
				info.add("application/x-mach-binary", "macho", |buf| {
					// a 32-bit mach-o header occupies 28 bits of space, so any input smaller than that cannot be a mach-o
					// binary, even if it starts with the magic numbers.

					// java class files also start with 0xCAFEBABE. since infer doesn't support detecting these files,
					// collisions are not an issue. if, however, infer does gain support for identifying java class files, the
					// 0xCAFEBABE check should be removed, as java bytecode files are far more prevalent than 32-bit universal
					// mach-o binaries [citation needed].

					// check for magic numbers (0xFEEDCACF, 0xFEEDFACE, 0xCAFEBABE) in both big and little endian forms
					buf.len() >= 28 && [b"\xFE\xED\xFA\xCF", b"\xFE\xED\xFA\xCE", b"\xCA\xFE\xBA\xBE", b"\xCF\xFA\xED\xFE",
						b"\xCE\xFA\xED\xFE", b"\xBE\xBA\xFE\xCA"].iter().any(|magic_numbers| buf.starts_with(&magic_numbers[..]))
				});

				// info.add("application/x-msi", "msi", |buf| {
				// TODO: find a way to detect MSI files properly - this just detects those weird windows OLE files and therefore
				// also picks up on .doc files
				// 	buf.starts_with(b"\xd0\xcf\x11\xe0\xa1\xb1\x1a\xe1")
				// });

				// Scalable Vector Graphics
				info.add("image/svg+xml", "svg", |buf| {
					// before doing the moderately expensive SVG check, we should make sure that the input is actually SGML-ish,
					// by which i mean, starts with the pattern "\s*<".

					// so, here comes our fancy pants """""SGML-ish validator"""""
					for c in buf {
						match c {
							// whitespace (according to https://www.w3.org/TR/xml/#NT-S)
							b'\t' | b'\r' | b'\n' | b'\x20' => continue,
							b'<' => break,
							_ => return false,
						}
					}

					// finally, to check whether or not the file is an SVG:
					// - split the buffer up into chunks separated by the less than sign
					// - check to see if this chunk starts with any of these identifiers:
					let identifiers: Vec<&[u8]> = vec![b"svg", b"SVG", b"!DOCTYPE svg", b"!DOCTYPE SVG"];
					// - if it does, the nested `any` will short circuit and immediately return true, causing the parent `any` to
					//   do the same
					// - and finally, if none of the chunks match, we'll return false

					// TODO: this is kind of messy, i'd like to clean it up somehow :(
					buf
						.split(|c| *c == b'<')
						.any(|buf| identifiers.iter().any(|id| buf.starts_with(id)))
				});

				Self { db: info }
			}

			fn get_type(&self, data: &[u8]) -> Option<Mime> {
				if let Some(mime) = self.db.get(data) {
					match Mime::from_str(mime.mime_type()) {
						Err(_) => None,
						Ok(m) => Some(m),
					}
				} else { None }
			}
		}
	} else {
		/// The [`xdg-mime`](https://docs.rs/xdg-mime/)-based implementation of [`MimeDb`].
		pub struct XdgDb {
			db: xdg_mime::SharedMimeInfo,
		}

		impl MimeDb for XdgDb {
			fn init() -> Self {
				Self { db: xdg_mime::SharedMimeInfo::new() }
			}

			fn get_type(&self, data: &[u8]) -> Option<Mime> {
				self.db.get_mime_type_for_data(data).map(|m| m.0)
			}
		}
	}
}