use crate::http::retrieve_asset; use base64::encode; use regex::Regex; use reqwest::blocking::Client; use std::collections::HashMap; use url::{ParseError, Url}; /// This monster of a regex is used to match any kind of URL found in CSS. /// /// There are roughly three different categories that a found URL could fit /// into: /// - Font [found after a src: property in an @font-family rule] /// - Stylesheet [denoted by an @import before the url /// - Image [covers all other uses of the url() function] /// /// This regex aims to extract the following information: /// - What type of URL is it (font/image/css) /// - Where is the part that needs to be replaced (incl any wrapping quotes) /// - What is the URL (excl any wrapping quotes) /// /// Essentially, the regex can be broken down into two parts: /// /// `(?:(?P@import)|(?Psrc\s*:)\s+)?` /// This matches the precursor to a font or CSS URL, and fills in a match under /// either `` (if it's a CSS URL) or `` (if it's a font). /// Determining whether or not it's an image can be done by the negation of both /// of these. Either zero or one of these can match. /// /// `url\((?P['"]?(?P[^"'\)]+)['"]?)\)` /// This matches the actual URL part of the url(), and must always match. It also /// sets `` and `` which correspond to everything within /// `url(...)` and a usable URL, respectively. /// /// Note, however, that this does not perform any validation of the found URL. /// Malformed CSS could lead to an invalid URL being present. It is therefore /// recomended that the URL gets manually validated. const CSS_URL_REGEX_STR: &str = r###"(?:(?:(?P@import)|(?Psrc\s*:))\s+)?url\((?P['"]?(?P[^"'\)]+)['"]?)\)"###; lazy_static! { static ref HAS_PROTOCOL: Regex = Regex::new(r"^[a-z0-9]+:").unwrap(); static ref REGEX_URL: Regex = Regex::new(r"^https?://").unwrap(); static ref REGEX_CSS_URL: Regex = Regex::new(CSS_URL_REGEX_STR).unwrap(); } const MAGIC: [[&[u8]; 2]; 19] = [ // Image [b"GIF87a", b"image/gif"], [b"GIF89a", b"image/gif"], [b"\xFF\xD8\xFF", b"image/jpeg"], [b"\x89PNG\x0D\x0A\x1A\x0A", b"image/png"], [b" String { let mimetype = if mime.is_empty() { detect_mimetype(data) } else { mime.to_string() }; format!("data:{};base64,{}", mimetype, encode(data)) } pub fn detect_mimetype(data: &[u8]) -> String { for item in MAGIC.iter() { if data.starts_with(item[0]) { return String::from_utf8(item[1].to_vec()).unwrap(); } } "".to_owned() } pub fn url_has_protocol>(url: T) -> bool { HAS_PROTOCOL.is_match(url.as_ref().to_lowercase().as_str()) } pub fn is_data_url>(url: T) -> Result { Url::parse(url.as_ref()).and_then(|u| Ok(u.scheme() == "data")) } pub fn is_http_url>(path: T) -> bool { REGEX_URL.is_match(path.as_ref()) } pub fn resolve_url, U: AsRef>(from: T, to: U) -> Result { let result = if is_http_url(to.as_ref()) { to.as_ref().to_string() } else { Url::parse(from.as_ref())? .join(to.as_ref())? .as_ref() .to_string() }; Ok(result) } pub fn resolve_css_imports( cache: &mut HashMap, client: &Client, css_string: &str, as_data_url: bool, href: &str, opt_no_images: bool, opt_silent: bool, ) -> String { let mut resolved_css = String::from(css_string); for link in REGEX_CSS_URL.captures_iter(&css_string) { let target_link = link.name("url").unwrap().as_str(); // Determine the type of link let is_stylesheet = link.name("stylesheet").is_some(); let is_font = link.name("font").is_some(); let is_image = !is_stylesheet && !is_font; // Generate absolute URL for content let embedded_url = match resolve_url(href, target_link) { Ok(url) => url, Err(_) => continue, // Malformed URL }; // Download the asset. If it's more CSS, resolve that too let content = if is_stylesheet { // The link is an @import link retrieve_asset( cache, client, &embedded_url, false, // Formating as data URL will be done later "text/css", // Expect CSS opt_silent, ) .map(|(content, _)| { resolve_css_imports( cache, client, &content, true, // Finally, convert to a data URL &embedded_url, opt_no_images, opt_silent, ) }) } else if (is_image && !opt_no_images) || is_font { // The link is some other, non-@import link retrieve_asset( cache, client, &embedded_url, true, // Format as data URL "", // Unknown MIME type opt_silent, ) .map(|(a, _)| a) } else { // If it's a datatype that has been opt_no'd out of, replace with // absolute URL Ok(embedded_url.clone()) } .unwrap_or_else(|e| { eprintln!("Warning: {}", e); // If failed to resolve, replace with absolute URL embedded_url }); let replacement = format!("\"{}\"", &content); let dest = link.name("to_repl").unwrap(); let offset = resolved_css.len() - css_string.len(); let target_range = (dest.start() + offset)..(dest.end() + offset); resolved_css.replace_range(target_range, &replacement); } if as_data_url { data_to_data_url("text/css", resolved_css.as_bytes()) } else { resolved_css } } pub fn clean_url>(url: T) -> String { let mut result = Url::parse(url.as_ref()).unwrap(); // Clear fragment result.set_fragment(None); // Get rid of stray question mark if result.query() == Some("") { result.set_query(None); } result.to_string() }