#[macro_use] extern crate lazy_static; extern crate regex; extern crate image; extern crate rayon; extern crate twox_hash; extern crate utils; extern crate errors; use std::path::{Path, PathBuf}; use std::hash::{Hash, Hasher}; use std::collections::HashMap; use std::collections::hash_map::Entry as HEntry; use std::fs::{self, File}; use regex::Regex; use image::{GenericImage, FilterType}; use image::jpeg::JPEGEncoder; use rayon::prelude::*; use twox_hash::XxHash; use utils::fs as ufs; use errors::{Result, ResultExt}; static RESIZED_SUBDIR: &'static str = "_resized_images"; lazy_static!{ pub static ref RESIZED_FILENAME: Regex = Regex::new(r#"([0-9a-f]{16})([0-9a-f]{2})[.]jpg"#).unwrap(); } /// Describes the precise kind of a resize operation #[derive(Debug, Clone, Copy, PartialEq, Eq)] pub enum ResizeOp { /// A simple scale operation that doesn't take aspect ratio into account Scale(u32, u32), /// Scales the image to a specified width with height computed such that aspect ratio is preserved FitWidth(u32), /// Scales the image to a specified height with width computed such that aspect ratio is preserved FitHeight(u32), /// Scales the image such that it fits within the specified width and height preserving aspect ratio. /// Either dimension may end up being smaller, but never larger than specified. Fit(u32, u32), /// Scales the image such that it fills the specified width and height. Output will always have the exact dimensions specified. /// The part of the image that doesn't fit in the thumbnail due to differing aspect ratio will be cropped away, if any. Fill(u32, u32), } impl ResizeOp { pub fn from_args(op: &str, width: Option, height: Option) -> Result { use ResizeOp::*; // Validate args: match op { "fit_width" => if width.is_none() { return Err(format!("op=\"fit_width\" requires a `width` argument").into()) }, "fit_height" => if height.is_none() { return Err(format!("op=\"fit_height\" requires a `height` argument").into()) }, "scale" | "fit" | "fill" => if width.is_none() || height.is_none() { return Err(format!("op={} requires a `width` and `height` argument", op).into()) }, _ => return Err(format!("Invalid image resize operation: {}", op).into()) }; Ok(match op { "scale" => Scale(width.unwrap(), height.unwrap()), "fit_width" => FitWidth(width.unwrap()), "fit_height" => FitHeight(height.unwrap()), "fit" => Fit(width.unwrap(), height.unwrap()), "fill" => Fill(width.unwrap(), height.unwrap()), _ => unreachable!(), }) } pub fn width(self) -> Option { use ResizeOp::*; match self { Scale(w, _) => Some(w), FitWidth(w) => Some(w), FitHeight(_) => None, Fit(w, _) => Some(w), Fill(w, _) => Some(w), } } pub fn height(self) -> Option { use ResizeOp::*; match self { Scale(_, h) => Some(h), FitWidth(_) => None, FitHeight(h) => Some(h), Fit(_, h) => Some(h), Fill(_, h) => Some(h), } } } impl From for u8 { fn from(op: ResizeOp) -> u8 { use ResizeOp::*; match op { Scale(_, _) => 1, FitWidth(_) => 2, FitHeight(_) => 3, Fit(_, _) => 4, Fill(_, _) => 5, } } } impl Hash for ResizeOp { fn hash(&self, hasher: &mut H) { hasher.write_u8(u8::from(*self)); if let Some(w) = self.width() { hasher.write_u32(w); } if let Some(h) = self.height() { hasher.write_u32(h); } } } /// Holds all data needed to perform a resize operation #[derive(Debug, PartialEq, Eq)] pub struct ImageOp { source: String, op: ResizeOp, quality: u8, /// Hash of the above parameters hash: u64, collision: Option, } impl ImageOp { pub fn new(source: String, op: ResizeOp, quality: u8) -> ImageOp { let mut hasher = XxHash::with_seed(0); hasher.write(source.as_ref()); op.hash(&mut hasher); hasher.write_u8(quality); let hash = hasher.finish(); ImageOp { source, op, quality, hash, collision: None } } pub fn from_args(source: String, op: &str, width: Option, height: Option, quality: u8) -> Result { let op = ResizeOp::from_args(op, width, height)?; Ok(Self::new(source, op, quality)) } fn num_collisions(&self) -> u32 { self.collision.unwrap_or(0) } fn perform(&self, content_path: &Path, target_path: &Path) -> Result<()> { use ResizeOp::*; let src_path = content_path.join(&self.source); if !ufs::file_stale(&src_path, target_path) { return Ok(()) } let mut img = image::open(&src_path)?; let (img_w, img_h) = img.dimensions(); const RESIZE_FILTER: FilterType = FilterType::Gaussian; const RATIO_EPSILLION: f32 = 0.1; let img = match self.op { Scale(w, h) => img.resize_exact(w, h, RESIZE_FILTER), FitWidth(w) => img.resize(w, u32::max_value(), RESIZE_FILTER), FitHeight(h) => img.resize(u32::max_value(), h, RESIZE_FILTER), Fit(w, h) => img.resize(w, h, RESIZE_FILTER), Fill(w, h) => { let factor_w = img_w as f32 / w as f32; let factor_h = img_h as f32 / h as f32; if (factor_w - factor_h).abs() <= RATIO_EPSILLION { // If the horizontal and vertical factor is very similar, that means the aspect is similar enough // that there's not much point in cropping, so just perform a simple scale in this case. img.resize_exact(w, h, RESIZE_FILTER) } else { // We perform the fill such that a crop is performed first and then resize_exact can be used, // which should be cheaper than resizing and then cropping (smaller number of pixels to resize). let (crop_w, crop_h) = if factor_w < factor_h { (img_w, (factor_w * h as f32).round() as u32) } else { ((factor_h * w as f32).round() as u32, img_h) }; let (offset_w, offset_h) = if factor_w < factor_h { (0, (img_h - crop_h) / 2) } else { ((img_w - crop_w) / 2, 0) }; img.crop(offset_w, offset_h, crop_w, crop_h).resize_exact(w, h, RESIZE_FILTER) } }, }; let mut f = File::create(target_path)?; let mut enc = JPEGEncoder::new_with_quality(&mut f, self.quality); let (img_w, img_h) = img.dimensions(); enc.encode(&img.raw_pixels(), img_w, img_h, img.color())?; Ok(()) } } /// A strcture into which image operations can be enqueued and then performed. /// All output is written in a subdirectory in `static_path`, /// taking care of file stale status based on timestamps and possible hash collisions. #[derive(Debug)] pub struct Processor { content_path: PathBuf, resized_path: PathBuf, resized_url: String, /// A map of a ImageOps by their stored hash. /// Note that this cannot be a HashSet, because hashest handles collisions and we don't want that, /// we need to be aware of and handle collisions ourselves. img_ops: HashMap, /// Hash collisions go here: img_ops_collisions: Vec, } impl Processor { pub fn new(content_path: PathBuf, static_path: &Path, base_url: &str) -> Processor { Processor { content_path, resized_path: static_path.join(RESIZED_SUBDIR), resized_url: Self::resized_url(base_url), img_ops: HashMap::new(), img_ops_collisions: Vec::new(), } } fn resized_url(base_url: &str) -> String { if base_url.ends_with('/') { format!("{}{}", base_url, RESIZED_SUBDIR) } else { format!("{}/{}", base_url, RESIZED_SUBDIR) } } pub fn set_base_url(&mut self, base_url: &str) { self.resized_url = Self::resized_url(base_url); } pub fn source_exists(&self, source: &str) -> bool { self.content_path.join(source).exists() } pub fn num_img_ops(&self) -> usize { self.img_ops.len() + self.img_ops_collisions.len() } fn insert_with_collisions(&mut self, mut img_op: ImageOp) -> u32 { match self.img_ops.entry(img_op.hash) { HEntry::Occupied(entry) => if *entry.get() == img_op { return 0; }, HEntry::Vacant(entry) => { entry.insert(img_op); return 0; }, } // If we get here, that means a hash collision. // This is detected when there is an ImageOp with the same hash in the `img_ops` map but which is not equal to this one. // To deal with this, all collisions get a (random) sequential ID number. // First try to look up this ImageOp in `img_ops_collisions`, maybe we've already seen the same ImageOp let mut num = 1; for op in self.img_ops_collisions.iter().filter(|op| op.hash == img_op.hash) { if *op == img_op { // This is a colliding ImageOp, but we've already seen the very same one, so just return its ID return num; } else { num += 1; } } // If we get here, that means this is a new colliding ImageOp and `num` has the next free ID if num == 1 { self.img_ops.get_mut(&img_op.hash).unwrap().collision = Some(0); } img_op.collision = Some(num); self.img_ops_collisions.push(img_op); num } fn op_filename(hash: u64, num_collisions: u32) -> String { // Please keep this in sync with RESIZED_FILENAME assert!(num_collisions < 256, "Unexpectedly large number of collisions: {}", num_collisions); format!("{:016x}{:02x}.jpg", hash, num_collisions) } fn op_url(&self, hash: u64, num_collisions: u32) -> String { format!("{}/{}", &self.resized_url, Self::op_filename(hash, num_collisions)) } pub fn insert(&mut self, img_op: ImageOp) -> String { let hash = img_op.hash; let num_collisions = self.insert_with_collisions(img_op); self.op_url(hash, num_collisions) } pub fn prune(&self) -> Result<()> { ufs::ensure_directory_exists(&self.resized_path)?; let entries = fs::read_dir(&self.resized_path)?; for entry in entries { let entry_path = entry?.path(); if entry_path.is_file() { let filename = entry_path.file_name().unwrap().to_string_lossy(); if let Some(capts) = RESIZED_FILENAME.captures(filename.as_ref()) { let hash = u64::from_str_radix(capts.get(1).unwrap().as_str(), 16).unwrap(); let num_collisions = u32::from_str_radix(capts.get(2).unwrap().as_str(), 16).unwrap(); if num_collisions > 0 || !self.img_ops.contains_key(&hash) { fs::remove_file(&entry_path)?; } } } } Ok(()) } pub fn do_process(&mut self) -> Result<()> { self.img_ops.par_iter().map(|(hash, op)| { let target = self.resized_path.join(Self::op_filename(*hash, op.num_collisions())); op.perform(&self.content_path, &target) .chain_err(|| format!("Failed to process image: {}", op.source)) }) .fold(|| Ok(()), Result::and) .reduce(|| Ok(()), Result::and) } } /// Looks at file's extension and returns whether it's a supported image format pub fn file_is_img>(p: P) -> bool { p.as_ref().extension().and_then(|s| s.to_str()).map(|ext| { match ext.to_lowercase().as_str() { "jpg" | "jpeg" => true, "png" => true, "gif" => true, "bmp" => true, _ => false, } }).unwrap_or(false) }