Image transformation

crates/pile-dataset/Cargo.toml

@@ -24,10 +24,17 @@ tokio-stream = { workspace = true }
 
 serde = { workspace = true, optional = true }
 axum = { workspace = true, optional = true }
+percent-encoding = { workspace = true, optional = true }
 utoipa = { workspace = true, optional = true }
 utoipa-swagger-ui = { workspace = true, optional = true }
 
 [features]
 default = []
 pdfium = ["pile-value/pdfium"]
-axum = ["dep:axum", "dep:utoipa", "dep:utoipa-swagger-ui", "dep:serde"]
+axum = [
+	"dep:axum",
+	"dep:utoipa",
+	"dep:utoipa-swagger-ui",
+	"dep:serde",
+	"dep:percent-encoding",
+]

crates/pile-dataset/src/serve/extract.rs

@@ -4,6 +4,7 @@ use axum::{
 	http::{StatusCode, header},
 	response::{IntoResponse, Response},
 };
+use percent_encoding::percent_decode_str;
 use pile_config::{Label, objectpath::ObjectPath};
 use pile_value::{extract::traits::ExtractState, value::PileValue};
 use serde::Deserialize;
@@ -61,6 +62,7 @@ pub async fn get_extract(
 			if let Some((k, v)) = part.split_once('=')
 				&& k == "path"
 			{
+				let v = percent_decode_str(v).decode_utf8_lossy();
 				match v.parse::<ObjectPath>() {
 					Ok(p) => result.push(p),
 					Err(e) => {

crates/pile-value/Cargo.toml

@@ -27,13 +27,15 @@ epub = { workspace = true }
 kamadak-exif = { workspace = true }
 pdf = { workspace = true }
 pdfium-render = { workspace = true, optional = true }
-image = { workspace = true, optional = true }
+image = { workspace = true }
 id3 = { workspace = true }
 tokio = { workspace = true }
 async-trait = { workspace = true }
 mime = { workspace = true }
 mime_guess = { workspace = true }
+serde = { workspace = true }
+strum = { workspace = true }
 
 [features]
 default = []
-pdfium = ["dep:pdfium-render", "dep:image"]
+pdfium = ["dep:pdfium-render"]

crates/pile-value/src/extract/item/image/mod.rs

@@ -0,0 +1,131 @@
+mod transform;
+use transform::{CropTransformer, ImageTransformer, MaxDimTransformer};
+
+use image::ImageFormat;
+use mime::Mime;
+use pile_config::Label;
+use pile_io::AsyncReader;
+use std::{
+	io::Cursor,
+	str::FromStr,
+	sync::{Arc, OnceLock},
+};
+use tracing::trace;
+
+use crate::{
+	extract::traits::{ExtractState, ObjectExtractor},
+	value::{Item, PileValue},
+};
+
+enum ImageSource {
+	Item(Item, OnceLock<Arc<Vec<u8>>>),
+	Blob(Arc<Vec<u8>>, Mime),
+}
+
+pub struct ImageExtractor {
+	source: ImageSource,
+}
+
+impl ImageExtractor {
+	pub fn new(item: &Item) -> Self {
+		Self {
+			source: ImageSource::Item(item.clone(), OnceLock::new()),
+		}
+	}
+
+	pub fn from_blob(bytes: Arc<Vec<u8>>, mime: Mime) -> Self {
+		Self {
+			source: ImageSource::Blob(bytes, mime),
+		}
+	}
+
+	fn mime(&self) -> &Mime {
+		match &self.source {
+			ImageSource::Item(item, _) => item.mime(),
+			ImageSource::Blob(_, mime) => mime,
+		}
+	}
+
+	async fn read_bytes(&self) -> Result<Arc<Vec<u8>>, std::io::Error> {
+		match &self.source {
+			ImageSource::Blob(bytes, _) => Ok(bytes.clone()),
+			ImageSource::Item(item, cache) => {
+				if let Some(x) = cache.get() {
+					return Ok(x.clone());
+				}
+				let mut reader = item.read().await?;
+				let bytes = reader.read_to_end().await?;
+				Ok(cache.get_or_init(|| Arc::new(bytes)).clone())
+			}
+		}
+	}
+
+	async fn apply<T: ImageTransformer + Send + 'static>(
+		&self,
+		args: &str,
+	) -> Result<Option<PileValue>, std::io::Error> {
+		let transformer = match T::parse_args(args) {
+			Ok(t) => t,
+			Err(_) => return Ok(None),
+		};
+
+		let mime = self.mime().clone();
+		let bytes = self.read_bytes().await?;
+
+		let Some(format) = ImageFormat::from_mime_type(&mime) else {
+			return Ok(Some(PileValue::Blob { mime, bytes }));
+		};
+
+		let bytes_for_closure = bytes.clone();
+		let result = tokio::task::spawn_blocking(move || {
+			let mut img = image::load_from_memory_with_format(&bytes_for_closure, format)?;
+			transformer.transform(&mut img);
+
+			let mut out = Cursor::new(Vec::new());
+			img.write_to(&mut out, format)?;
+
+			let out_mime =
+				Mime::from_str(format.to_mime_type()).unwrap_or(mime::APPLICATION_OCTET_STREAM);
+			Ok::<_, image::ImageError>((out_mime, out.into_inner()))
+		})
+		.await?;
+
+		match result {
+			Ok((out_mime, out_bytes)) => Ok(Some(PileValue::Blob {
+				mime: out_mime,
+				bytes: Arc::new(out_bytes),
+			})),
+			Err(_) => Ok(Some(PileValue::Blob { mime, bytes })),
+		}
+	}
+}
+
+#[async_trait::async_trait]
+impl ObjectExtractor for ImageExtractor {
+	async fn field(
+		&self,
+		_state: &ExtractState,
+		name: &Label,
+		args: Option<&str>,
+	) -> Result<Option<PileValue>, std::io::Error> {
+		let Some(args) = args else {
+			return Ok(None);
+		};
+
+		trace!(?args, "Getting field {name:?} from ImageExtractor",);
+
+		match name.as_str() {
+			"maxdim" => self.apply::<MaxDimTransformer>(args).await,
+			"crop" => self.apply::<CropTransformer>(args).await,
+			_ => Ok(None),
+		}
+	}
+
+	#[expect(clippy::unwrap_used)]
+	async fn fields(&self) -> Result<Vec<Label>, std::io::Error> {
+		Ok(vec![
+			Label::new("maxdim").unwrap(),
+			Label::new("crop").unwrap(),
+		])
+	}
+}

crates/pile-value/src/extract/item/image/transform/mod.rs

@@ -0,0 +1,4 @@
+mod pixeldim;
+
+pub mod transformers;
+pub use transformers::{CropTransformer, ImageTransformer, MaxDimTransformer};

crates/pile-value/src/extract/item/image/transform/pixeldim.rs

@@ -0,0 +1,68 @@
+use serde::{Deserialize, Deserializer};
+use std::fmt;
+use std::str::FromStr;
+
+// TODO: parse -, + (100vw - 10px)
+// TODO: parse 100vw [min] 10
+// TODO: parse 100vw [max] 10
+
+#[derive(Debug, Clone, PartialEq)]
+pub enum PixelDim {
+	Pixels(u32),
+	WidthPercent(f32),
+	HeightPercent(f32),
+}
+
+impl FromStr for PixelDim {
+	type Err = String;
+
+	fn from_str(s: &str) -> Result<Self, Self::Err> {
+		let numeric_end = s.find(|c: char| !c.is_ascii_digit() && c != '.');
+
+		let (quantity, unit) = numeric_end.map(|x| s.split_at(x)).unwrap_or((s, "px"));
+		let quantity = quantity.trim();
+		let unit = unit.trim();
+
+		match unit {
+			"vw" => Ok(PixelDim::WidthPercent(
+				quantity
+					.parse()
+					.map_err(|_err| format!("invalid quantity {quantity}"))?,
+			)),
+
+			"vh" => Ok(PixelDim::HeightPercent(
+				quantity
+					.parse()
+					.map_err(|_err| format!("invalid quantity {quantity}"))?,
+			)),
+
+			"px" => Ok(PixelDim::Pixels(
+				quantity
+					.parse()
+					.map_err(|_err| format!("invalid quantity {quantity}"))?,
+			)),
+
+			_ => Err(format!("invalid unit {unit}")),
+		}
+	}
+}
+
+impl<'de> Deserialize<'de> for PixelDim {
+	fn deserialize<D>(deserializer: D) -> Result<Self, D::Error>
+	where
+		D: Deserializer<'de>,
+	{
+		let s = String::deserialize(deserializer)?;
+		FromStr::from_str(&s).map_err(serde::de::Error::custom)
+	}
+}
+
+impl fmt::Display for PixelDim {
+	fn fmt(&self, f: &mut fmt::Formatter<'_>) -> fmt::Result {
+		match self {
+			PixelDim::Pixels(px) => write!(f, "{px}"),
+			PixelDim::WidthPercent(p) => write!(f, "{p:.2}vw"),
+			PixelDim::HeightPercent(p) => write!(f, "{p:.2}vh"),
+		}
+	}
+}

crates/pile-value/src/extract/item/image/transform/transformers/crop.rs

@@ -0,0 +1,188 @@
+use image::DynamicImage;
+use serde::{Deserialize, Serialize};
+use std::{fmt::Display, str::FromStr};
+use strum::{Display, EnumString};
+
+use super::super::{pixeldim::PixelDim, transformers::ImageTransformer};
+
+#[derive(Debug, Clone, Copy, PartialEq, Eq, EnumString, Serialize, Deserialize, Display)]
+pub enum Direction {
+	#[serde(rename = "n")]
+	#[strum(to_string = "n")]
+	#[strum(serialize = "north")]
+	North,
+
+	#[serde(rename = "e")]
+	#[strum(serialize = "e")]
+	#[strum(serialize = "east")]
+	East,
+
+	#[serde(rename = "s")]
+	#[strum(serialize = "s")]
+	#[strum(serialize = "south")]
+	South,
+
+	#[serde(rename = "w")]
+	#[strum(to_string = "w")]
+	#[strum(serialize = "west")]
+	West,
+
+	#[serde(rename = "c")]
+	#[strum(serialize = "c")]
+	#[strum(serialize = "center")]
+	Center,
+
+	#[serde(rename = "ne")]
+	#[strum(serialize = "ne")]
+	#[strum(serialize = "northeast")]
+	NorthEast,
+
+	#[serde(rename = "se")]
+	#[strum(serialize = "se")]
+	#[strum(serialize = "southeast")]
+	SouthEast,
+
+	#[serde(rename = "nw")]
+	#[strum(serialize = "nw")]
+	#[strum(serialize = "northwest")]
+	NorthWest,
+
+	#[serde(rename = "sw")]
+	#[strum(serialize = "sw")]
+	#[strum(serialize = "southwest")]
+	SouthWest,
+}
+
+/// Crop an image to (at most) the given size.
+/// See [Self::new] for details.
+#[derive(Debug, Clone, PartialEq)]
+pub struct CropTransformer {
+	w: PixelDim,
+	h: PixelDim,
+	float: Direction,
+}
+
+impl CropTransformer {
+	/// Create a new [CropTransformer] with the given parameters.
+	///
+	/// A [CropTransformer] creates an image of size `w x h`, but...
+	/// - does not reduce width if `w` is greater than image width
+	/// - does not reduce height if `h` is greater than image height
+	/// - does nothing if `w` or `h` is less than or equal to zero.
+	#[expect(dead_code)]
+	pub fn new(w: PixelDim, h: PixelDim, float: Direction) -> Self {
+		Self { w, h, float }
+	}
+
+	fn crop_dim(&self, img_width: u32, img_height: u32) -> (u32, u32) {
+		let crop_width = match self.w {
+			PixelDim::Pixels(w) => w,
+			PixelDim::WidthPercent(pct) => ((img_width as f32) * pct / 100.0) as u32,
+			PixelDim::HeightPercent(pct) => ((img_height as f32) * pct / 100.0) as u32,
+		};
+
+		let crop_height = match self.h {
+			PixelDim::Pixels(h) => h,
+			PixelDim::WidthPercent(pct) => ((img_width as f32) * pct / 100.0) as u32,
+			PixelDim::HeightPercent(pct) => ((img_height as f32) * pct / 100.0) as u32,
+		};
+
+		(crop_width, crop_height)
+	}
+
+	#[expect(clippy::integer_division)]
+	fn crop_pos(
+		&self,
+		img_width: u32,
+		img_height: u32,
+		crop_width: u32,
+		crop_height: u32,
+	) -> (u32, u32) {
+		match self.float {
+			Direction::North => {
+				let x = (img_width - crop_width) / 2;
+				let y = 0;
+				(x, y)
+			}
+			Direction::East => {
+				let x = img_width - crop_width;
+				let y = (img_height - crop_height) / 2;
+				(x, y)
+			}
+			Direction::South => {
+				let x = (img_width - crop_width) / 2;
+				let y = img_height - crop_height;
+				(x, y)
+			}
+			Direction::West => {
+				let x = 0;
+				let y = (img_height - crop_height) / 2;
+				(x, y)
+			}
+			Direction::Center => {
+				let x = (img_width - crop_width) / 2;
+				let y = (img_height - crop_height) / 2;
+				(x, y)
+			}
+			Direction::NorthEast => {
+				let x = img_width - crop_width;
+				let y = 0;
+				(x, y)
+			}
+			Direction::SouthEast => {
+				let x = img_width - crop_width;
+				let y = img_height - crop_height;
+				(x, y)
+			}
+			Direction::NorthWest => {
+				let x = 0;
+				let y = 0;
+				(x, y)
+			}
+			Direction::SouthWest => {
+				let x = 0;
+				let y = img_height - crop_height;
+				(x, y)
+			}
+		}
+	}
+}
+
+impl Display for CropTransformer {
+	fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+		write!(f, "crop({},{},{})", self.w, self.h, self.float)
+	}
+}
+
+impl ImageTransformer for CropTransformer {
+	fn parse_args(args: &str) -> Result<Self, String> {
+		let args: Vec<&str> = args.split(",").collect();
+		if args.len() != 3 {
+			return Err(format!("expected 3 args, got {}", args.len()));
+		}
+
+		let w = args[0].trim().parse::<PixelDim>()?;
+		let h = args[1].trim().parse::<PixelDim>()?;
+
+		let direction = args[2].trim();
+		let direction = Direction::from_str(direction)
+			.map_err(|_err| format!("invalid direction {direction}"))?;
+
+		Ok(Self {
+			w,
+			h,
+			float: direction,
+		})
+	}
+
+	fn transform(&self, input: &mut DynamicImage) {
+		let (img_width, img_height) = (input.width(), input.height());
+		let (crop_width, crop_height) = self.crop_dim(img_width, img_height);
+
+		if (crop_width < img_width || crop_height < img_height) && crop_width > 0 && crop_height > 0
+		{
+			let (x, y) = self.crop_pos(img_width, img_height, crop_width, crop_height);
+			*input = input.crop(x, y, crop_width, crop_height);
+		}
+	}
+}

crates/pile-value/src/extract/item/image/transform/transformers/maxdim.rs

@@ -0,0 +1,87 @@
+use image::{DynamicImage, imageops::FilterType};
+use std::fmt::Display;
+
+use super::super::{pixeldim::PixelDim, transformers::ImageTransformer};
+
+/// Scale an image until it fits in a configured bounding box.
+#[derive(Debug, Clone, PartialEq)]
+pub struct MaxDimTransformer {
+	w: PixelDim,
+	h: PixelDim,
+}
+
+impl MaxDimTransformer {
+	/// Create a new [MaxDimTransformer] that scales an image down
+	/// until it fits in a box of dimension `w x h`.
+	///
+	/// Images are never scaled up.
+	#[expect(dead_code)]
+	pub fn new(w: PixelDim, h: PixelDim) -> Self {
+		Self { w, h }
+	}
+
+	fn target_dim(&self, img_width: u32, img_height: u32) -> (u32, u32) {
+		let max_width = match self.w {
+			PixelDim::Pixels(w) => Some(w),
+			PixelDim::WidthPercent(pct) => Some(((img_width as f32) * pct / 100.0) as u32),
+			PixelDim::HeightPercent(_) => None,
+		};
+
+		let max_height = match self.h {
+			PixelDim::Pixels(h) => Some(h),
+			PixelDim::HeightPercent(pct) => Some(((img_height as f32) * pct / 100.0) as u32),
+			PixelDim::WidthPercent(_) => None,
+		};
+
+		if max_width.map(|x| img_width <= x).unwrap_or(true)
+			&& max_height.map(|x| img_height <= x).unwrap_or(true)
+		{
+			return (img_width, img_height);
+		}
+
+		let width_ratio = max_width
+			.map(|x| x as f32 / img_width as f32)
+			.unwrap_or(1.0);
+
+		let height_ratio = max_height
+			.map(|x| x as f32 / img_height as f32)
+			.unwrap_or(1.0);
+
+		let ratio = width_ratio.min(height_ratio);
+
+		(
+			(img_width as f32 * ratio) as u32,
+			(img_height as f32 * ratio) as u32,
+		)
+	}
+}
+
+impl Display for MaxDimTransformer {
+	fn fmt(&self, f: &mut std::fmt::Formatter<'_>) -> std::fmt::Result {
+		write!(f, "maxdim({},{})", self.w, self.h)
+	}
+}
+
+impl ImageTransformer for MaxDimTransformer {
+	fn parse_args(args: &str) -> Result<Self, String> {
+		let args: Vec<&str> = args.split(",").collect();
+		if args.len() != 2 {
+			return Err(format!("expected 2 args, got {}", args.len()));
+		}
+
+		let w = args[0].parse::<PixelDim>()?;
+		let h = args[1].parse::<PixelDim>()?;
+
+		Ok(Self { w, h })
+	}
+
+	fn transform(&self, input: &mut DynamicImage) {
+		let (img_width, img_height) = (input.width(), input.height());
+		let (target_width, target_height) = self.target_dim(img_width, img_height);
+
+		// Only resize if needed
+		if target_width != img_width || target_height != img_height {
+			*input = input.resize(target_width, target_height, FilterType::Lanczos3);
+		}
+	}
+}

crates/pile-value/src/extract/item/image/transform/transformers/mod.rs

@@ -0,0 +1,26 @@
+//! Defines all transformation steps we can apply to an image
+
+use image::DynamicImage;
+use std::fmt::{Debug, Display};
+
+mod crop;
+pub use crop::*;
+
+mod maxdim;
+pub use maxdim::*;
+
+/// A single transformation that may be applied to an image.
+pub trait ImageTransformer
+where
+	Self: PartialEq,
+	Self: Sized + Clone,
+	Self: Display + Debug,
+{
+	/// Transform the given image in place
+	fn transform(&self, input: &mut DynamicImage);
+
+	/// Parse an arg string.
+	///
+	/// `name({arg_string})`
+	fn parse_args(args: &str) -> Result<Self, String>;
+}

crates/pile-value/src/extract/item/mod.rs

@@ -31,6 +31,9 @@ pub use group::*;
 mod text;
 pub use text::*;
 
+mod image;
+pub use image::*;
+
 use crate::{
 	extract::{
 		misc::MapExtractor,
@@ -41,6 +44,7 @@ use crate::{
 
 pub struct ItemExtractor {
 	inner: MapExtractor,
+	image: Arc<ImageExtractor>,
 }
 
 impl ItemExtractor {
@@ -91,7 +95,10 @@ impl ItemExtractor {
 			]),
 		};
 
-		Self { inner }
+		Self {
+			inner,
+			image: Arc::new(ImageExtractor::new(item)),
+		}
 	}
 }
 
@@ -103,22 +110,16 @@ impl ObjectExtractor for ItemExtractor {
 		name: &pile_config::Label,
 		args: Option<&str>,
 	) -> Result<Option<PileValue>, std::io::Error> {
-		self.inner.field(state, name, args).await
+		if self.image.fields().await?.contains(name) {
+			self.image.field(state, name, args).await
+		} else {
+			self.inner.field(state, name, args).await
+		}
 	}
 
-	#[expect(clippy::unwrap_used)]
 	async fn fields(&self) -> Result<Vec<Label>, std::io::Error> {
-		return Ok(vec![
-			Label::new("flac").unwrap(),
-			Label::new("id3").unwrap(),
-			Label::new("fs").unwrap(),
-			Label::new("epub").unwrap(),
-			Label::new("exif").unwrap(),
-			Label::new("pdf").unwrap(),
-			Label::new("json").unwrap(),
-			Label::new("toml").unwrap(),
-			Label::new("text").unwrap(),
-			Label::new("groups").unwrap(),
-		]);
+		let mut fields = self.inner.fields().await?;
+		fields.extend(self.image.fields().await?);
+		Ok(fields)
 	}
 }

crates/pile-value/src/value/value.rs

@@ -6,7 +6,7 @@ use std::sync::Arc;
 
 use crate::{
 	extract::{
-		item::ItemExtractor,
+		item::{ImageExtractor, ItemExtractor},
 		misc::{ArrayExtractor, MapExtractor, VecExtractor},
 		string::StringExtractor,
 		traits::{ExtractState, ListExtractor, ObjectExtractor},
@@ -70,7 +70,10 @@ impl PileValue {
 			Self::I64(_) => Arc::new(MapExtractor::default()),
 			Self::Array(_) => Arc::new(MapExtractor::default()),
 			Self::String(s) => Arc::new(StringExtractor::new(s)),
-			Self::Blob { .. } => Arc::new(MapExtractor::default()),
+			Self::Blob { mime, bytes } => {
+				// TODO: make a blobextractor (with pdf, epub, etc; like item)
+				Arc::new(ImageExtractor::from_blob(bytes.clone(), mime.clone()))
+			}
 			Self::ListExtractor(_) => Arc::new(MapExtractor::default()),
 			Self::ObjectExtractor(e) => e.clone(),
 			Self::Item(i) => Arc::new(ItemExtractor::new(i)),