Overhaul the serializer pipeline (#23)

Author: LaurenzV

Cargo.lock

@@ -7,7 +7,7 @@ edition = "2021"
 image = { version = "0.25.1", default-features = false, features = ["jpeg", "png", "gif", "webp"] }
 miniz_oxide = "0.7.4"
 once_cell = "1.19.0"
-pdf-writer = {git = "https://github.com/LaurenzV/pdf-writer", rev = "f83b414"}
+pdf-writer = {git = "https://github.com/LaurenzV/pdf-writer", rev = "34c991f5"}
 resvg = {git = "https://github.com/LaurenzV/resvg", rev = "1c2b6bd0"}
 siphasher = "1.0.1"
 skrifa = {git="https://github.com/LaurenzV/fontations", rev="0ed7955"}

Cargo.toml

@@ -0,0 +1,151 @@
+use crate::serialize::SerializeSettings;
+use pdf_writer::{Chunk, Finish, Name, Pdf, Ref};
+use std::collections::HashMap;
+
+pub struct ChunkContainer {
+    pub(crate) page_label_tree: Option<(Ref, Chunk)>,
+    pub(crate) page_tree: Option<(Ref, Chunk)>,
+    pub(crate) outline: Option<(Ref, Chunk)>,
+
+    pub(crate) pages: Vec<Chunk>,
+    pub(crate) page_labels: Vec<Chunk>,
+    pub(crate) annotations: Vec<Chunk>,
+    pub(crate) fonts: Vec<Chunk>,
+    pub(crate) color_spaces: Vec<Chunk>,
+    pub(crate) destinations: Vec<Chunk>,
+    pub(crate) ext_g_states: Vec<Chunk>,
+    pub(crate) images: Vec<Chunk>,
+    pub(crate) masks: Vec<Chunk>,
+    pub(crate) x_objects: Vec<Chunk>,
+    pub(crate) shading_functions: Vec<Chunk>,
+    pub(crate) patterns: Vec<Chunk>,
+}
+
+impl ChunkContainer {
+    pub fn new() -> Self {
+        Self {
+            page_tree: None,
+            outline: None,
+            page_label_tree: None,
+
+            pages: vec![],
+            page_labels: vec![],
+            annotations: vec![],
+            fonts: vec![],
+            color_spaces: vec![],
+            destinations: vec![],
+            ext_g_states: vec![],
+            images: vec![],
+            masks: vec![],
+            x_objects: vec![],
+            shading_functions: vec![],
+            patterns: vec![],
+        }
+    }
+
+    pub fn finish(mut self, serialize_settings: SerializeSettings) -> Pdf {
+        let mut remapped_ref = Ref::new(1);
+        let mut remapper = HashMap::new();
+
+        // Two utility macros, that basically traverses the fields in the order that we
+        // will write them to the PDF and assigns new references as we go.
+        // This gives us the advantage that the PDF will be numbered with
+        // monotonically increasing numbers, which, while it is not a strict requirement
+        // for a valid PDF, makes it a lot cleaner and might make implementing features
+        // like object streams easier down the road.
+        //
+        // It also allows us to estimate the capacity we will need for the new PDF.
+        let mut chunks_len = 0;
+        macro_rules! remap_field {
+            ($self:expr, $remapper:expr, $remapped_ref:expr; $($field:ident),+) => {
+                $(
+                    if let Some((original_ref, chunk)) = &mut $self.$field {
+                        chunks_len += chunk.len();
+                        for object_ref in chunk.object_refs() {
+                            debug_assert!(!remapper.contains_key(&object_ref));
+
+                            $remapper.insert(object_ref, $remapped_ref.bump());
+                            *original_ref = *remapper.get(&object_ref).unwrap();
+                        }
+                    }
+                )+
+            };
+        }
+
+        macro_rules! remap_fields {
+            ($self:expr, $remapper:expr, $remapped_ref:expr; $($field:ident),+) => {
+                $(
+                    for chunk in &$self.$field {
+                        chunks_len += chunk.len();
+                        for ref_ in chunk.object_refs() {
+                            debug_assert!(!remapper.contains_key(&ref_));
+
+                            $remapper.insert(ref_, $remapped_ref.bump());
+                        }
+                    }
+                )+
+            };
+        }
+
+        // Chunk length is not an exact number because the length might change as we renumber,
+        // so we add a bit of a padding by multiplying with 1.1. The 200 is additional padding
+        // for the document catalog. This hopefully allows us to avoid realloactions in the general
+        // case, and thus give us better performance.
+        let mut pdf = Pdf::with_capacity((chunks_len as f32 * 1.1 + 200.0) as usize);
+
+        if serialize_settings.ascii_compatible {
+            pdf.set_binary_marker(&[b'A', b'A', b'A', b'A'])
+        }
+
+        // We only write a catalog if a page tree exists. Every valid PDF must have one
+        // and krilla ensures that there always is one, but for snapshot tests, it can be
+        // useful to not write a document catalog if we don't actually need it for the test.
+        if self.page_tree.is_some() || self.outline.is_some() || self.page_label_tree.is_some() {
+            let catalog_ref = remapped_ref.bump();
+
+            let mut catalog = pdf.catalog(catalog_ref);
+            remap_field!(self, remapper, remapped_ref; page_tree, outline, page_label_tree);
+
+            if let Some(pt) = &self.page_tree {
+                catalog.pages(pt.0);
+            }
+
+            if let Some(pl) = &self.page_label_tree {
+                catalog.pair(Name(b"PageLabels"), pl.0);
+            }
+
+            if let Some(ol) = &self.outline {
+                catalog.outlines(ol.0);
+            }
+
+            catalog.finish();
+        }
+
+        remap_fields!(self, remapper, remapped_ref; pages, page_labels, annotations, fonts, color_spaces, destinations, ext_g_states, images, masks, x_objects, shading_functions, patterns);
+
+        macro_rules! write_field {
+            ($self:expr, $remapper:expr, $pdf:expr; $($field:ident),+) => {
+                $(
+                    if let Some((_, chunk)) = $self.$field {
+                        chunk.renumber_into($pdf, |old| *$remapper.get(&old).unwrap());
+                    }
+                )+
+            };
+        }
+
+        macro_rules! write_fields {
+            ($self:expr, $remapper:expr, $pdf:expr; $($field:ident),+) => {
+                $(
+                    for chunk in $self.$field {
+                        chunk.renumber_into($pdf, |old| *$remapper.get(&old).unwrap());
+                    }
+                )+
+            };
+        }
+
+        write_field!(self, remapper, &mut pdf; page_tree, outline, page_label_tree);
+        write_fields!(self, remapper, &mut pdf; pages, page_labels, annotations, fonts, color_spaces, destinations, ext_g_states, images, masks, x_objects, shading_functions, patterns);
+
+        pdf
+    }
+}

src/chunk_container.rs

@@ -1,3 +1,4 @@
+use crate::chunk_container::ChunkContainer;
 use crate::serialize::{Object, SerializerContext, SvgSettings};
 use crate::surface::Surface;
 use crate::type3_font::Type3ID;
@@ -302,6 +303,10 @@ impl FontIdentifier {
 
 // TODO: Remove?
 impl Object for FontIdentifier {
+    fn chunk_container<'a>(&self, _: &'a mut ChunkContainer) -> &'a mut Vec<Chunk> {
+        unreachable!()
+    }
+
     fn serialize_into(&self, _: &mut SerializerContext, _: Ref) -> Chunk {
         unreachable!()
     }

src/font/mod.rs

@@ -1,3 +1,4 @@
+pub mod chunk_container;
 pub mod document;
 pub mod font;
 mod graphics_state;

src/lib.rs

@@ -1,5 +1,5 @@
 use crate::font::{CIDIdentifer, Font, FontIdentifier};
-use crate::serialize::{Object, SerializerContext, SipHashable};
+use crate::serialize::{SerializerContext, SipHashable};
 use crate::util::{RectExt, SliceExt};
 use pdf_writer::types::{CidFontType, FontFlags, SystemInfo, UnicodeCmap};
 use pdf_writer::{Chunk, Filter, Finish, Name, Ref, Str};
@@ -90,10 +90,8 @@ impl CIDFont {
     pub fn identifier(&self) -> FontIdentifier {
         FontIdentifier::Cid(CIDIdentifer(self.font.clone()))
     }
-}
 
-impl Object for CIDFont {
-    fn serialize_into(&self, sc: &mut SerializerContext, root_ref: Ref) -> Chunk {
+    pub(crate) fn serialize_into(&self, sc: &mut SerializerContext, root_ref: Ref) -> Chunk {
         let mut chunk = Chunk::new();
 
         let cid_ref = sc.new_ref();