Generate PDF ID automatically unless we really have a stable ID (#3591)

5 files changed, 40 insertions, 38 deletions

diff --git a/crates/typst-cli/src/compile.rs b/crates/typst-cli/src/compile.rs
index 8e628b4e..9a252a53 100644
--- a/crates/typst-cli/src/compile.rs
+++ b/crates/typst-cli/src/compile.rs
@@ -9,7 +9,7 @@ use parking_lot::RwLock;
 use rayon::iter::{IndexedParallelIterator, IntoParallelRefIterator, ParallelIterator};
 use typst::diag::{bail, At, Severity, SourceDiagnostic, StrResult};
 use typst::eval::Tracer;
-use typst::foundations::Datetime;
+use typst::foundations::{Datetime, Smart};
 use typst::layout::Frame;
 use typst::model::Document;
 use typst::syntax::{FileId, Source, Span};
@@ -157,18 +157,13 @@ fn export(
         OutputFormat::Svg => {
             export_image(world, document, command, watching, ImageExportFormat::Svg)
         }
-        OutputFormat::Pdf => export_pdf(document, command, world),
+        OutputFormat::Pdf => export_pdf(document, command),
     }
 }
 
 /// Export to a PDF.
-fn export_pdf(
-    document: &Document,
-    command: &CompileCommand,
-    world: &SystemWorld,
-) -> StrResult<()> {
-    let ident = world.input().map(|i| i.to_string_lossy());
-    let buffer = typst_pdf::pdf(document, ident.as_deref(), now());
+fn export_pdf(document: &Document, command: &CompileCommand) -> StrResult<()> {
+    let buffer = typst_pdf::pdf(document, Smart::Auto, now());
     let output = command.output();
     fs::write(output, buffer)
         .map_err(|err| eco_format!("failed to write PDF file ({err})"))?;
diff --git a/crates/typst-cli/src/world.rs b/crates/typst-cli/src/world.rs
index 893b02bd..4e0bcd54 100644
--- a/crates/typst-cli/src/world.rs
+++ b/crates/typst-cli/src/world.rs
@@ -29,8 +29,6 @@ static STDIN_ID: Lazy<FileId> =
 pub struct SystemWorld {
     /// The working directory.
     workdir: Option<PathBuf>,
-    /// The canonical path to the input file.
-    input: Option<PathBuf>,
     /// The root relative to which absolute paths are resolved.
     root: PathBuf,
     /// The input path.
@@ -108,7 +106,6 @@ impl SystemWorld {
 
         Ok(Self {
             workdir: std::env::current_dir().ok(),
-            input,
             root,
             main,
             library: Prehashed::new(library),
@@ -152,11 +149,6 @@ impl SystemWorld {
         self.now.take();
     }
 
-    /// Return the canonical path to the input file.
-    pub fn input(&self) -> Option<&PathBuf> {
-        self.input.as_ref()
-    }
-
     /// Lookup a source file by id.
     #[track_caller]
     pub fn lookup(&self, id: FileId) -> Source {
diff --git a/crates/typst-pdf/src/lib.rs b/crates/typst-pdf/src/lib.rs
index cc2edba2..1e425f01 100644
--- a/crates/typst-pdf/src/lib.rs
+++ b/crates/typst-pdf/src/lib.rs
@@ -19,7 +19,7 @@ use ecow::{eco_format, EcoString};
 use pdf_writer::types::Direction;
 use pdf_writer::writers::Destination;
 use pdf_writer::{Finish, Name, Pdf, Ref, Str, TextStr};
-use typst::foundations::{Datetime, Label, NativeElement};
+use typst::foundations::{Datetime, Label, NativeElement, Smart};
 use typst::introspection::Location;
 use typst::layout::{Abs, Dir, Em, Transform};
 use typst::model::{Document, HeadingElem};
@@ -39,12 +39,17 @@ use crate::pattern::PdfPattern;
 ///
 /// Returns the raw bytes making up the PDF file.
 ///
-/// The `ident` parameter shall be a string that uniquely and stably identifies
-/// the document. It should not change between compilations of the same
-/// document. Its hash will be used to create a PDF document identifier (the
-/// identifier itself is not leaked). If `ident` is `None`, a hash of the
-/// document is used instead (which means that it _will_ change across
-/// compilations).
+/// The `ident` parameter, if given, shall be a string that uniquely and stably
+/// identifies the document. It should not change between compilations of the
+/// same document.  **If you cannot provide such a stable identifier, just pass
+/// `Smart::Auto` rather than trying to come up with one.** The CLI, for
+/// example, does not have a well-defined notion of a long-lived project and as
+/// such just passes `Smart::Auto`.
+///
+/// If an `ident` is given, the hash of it will be used to create a PDF document
+/// identifier (the identifier itself is not leaked). If `ident` is `Auto`, a
+/// hash of the document's title and author is used instead (which is reasonably
+/// unique and stable).
 ///
 /// The `timestamp`, if given, is expected to be the creation date of the
 /// document as a UTC datetime. It will only be used if `set document(date: ..)`
@@ -52,7 +57,7 @@ use crate::pattern::PdfPattern;
 #[typst_macros::time(name = "pdf")]
 pub fn pdf(
     document: &Document,
-    ident: Option<&str>,
+    ident: Smart<&str>,
     timestamp: Option<Datetime>,
 ) -> Vec<u8> {
     let mut ctx = PdfContext::new(document);
@@ -158,7 +163,7 @@ impl<'a> PdfContext<'a> {
 }
 
 /// Write the document catalog.
-fn write_catalog(ctx: &mut PdfContext, ident: Option<&str>, timestamp: Option<Datetime>) {
+fn write_catalog(ctx: &mut PdfContext, ident: Smart<&str>, timestamp: Option<Datetime>) {
     let lang = ctx.languages.iter().max_by_key(|(_, &count)| count).map(|(&l, _)| l);
 
     let dir = if lang.map(Lang::dir) == Some(Dir::RTL) {
@@ -236,18 +241,25 @@ fn write_catalog(ctx: &mut PdfContext, ident: Option<&str>, timestamp: Option<Da
     // changes in the frames.
     let instance_id = hash_base64(&ctx.pdf.as_bytes());
 
-    if let Some(ident) = ident {
-        // A unique ID for the document that stays stable across compilations.
-        let doc_id = hash_base64(&("PDF-1.7", ident));
-        xmp.document_id(&doc_id);
-        xmp.instance_id(&instance_id);
-        ctx.pdf
-            .set_file_id((doc_id.clone().into_bytes(), instance_id.into_bytes()));
+    // Determine the document's ID. It should be as stable as possible.
+    const PDF_VERSION: &str = "PDF-1.7";
+    let doc_id = if let Smart::Custom(ident) = ident {
+        // We were provided with a stable ID. Yay!
+        hash_base64(&(PDF_VERSION, ident))
+    } else if ctx.document.title.is_some() && !ctx.document.author.is_empty() {
+        // If not provided from the outside, but title and author were given, we
+        // compute a hash of them, which should be reasonably stable and unique.
+        hash_base64(&(PDF_VERSION, &ctx.document.title, &ctx.document.author))
     } else {
-        // This is not spec-compliant, but some PDF readers really want an ID.
-        let bytes = instance_id.into_bytes();
-        ctx.pdf.set_file_id((bytes.clone(), bytes));
-    }
+        // The user provided no usable metadata which we can use as an `/ID`.
+        instance_id.clone()
+    };
+
+    // Write IDs.
+    xmp.document_id(&doc_id);
+    xmp.instance_id(&instance_id);
+    ctx.pdf
+        .set_file_id((doc_id.clone().into_bytes(), instance_id.into_bytes()));
 
     xmp.rendition_class(RenditionClass::Proof);
     xmp.pdf_version("1.7");
diff --git a/crates/typst/src/model/document.rs b/crates/typst/src/model/document.rs
index edd5e6c9..1e614399 100644
--- a/crates/typst/src/model/document.rs
+++ b/crates/typst/src/model/document.rs
@@ -51,6 +51,9 @@ pub struct DocumentElem {
     ///
     /// The year component must be at least zero in order to be embedded into a
     /// PDF.
+    ///
+    /// If you want to create byte-by-byte reproducible PDFs, set this to
+    /// something other than `{auto}`.
     #[ghost]
     pub date: Smart<Option<Datetime>>,
 
diff --git a/tests/src/tests.rs b/tests/src/tests.rs
index 4c3975bb..e4f60bb6 100644
--- a/tests/src/tests.rs
+++ b/tests/src/tests.rs
@@ -501,7 +501,7 @@ fn test(
         if let Some(pdf_path) = pdf_path {
             let pdf_data = typst_pdf::pdf(
                 &document,
-                Some(&format!("typst-test: {}", name.display())),
+                Smart::Custom(&format!("typst-test: {}", name.display())),
                 world.today(Some(0)),
             );
             fs::create_dir_all(pdf_path.parent().unwrap()).unwrap();