vectorcraft/crates/pdf/src/post.rs

//! What the export does to the file once it is written, in this order: the pages' thumbnails
//! (General › Embed page thumbnails), the seal of the editing data, fast web view (a linearised
//! file, [`crate::linearize`]) and encryption ([`crate::encrypt`]), which a linearised file goes
//! through between its two steps so that it stays linearised. Also here: what makes a file say it
//! is PDF 1.3 ([`declare_pdf13`]), for PDF 1.3 and PDF/X-1a and PDF/X-3 files, the file's
//! modification date ([`modification_date`]), its copyright ([`rights`]) and the pages' user units
//! ([`user_units`]), which the export writes in that order before the thumbnails.

use std::borrow::Cow;

use vectorcraft_doc::{CopyrightStatus, DocMetadata};

use crate::encrypt::Obj;
use crate::linearize::{self, Parsed};
use crate::patch::Patch;
use crate::{PdfError, PdfOptions, PdfSettings};

/// Longest side of a page thumbnail, in pixels.
pub const THUMBNAIL_SIZE: u32 = 106;

/// A page thumbnail: 8-bit RGB pixels, row by row, at most [`THUMBNAIL_SIZE`] pixels a side.
#[derive(Clone, Debug, Default, PartialEq)]
pub struct Thumbnail {
    pub width: u32,
    pub height: u32,
    pub rgb: Vec<u8>,
}

impl Thumbnail {
    /// The image XObject of a page's `/Thumb` entry; `gray`: in DeviceGray (for PDF/X files, whose
    /// device colours must suit the printing condition, an RGB thumbnail wouldn't).
    fn object(&self, gray: bool) -> Result<Vec<u8>, PdfError> {
        let (w, h) = (self.width, self.height);
        let sized = (1..=THUMBNAIL_SIZE).contains(&w) && (1..=THUMBNAIL_SIZE).contains(&h);
        if !sized || self.rgb.len() as u64 != u64::from(w) * u64::from(h) * 3 {
            return Err(PdfError::BadSetting(format!("a page thumbnail must be RGB pixels, 1 to {THUMBNAIL_SIZE} a side (got {w} × {h})")));
        }
        let (space, pixels) = if gray {
            // Rec. 601 luma.
            let luma = |p: &[u8; 3]| ((299 * u32::from(p[0]) + 587 * u32::from(p[1]) + 114 * u32::from(p[2]) + 500) / 1000) as u8;
            ("DeviceGray", Cow::Owned(self.rgb.as_chunks::<3>().0.iter().map(luma).collect()))
        } else {
            ("DeviceRGB", Cow::Borrowed(self.rgb.as_slice()))
        };
        let data = crate::output::deflate(&pixels).map_err(|e| PdfError::Write(format!("page thumbnail: {e}")))?;
        let mut out = format!("<</Width {w}/Height {h}/ColorSpace/{space}/BitsPerComponent 8/Filter/FlateDecode/Length {}>>\nstream\n", data.len())
            .into_bytes();
        out.extend_from_slice(&data);
        out.extend_from_slice(b"\nendstream");
        Ok(out)
    }
}

/// `pdf` with `thumbs` as the thumbnails of its pages, in page order (in grey when `gray`).
fn thumbnails(pdf: Vec<u8>, thumbs: &[Thumbnail], gray: bool) -> Result<Vec<u8>, PdfError> {
    let failed = |why: &str| PdfError::Write(format!("can't add page thumbnails: {why}"));
    let Parsed { xref, pages, .. } = linearize::parse(&pdf).map_err(failed)?;
    let mut patch = Patch::new(&xref);
    for (page, thumb) in pages.iter().zip(thumbs) {
        let (dict, _) = xref.dict(&pdf, *page).ok_or_else(|| failed("a page can't be read"))?;
        let n = patch.add_object(thumb.object(gray)?);
        patch.replace(dict, dict, format!("/Thumb {n} 0 R").into_bytes());
    }
    patch.apply(&pdf, &xref)
}

/// `pdf` with `units` as the `UserUnit` entries of its pages, in page order (a unit of 1 writes
/// none); unchanged when every unit is 1.
pub(crate) fn user_units(pdf: Vec<u8>, units: &[f64]) -> Result<Vec<u8>, PdfError> {
    if units.iter().all(|u| *u == 1.0) {
        return Ok(pdf);
    }
    let failed = |why: &str| PdfError::Write(format!("can't write the pages' user units: {why}"));
    let Parsed { xref, pages, .. } = linearize::parse(&pdf).map_err(failed)?;
    let mut patch = Patch::new(&xref);
    for (page, unit) in pages.iter().zip(units).filter(|(_, u)| **u != 1.0) {
        let (dict, _) = xref.dict(&pdf, *page).ok_or_else(|| failed("a page can't be read"))?;
        patch.replace(dict, dict, format!("/UserUnit {unit}").into_bytes());
    }
    patch.apply(&pdf, &xref)
}

/// `pdf` dated as modified at `modified` (Unix seconds, UTC), its creation dates kept: the
/// document information's `/ModDate` (a PDF date, ISO 32000-1 §7.9.4) and every `xmp:ModifyDate`,
/// `xmp:MetadataDate` and `stEvt:when` of the catalog's XMP metadata (ISO 8601 dates), in UTC as
/// the PDF writer writes the creation date. The writer gives `/ModDate`, `xmp:ModifyDate` and
/// `stEvt:when` the creation date.
pub(crate) fn modification_date(pdf: Vec<u8>, modified: i64) -> Result<Vec<u8>, PdfError> {
    let failed = |why: &str| PdfError::Write(format!("can't write the file's modification date: {why}"));
    let Parsed { xref, objects, trailer, root, .. } = linearize::parse(&pdf).map_err(failed)?;
    let [y, mo, d, h, mi, s] = vectorcraft_doc::metadata::civil(modified);
    let mut patch = Patch::new(&xref);
    if let Some(&Obj::Ref(n, _)) = trailer.get(b"Info")
        && let Some(Obj::Str { start, end, .. }) = objects.get(&n).and_then(|o| o.value.get(b"ModDate"))
    {
        patch.replace(*start, *end, format!("(D:{y:04}{mo:02}{d:02}{h:02}{mi:02}{s:02}Z)").into_bytes());
    }
    let metadata = match objects.get(&root).and_then(|c| c.value.get(b"Metadata")) {
        Some(&Obj::Ref(n, _)) => objects.get(&n),
        _ => None,
    };
    if let Some(o) = metadata
        && let Some(&Obj::Int { value, start, end }) = o.value.get(b"Length")
    {
        let date = format!("{y:04}-{mo:02}-{d:02}T{h:02}:{mi:02}:{s:02}+00:00");
        let xmp = pdf.get(..o.endobj).unwrap_or_default();
        let mut grown = 0;
        for tag in ["xmp:ModifyDate", "xmp:MetadataDate", "stEvt:when"] {
            let (open, close) = (format!("<{tag}>"), format!("</{tag}>"));
            let mut from = o.value_span.1;
            while let Some(at) = crate::lab_spot::find(xmp, open.as_bytes(), from).map(|i| i + open.len())
                && let Some(to) = crate::lab_spot::find(xmp, close.as_bytes(), at)
            {
                grown += date.len() as i64 - (to - at) as i64;
                patch.replace(at, to, date.clone().into_bytes());
                from = to + close.len();
            }
        }
        if grown != 0 {
            patch.replace(start, end, (value + grown).to_string().into_bytes());
        }
    }
    patch.apply(&pdf, &xref)
}

/// `pdf` with File Info's copyright (`info`) added to the catalog's XMP metadata, after the
/// writer's properties: the notice as `dc:rights`, the URL as `xmpRights:WebStatement` and the
/// status as `xmpRights:Marked` (`True` for a copyrighted work, `False` for a public-domain one,
/// none when it is unknown), as [`crate::file_info`] reads them. Unchanged when `info` gives none of
/// them or the file has no XMP metadata.
pub(crate) fn rights(pdf: Vec<u8>, info: &DocMetadata) -> Result<Vec<u8>, PdfError> {
    let props = rights_properties(info);
    if props.is_empty() {
        return Ok(pdf);
    }
    let failed = |why: &str| PdfError::Write(format!("can't write the file's copyright: {why}"));
    let Parsed { xref, objects, root, .. } = linearize::parse(&pdf).map_err(failed)?;
    let metadata = match objects.get(&root).and_then(|c| c.value.get(b"Metadata")) {
        Some(&Obj::Ref(n, _)) => objects.get(&n),
        _ => None,
    };
    let Some(o) = metadata else { return Ok(pdf) };
    let Some(&Obj::Int { value, start, end }) = o.value.get(b"Length") else { return Err(failed("its metadata has no direct length")) };
    let xmp = pdf.get(o.value_span.1..o.endobj).unwrap_or_default();
    let close = crate::lab_spot::rfind(xmp, b"</rdf:Description>").ok_or_else(|| failed("its metadata has no description"))?;
    let at = o.value_span.1 + close;
    let mut patch = Patch::new(&xref);
    patch.replace(start, end, (value + props.len() as i64).to_string().into_bytes());
    patch.replace(at, at, props.into_bytes());
    patch.apply(&pdf, &xref)
}

/// The XMP properties [`rights`] adds, each declaring its namespace; empty when `info` has no
/// copyright notice or URL and its status is unknown.
fn rights_properties(info: &DocMetadata) -> String {
    use crate::file_info::{DC, XMP_RIGHTS};
    let mut out = String::new();
    let notice = info.copyright_notice.trim();
    if !notice.is_empty() {
        out.push_str(&format!(
            "<dc:rights xmlns:dc=\"{DC}\"><rdf:Alt><rdf:li xml:lang=\"x-default\">{}</rdf:li></rdf:Alt></dc:rights>",
            xml_text(notice)
        ));
    }
    let url = info.copyright_url.trim();
    if !url.is_empty() {
        out.push_str(&format!("<xmpRights:WebStatement xmlns:xmpRights=\"{XMP_RIGHTS}\">{}</xmpRights:WebStatement>", xml_text(url)));
    }
    let marked = match info.copyright_status {
        CopyrightStatus::Copyrighted => Some("True"),
        CopyrightStatus::PublicDomain => Some("False"),
        CopyrightStatus::Unknown => None,
    };
    if let Some(m) = marked {
        out.push_str(&format!("<xmpRights:Marked xmlns:xmpRights=\"{XMP_RIGHTS}\">{m}</xmpRights:Marked>"));
    }
    out
}

/// `s` as XML character data: `&`, `<`, `>` and `"` escaped, a carriage return written as the
/// character reference `&#xD;`, and the characters XML 1.0 doesn't allow left out.
fn xml_text(s: &str) -> String {
    let mut out = String::with_capacity(s.len());
    for c in s.chars() {
        match c {
            '&' => out.push_str("&amp;"),
            '<' => out.push_str("&lt;"),
            '>' => out.push_str("&gt;"),
            '"' => out.push_str("&quot;"),
            '\r' => out.push_str("&#xD;"),
            '\t' | '\n' | '\u{20}'..='\u{D7FF}' | '\u{E000}'..='\u{FFFD}' | '\u{10000}'.. => out.push(c),
            _ => {}
        }
    }
    out
}

/// Linearise `pdf`, encrypting it on the way as `set` says.
fn linearised(pdf: &[u8], set: &PdfSettings) -> Result<Vec<u8>, PdfError> {
    let (prepared, plan) = linearize::prepare(pdf)?;
    let (encrypted, mut cipher) = crate::encrypt::protect_keeping(prepared, set)?;
    linearize::write(&encrypted, &plan, cipher.as_mut())
}

/// The file the export wrote (`pdf`), finished as `opts` say; `editing`: it carries the editing
/// data, to seal once the pages are final. What can't be done comes back in `warnings`.
pub(crate) fn finish(pdf: Vec<u8>, opts: &PdfOptions, editing: bool, warnings: &mut Vec<String>) -> Result<Vec<u8>, PdfError> {
    let set = &opts.settings;
    let pdf = match (set.thumbnails, opts.thumbnails.is_empty()) {
        (true, false) => thumbnails(pdf, &opts.thumbnails, set.standard.is_pdfx())?,
        (true, true) => {
            warnings.push("page thumbnails need the pages drawn, which weren't given: none are embedded".into());
            pdf
        }
        (false, _) => pdf,
    };
    let pdf = if editing { crate::editing::seal(pdf)? } else { pdf };
    if !set.fast_web_view {
        return crate::encrypt::protect(pdf, set);
    }
    match linearised(&pdf, set) {
        Ok(out) => Ok(out),
        Err(e) => {
            warnings.push(format!("the file isn't optimized for fast web view: {e}"));
            crate::encrypt::protect(pdf, set)
        }
    }
}

/// Make a file written with the PDF 1.4 settings say it is PDF 1.3: its header and the version
/// its XMP metadata gives, rewritten in place.
pub(crate) fn declare_pdf13(pdf: &mut [u8]) {
    if let Some(v) = pdf.get_mut(..8).filter(|h| h.starts_with(b"%PDF-1.4")) {
        v.copy_from_slice(b"%PDF-1.3");
    }
    if let Some(digit) = metadata_version(pdf).and_then(|at| pdf.get_mut(at)) {
        *digit = b'3';
    }
}

/// Where the minor version digit is in the `1.4` of the XMP metadata stream of `pdf`'s catalog.
fn metadata_version(pdf: &[u8]) -> Option<usize> {
    const VERSION: &[u8] = b"<pdf:PDFVersion>1.4<";
    let p = linearize::parse(pdf).ok()?;
    let Some(&Obj::Ref(n, _)) = p.objects.get(&p.root)?.value.get(b"Metadata") else { return None };
    let o = p.objects.get(&n)?;
    let at = crate::lab_spot::find(pdf.get(..o.endobj)?, VERSION, o.value_span.1)?;
    Some(at + VERSION.len() - 2)
}