diff --git a/crates/export/src/as_docx.rs b/crates/export/src/as_docx.rs index 0dd395d08a..24b1c5eb98 100644 --- a/crates/export/src/as_docx.rs +++ b/crates/export/src/as_docx.rs @@ -150,7 +150,7 @@ pub(crate) const MAX_MEDIA: usize = 1000; const MAX_MEDIA_BYTES: usize = 512 << 20; /// Whether `im` can still be embedded: it has data, and the package stays within its limits. -fn room_for(media: &[(String, &Image)], im: &Image) -> bool { +pub(crate) fn room_for(media: &[(String, &Image)], im: &Image) -> bool { !im.bytes.is_empty() && media.len() < MAX_MEDIA && media.iter().map(|(_, m)| m.bytes.len()).fold(im.bytes.len(), usize::saturating_add) <= MAX_MEDIA_BYTES diff --git a/crates/export/src/as_html.rs b/crates/export/src/as_html.rs index 7d70359edd..5d51e3b6fe 100644 --- a/crates/export/src/as_html.rs +++ b/crates/export/src/as_html.rs @@ -1,5 +1,5 @@ -use super::as_docx::{hex, rgb8}; -use super::layout::{Item, Page, Run, Table, items}; +use super::as_docx::{hex, rgb8, room_for}; +use super::layout::{Image, Item, Page, Run, Table, items}; /// Whether a character may appear in the output. XML 1.0 forbids the C0 controls other than tab, /// line feed and carriage return, and U+FFFE/U+FFFF. Text extracted from PDFs often holds them @@ -152,6 +152,7 @@ fn html_with_mode(pages: &[Page], title: &str, simple: bool, language: Option<&s "" } ); + let mut embedded: Vec<(String, &Image)> = Vec::new(); for it in items(pages) { match it { Item::Para(runs, lvl) => { @@ -160,10 +161,13 @@ fn html_with_mode(pages: &[Page], title: &str, simple: bool, language: Option<&s } } Item::Table(t) => s.push_str(&table_html(&t, simple)), - Item::Img(im) => { + // HTML embeds each placement in full, so repeated pictures also consume the budget. + Item::Img(im) if room_for(&embedded, im) => { let mime = if im.ext == "jpg" { "image/jpeg" } else { "image/png" }; s.push_str(&format!("

\"\"

\n", base64(&im.bytes))); + embedded.push((String::new(), im)); } + Item::Img(_) => {} Item::PageBreak => s.push_str("
\n"), } } diff --git a/crates/export/src/pptx.rs b/crates/export/src/pptx.rs index abfdb833c4..59d9f858d9 100644 --- a/crates/export/src/pptx.rs +++ b/crates/export/src/pptx.rs @@ -1,7 +1,9 @@ //! PowerPoint presentation (.pptx, Office Open XML PresentationML): slides per page, //! largest text as slide title, remaining paragraphs as body text, images and tables between them. -use crate::as_docx::{core_properties_xml, image_mime}; +use std::collections::HashMap; + +use crate::as_docx::{core_properties_xml, image_mime, room_for}; use crate::{Image, Page, Run, Table, hex}; use craft_opc::{PackageOut, RT_BASE, RelOut, RelsOut}; use craft_xml::{DECL_STANDALONE, esc}; @@ -127,6 +129,8 @@ pub fn pptx(pages: &[Page], title: &str) -> Vec { let content_w = sw - 2 * mx; let mut global_media: Vec<(String, &Image)> = Vec::new(); + // Shared picture bytes remain alive throughout export; only admitted media enter this map. + let mut media_of: HashMap<*const u8, usize> = HashMap::new(); let mut slides: Vec<(String, RelsOut)> = Vec::new(); // (slide_xml, slide_rels) let body = crate::layout::body_size(pages); @@ -201,9 +205,19 @@ pub fn pptx(pages: &[Page], title: &str) -> Vec { current_y = current_y.saturating_add(box_h).saturating_add(height_emu(12.0)); } SlideBodyItem::Img(im) => { - let global_img_idx = global_media.len() + 1; - let img_name = format!("image{global_img_idx}.{}", im.ext); - global_media.push((img_name, *im)); + let global_img_idx = match media_of.get(&im.bytes.as_ptr()) { + Some(&i) => i, + None if room_for(&global_media, im) => { + let n = global_media.len() + 1; + global_media.push((format!("image{n}.{}", im.ext), *im)); + media_of.insert(im.bytes.as_ptr(), n); + n + } + None => { + i += 1; + continue; + } + }; let r_id = format!("rIdImg{}", slide_media_rels.len() + 1); slide_media_rels.push((r_id.clone(), global_img_idx)); diff --git a/crates/export/src/pptx/tests.rs b/crates/export/src/pptx/tests.rs index 2ed0235778..d703c682f3 100644 --- a/crates/export/src/pptx/tests.rs +++ b/crates/export/src/pptx/tests.rs @@ -403,3 +403,40 @@ fn pptx_opens_in_libreoffice() { } let _ = std::fs::remove_dir_all(&dir); } + +/// A picture drawn in many places is one media part, referred to from each place; the package +/// stays within the media limits however many places there are. +#[test] +fn pptx_packages_a_repeated_picture_once() { + let mut p = sample_page(); + let one = p.images[0].clone(); + p.images = std::iter::repeat_n(one, 2 * crate::as_docx::MAX_MEDIA).collect(); + let bytes = pptx(&[p], "Deck"); + let rels = part(&bytes, "ppt/slides/_rels/slide1.xml.rels"); + assert_eq!(rels.matches("relationships/image\"").count(), 2 * crate::as_docx::MAX_MEDIA, "every place refers to the picture"); + assert!(rels.contains("Target=\"../media/image1.png\"") && !rels.contains("image2."), "one media part"); + assert_eq!(part(&bytes, "ppt/slides/slide1.xml").matches("").count(), 2 * crate::as_docx::MAX_MEDIA); + assert_eq!(bytes.windows(20).filter(|w| *w == b"ppt/media/image1.png").count(), 2, "its local header and central directory entry"); +} + +/// Distinct allocations remain distinct media, even when their bytes match. Once the media +/// budget is full, an existing picture remains usable on a later slide, but a new one is omitted. +#[test] +fn pptx_bounds_distinct_pictures_and_reuses_media_across_slides() { + let mut first = sample_page(); + let one = first.images[0].clone(); + first.images = (0..crate::as_docx::MAX_MEDIA).map(|_| Image { bytes: one.bytes.as_ref().to_vec().into(), ..one.clone() }).collect(); + // Media names follow reading order, not the input vector. Give the reused picture a + // distinct top edge so it is unambiguously image1 rather than tying every rectangle. + first.images[0].rect[3] += 1.0; + let mut second = sample_page(); + second.images = + vec![first.images[0].clone(), Image { bytes: one.bytes.as_ref().to_vec().into(), ..one.clone() }, Image { bytes: Vec::new().into(), ..one }]; + let bytes = pptx(&[first, second], "Deck"); + assert_eq!(part(&bytes, "ppt/slides/slide1.xml").matches("").count(), crate::as_docx::MAX_MEDIA); + assert_eq!(part(&bytes, "ppt/slides/slide2.xml").matches("").count(), 1); + let rels = part(&bytes, "ppt/slides/_rels/slide2.xml.rels"); + assert_eq!(rels.matches("relationships/image\"").count(), 1); + assert!(rels.contains("Target=\"../media/image1.png\"")); + assert!(part(&bytes, "ppt/slides/_rels/slide1.xml.rels").contains("../media/image1000.png")); +} diff --git a/crates/export/src/tests.rs b/crates/export/src/tests.rs index 68d3a569ee..3bae8571d3 100644 --- a/crates/export/src/tests.rs +++ b/crates/export/src/tests.rs @@ -1442,3 +1442,15 @@ fn shared_base64_keeps_padded_unwrapped_export_bytes() { assert!(!encoded.contains(['\r', '\n'])); assert!(encoded.ends_with("/w==")); } + +/// Each place a picture is drawn embeds it whole in the HTML, so the places count against the +/// media limits: past them, pictures are left out instead of multiplying one into gigabytes. +#[test] +fn html_bounds_the_pictures_it_embeds() { + let mut p = page(); + let one = p.images[0].clone(); + p.images = std::iter::repeat_n(one, MAX_MEDIA + 10).collect(); + for h in [html(std::slice::from_ref(&p), "Report"), html_simple(std::slice::from_ref(&p), "Report")] { + assert_eq!(h.matches("