Struct PdfStream 

pub struct PdfStream { /* private fields */ }

On-demand page extractor backed by a single parsed pdf_oxide::PdfDocument.

pdf_oxide parses the file lazily — open does the xref + catalog and returns in tens of milliseconds even on the 31 MB / 1310-page PDF reference, where lopdf::Document::load (the old backend) took ~40 s because it eagerly decompressed every content stream. Per-page extraction is sub-millisecond warm, hundreds of micros cold.

pdf_oxide::PdfDocument is Send + Sync (its interior-mutable caches are Mutex-guarded), so a PdfStream can be wrapped in Arc and shared between the main thread (rendering the first visible page) and the background loader thread (extracting the rest of the document) the same way the lopdf-backed version was.

Implementations

impl PdfStream

pub fn open(pdf_path: &str) -> Result<Self, Box<dyn Error>>

Open a PDF and parse its catalog. Does not extract any page text.

pub fn open_with_bundled_ocr(pdf_path: &str) -> Result<Self, Box<dyn Error>>

pub fn total_pages(&self) -> usize

pub fn canonical_path(&self) -> &Path

pub fn extract_page(&self, page_index: usize) -> Option<String>

Extract sanitized text for a single page.

page_index is 1-based to match the historical lopdf-backed API (the rest of hygg counts pages from 1 in saved progress, status line, etc.). Returns None if the index is out of range, the page has no extractable text, or extraction panicked. pdf_oxide claims a 100 % pass rate on its 3 830-PDF corpus, but we still wrap in catch_unwind so a misbehaving page can’t take down the background loader thread and leave every later page stuck on “loading”.

Uses pdf_oxide’s positional extract_text_lines rather than the simpler extract_text. The former returns each visual line with its bounding box; we group lines that share a row (overlapping y ranges) and join them left-to-right. Without that step pdf_oxide can interleave adjacent TOC entries — “1.3 Foo1.4 Bar 3231” — and the downstream sanitizer can’t recover them.

pub fn extract_page_with_images( &self, page_index: usize, col: usize, ) -> Option<PdfRenderedPage>