Struct PageTextParser 

pub struct PageTextParser { /* private fields */ }

The parser text layer, driven one page at a time: the document is loaded (and repaired, see [load_document]) once, the font/form caches persist across pages, and each page’s glyphs are parsed only when asked for.

The eager whole-document walk this replaces ran before the first page was rendered, so on a long PDF it was a serial prefix the page-worker pool sat idle through — 6.2 s on the 1913-page .NET reference, in front of a pipeline that otherwise overlaps parsing with inference — and a --pages window still paid for every page in the file. Pulling pages on demand keeps the parse on the producer thread but interleaved with rendering, and skips unselected pages entirely. Output per page is unchanged: same glyph walk, same shared caches, same contraction.

Implementations

impl PageTextParser

pub fn open(bytes: &[u8]) -> Option<Self>

Load the document; None when it has no parseable text layer at all (the caller then keeps pdfium’s cells, as before).

pub fn cells(&mut self, index: usize) -> PageParserCells

Prose, word and code cells of the 0-based page index — empty for an index the parser’s page tree doesn’t have (pdfium and lopdf can disagree on a damaged file; the caller falls back to pdfium’s text).