Struct ExtractionOptions 

pub struct ExtractionOptions {

    pub preserve_layout: bool,
    pub space_threshold: f64,
    pub tj_space_threshold: f64,
    pub newline_threshold: f64,
    pub sort_by_position: bool,
    pub detect_columns: bool,
    pub column_threshold: f64,
    pub merge_hyphenated: bool,
    pub track_space_decisions: bool,
    pub reconstruct_paragraphs: bool,
    pub include_artifacts: bool,
    pub reorder_columns: bool,
    pub max_extracted_bytes: Option<usize>,
}

Text extraction options

Fields

preserve_layout: bool

Preserve the original layout (spacing and positioning)

space_threshold: f64

Minimum space width to insert space character (in text space units)

tj_space_threshold: f64

Threshold for synthesising an implicit U+0020 from a TJ numeric kerning offset, expressed as a fraction of the current font size. A TJ kern advances the text matrix by -adjustment/1000 * font_size without rendering any glyph; many PDFs (academic publishers, LaTeX, kerned typography) encode inter-word gaps purely as wide negative kerns rather than literal space bytes. When the synthesised advance exceeds tj_space_threshold * font_size, the extractor inserts one U+0020. Default 0.2 (200 milli-em) sits well between typical intra-word kerning (10-50 milli-em) and the width of a space glyph in most fonts (250-300 milli-em). Lower values catch tighter spaces; higher values reduce false positives in fonts with unusually wide kerning. Separate from space_threshold (which governs the post-glyph gap between separate text-show operators) because the TJ numeric kern is measured without any glyph advance baseline and needs a more sensitive threshold (issue #272).

newline_threshold: f64

Minimum vertical distance to insert newline (in text space units)

sort_by_position: bool

Sort text fragments by position (useful for multi-column layouts)

detect_columns: bool

Detect and handle columns

column_threshold: f64

Column separation threshold (in page units)

merge_hyphenated: bool

Merge hyphenated words at line ends

track_space_decisions: bool

Track space insertion decisions in each TextFragment (default: false). When false: zero overhead. When true: populates TextFragment::space_decisions.

reconstruct_paragraphs: bool

Reconstruct visual lines and paragraphs from the raw text fragments produced by PDF text-show operators. When true, the extractor groups fragments by baseline into single-line fragments, then groups consecutive lines with normal leading into paragraph-level fragments. This is what the partition pipeline needs to produce Element values at paragraph granularity rather than at per-Tj granularity (see issue #261).

Default false for backward compatibility with direct extract_text callers. The PdfDocument::partition* entry points force this to true.

include_artifacts: bool

Include content inside /Artifact marked-content scopes (page headers, footers, watermarks, decorative content). Default false — Artifact content is filtered out, as the PDF/UA conformance level recommends for accessibility tooling and as RAG callers consistently want (issue #269 Phase 1). Opt-in by setting true when extracting page furniture matters (e.g. forensic auditing, redaction tools).

reorder_columns: bool

Reorder flat-text output by column so per-column tokens stay adjacent in multi-column layouts (issue #389). Only affects the flat path (preserve_layout = false); in layout mode detect_columns already reorders. Default false → the flat path is byte-identical to before. When on, .text is produced by the fragment pipeline (its shape matches the layout path’s reconstruction, not stream order); .fragments stays empty.

Column reflow only triggers for column blocks whose rows are spaced at least one line height apart. Layouts pitched tighter than that are geometrically indistinguishable from tight-leading prose that merely contains a wide gap, so they are intentionally left in reading order rather than risk shredding prose (issue #417); text is never corrupted.

Column blocks require gaps that align horizontally across rows: a set of unrelated wide gaps at different X (e.g. a label/value form with varying label lengths) is left in reading order, never reflowed (#422). A genuine table whose column corridor drifts more than ~10pt between rows may also be left un-reordered; text is never corrupted.

max_extracted_bytes: Option<usize>

Stop accumulating decoded text for a page once this many bytes have been collected, bounding the per-page peak memory of extraction. The limit is enforced during accumulation, not by truncating the finished string, so a single page with a huge or adversarially inflated content stream cannot materialise an unbounded String before the caller sees it (issue #382).

Semantics are undershoot: extraction stops before the fragment that would push the accumulated bytes past the limit, so the returned text.len() <= max_extracted_bytes and a multi-byte UTF-8 character is never split. When the limit cuts extraction short, ExtractedText::truncated is set to true.

None (default) means no limit — output is byte-identical to before. The text.len() <= max_extracted_bytes invariant holds on every path (flat, reorder_columns, preserve_layout): the layout paths rebuild .text from the already-bounded fragment set and are then clamped to the limit at a UTF-8 char boundary as a final safety net.

Because whole decoded runs are the unit of truncation, a page whose text is a single run larger than the whole budget (e.g. one huge Tj, or an /ActualText override) comes back with text == "" and truncated == true rather than a partial run — the limit is never satisfied by splitting a run mid-character.

Trait Implementations

impl Clone for ExtractionOptions

fn clone(&self) -> ExtractionOptions

Returns a duplicate of the value.

fn clone_from(&mut self, source: &Self)

Performs copy-assignment from source.

impl Debug for ExtractionOptions

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.

impl Default for ExtractionOptions

fn default() -> Self

Returns the “default value” for a type.