Enum OcrLang 

pub enum OcrLang {
    En,
    Ch,
}

OCR recognition language: which PP-OCRv3 model + dictionary pair runs when the PP-OCRv6 recognizer is not installed.

With .models/ocr_rec_v6.onnx + .models/ocr_rec_v6_dict.txt on disk (#570; download_dependencies.sh fetches them) both languages run that one multilingual model — RapidOCR’s PP-OCRv6_rec_small, the recognizer docling runs for English and Chinese alike, and the single largest factor in the FUNSD word-recall gap once the detector’s lines are the crops (0.70 → 0.77 on 30 forms). Without it, the PP-OCRv3 pairs: the default is English (.models/ocr_rec_en.onnx + .models/en_dict.txt) — the multilingual ch_ v3 model reads Latin scripts with badly degraded word spacing (glued words on ordinary English scans) — and Ch selects the ch_ pair (.models/ocr_rec.onnx + .models/ppocr_keys_v1.txt), what the PDF conformance baselines were pinned against; scripts/conformance/pdf_*.sh pin it explicitly by path, which wins over this selector and over the v6 preference.

Variants

En

en_PP-OCRv3 — English-only, proper Latin word spacing.

Ch

ch_PP-OCRv3 — multilingual; the docling-conformance model.

Implementations

impl OcrLang

pub const ACCEPTED: &'static str = "en | ch, or a BCP-47 tag for English or Chinese such as en-US, eng, zh, zh-Hans, zh-TW"

The accepted spellings, for error messages and docs.

pub fn parse(s: &str) -> Option<Self>

Parse a user-supplied language id (#388): the engine’s own codes (en, ch) and BCP-47 tags naming a language one of the two recognizers reads, trimmed and case-insensitive. None for anything else — callers surface their own error/warning.

docling canonicalizes OCR languages across its engines (docling#4075): a bare value is the engine’s native code, an iso:-prefixed value a BCP-47 tag reduced to a language-script pair with the region dropped (zh-CN and zh-Hans are the same recognizer, en-GB is en), and the RapidOCR adapter maps en → its en model and zh-Hans → ch. With only those two PP-OCRv3 pairs on board there is no ambiguity, so the prefix is optional here: en-US, eng, zh, zh-Hans, iso:zh-CN all resolve without a warning. Accepted primary subtags: English as en / ISO 639-2/3 eng / docling’s legacy english; Chinese as the engine code ch (and RapidOCR’s chinese_cht), zh / zho / chi / cmn / legacy chinese / EasyOCR’s ch_sim / ch_tra. Script, region and variant subtags (-Hans, -Hant, -CN, -TW, _US) are ignored: a traditional-script request (zh-Hant, zh-TW) gets the multilingual ch recognizer too, the closest model shipped — upstream would pick RapidOCR’s separate chinese_cht, which this engine does not carry. Genuinely unsupported languages (de, fr, ja, …) parse to None and keep warning.

pub fn from_env() -> Self

The process-level choice from DOCLING_RS_OCR_LANG (empty/unset → the English default; unknown values warn and use English).

Trait Implementations

impl Clone for OcrLang

fn clone(&self) -> Self

Returns a duplicate of the value.

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

Performs copy-assignment from source.

impl Copy for OcrLang

impl Debug for OcrLang

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

Formats the value using the given formatter.

impl Default for OcrLang

fn default() -> Self

Returns the “default value” for a type.

impl Eq for OcrLang

impl PartialEq for OcrLang

fn eq(&self, other: &Self) -> bool

Equality operator ==.

fn ne(&self, other: &Rhs) -> bool

Inequality operator !=.

impl StructuralPartialEq for OcrLang