feat(kebab-parse-pdf): P7-1 text PDF extractor — per-page CanonicalDocument

`PdfTextExtractor`(MediaType::Pdf) lopdf 기반 per-page 텍스트 추출.
페이지마다 `Block::Paragraph` + `SourceSpan::Page { page, char_start, char_end }`
emit. 본문이 비거나 추출 panic 인 페이지는 빈 paragraph + `Provenance::Warning`
("scanned candidate") 로 표시 — 이후 OCR fallback (별도 task) 의 입력.

핵심 동작:
- `lopdf::Document::load_mem` + `is_encrypted()` → 암호화 PDF 는 명시 에러
  (`qpdf --decrypt` 안내).
- 페이지 단위 `extract_text(&[page])` 를 `catch_unwind` 로 감싸 malformed
  page panic 을 recoverable warning 으로 변환.
- `/Info` dict 에서 Title/Producer/Creator best-effort 추출. UTF-16BE BOM
  prefixed 문자열도 디코드 (한국어 등 non-ASCII Title 정상 처리).
- 9개 통합 테스트: 3-page emit, scanned-mixed warning, encrypted refuse,
  corrupt header error, page_count 메타, UTF-16BE Title, filename
  fallback, determinism, snapshot.

`parser_version = "pdf-text-v1"`. Allowed deps: `lopdf 0.32` + `pdf-extract 0.7`
(원본 spec 그대로). 본문 다국어 OCR fallback 은 §9.2 후속 task (out of scope).

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>
This commit is contained in:
2026-05-02 08:34:55 +00:00
parent fd89777c83
commit 5a158d7343
9 changed files with 936 additions and 1 deletions

View File

@@ -0,0 +1,228 @@
//! `kebab-parse-pdf` — text PDF extractor (P7-1).
//!
//! Implements [`kebab_core::Extractor`] for [`MediaType::Pdf`]. Extracts
//! text page-by-page via `lopdf`'s per-page API and emits one
//! [`Block::Paragraph`] per page with [`SourceSpan::Page`] (1-based page,
//! `char_start = 0`, `char_end = chars().count()`).
//!
//! Pages where text extraction fails or returns empty get an empty
//! `Block::Paragraph` plus a `Provenance::Warning` flagging the page as
//! a "scanned candidate" — out-of-scope OCR fallback can pick those up.
//!
//! Scope is intentionally narrow: page text + page numbers. Layout
//! reconstruction (multi-column reading order, tables, math), form
//! fields, bookmarks, and OCR for scanned PDFs are explicitly **not**
//! in this task. See `tasks/p7/p7-1-pdf-text-extractor.md`.
//!
//! Per design §3.4 (`SourceSpan::Page` / `Block::Paragraph`),
//! §9.2 (PDF text extraction), §9 versioning.
mod info;
mod page_text;
use anyhow::{Context, Result};
use kebab_core::{
Block, CanonicalDocument, CommonBlock, Extractor, Inline, Lang, MediaType, Metadata,
ParserVersion, Provenance, ProvenanceEvent, ProvenanceKind, SourceSpan, SourceType, TextBlock,
TrustLevel, id_for_block, id_for_doc,
};
use serde_json::{Map, Value};
use time::OffsetDateTime;
pub const PARSER_VERSION: &str = "pdf-text-v1";
/// Text-PDF extractor. Per-page text via `lopdf::Document::extract_text`
/// (the only stable per-page API in the lopdf / pdf-extract pair —
/// pdf-extract 0.7 only exposes whole-document calls).
pub struct PdfTextExtractor;
impl PdfTextExtractor {
pub fn new() -> Self {
Self
}
}
impl Default for PdfTextExtractor {
fn default() -> Self {
Self::new()
}
}
impl Extractor for PdfTextExtractor {
fn supports(&self, m: &MediaType) -> bool {
matches!(m, MediaType::Pdf)
}
fn parser_version(&self) -> ParserVersion {
ParserVersion(PARSER_VERSION.to_string())
}
fn extract(
&self,
ctx: &kebab_core::ExtractContext<'_>,
bytes: &[u8],
) -> Result<CanonicalDocument> {
let asset = ctx.asset;
if !self.supports(&asset.media_type) {
anyhow::bail!(
"kebab-parse-pdf: unsupported media_type for PdfTextExtractor: {:?}",
asset.media_type
);
}
let parser_version = self.parser_version();
let doc_id = id_for_doc(&asset.workspace_path, &asset.asset_id, &parser_version);
// Catastrophic-decode guard via lopdf. `pdf-extract` is intentionally
// not used for parsing here — it only exposes whole-doc text and
// would re-parse the bytes a second time.
let pdf_doc = lopdf::Document::load_mem(bytes)
.context("kebab-parse-pdf: failed to parse PDF (corrupt header or not a PDF)")?;
if pdf_doc.is_encrypted() {
anyhow::bail!(
"kebab-parse-pdf: encrypted PDF; remove encryption (e.g. `qpdf --decrypt`) before ingest"
);
}
let info = info::extract_info(&pdf_doc);
// `get_pages()` returns BTreeMap<u32, ObjectId> with 1-based page
// numbers. We iterate keys in BTreeMap natural order so output is
// deterministic.
let pages = pdf_doc.get_pages();
let page_count = pages.len() as u32;
let now = OffsetDateTime::now_utc();
let mut events: Vec<ProvenanceEvent> = Vec::with_capacity(2 + pages.len());
events.push(ProvenanceEvent {
at: asset.discovered_at,
agent: "kb-source-fs".to_string(),
kind: ProvenanceKind::Discovered,
note: None,
});
events.push(ProvenanceEvent {
at: now,
agent: "kb-parse-pdf".to_string(),
kind: ProvenanceKind::Parsed,
note: Some(format!(
"parser_version={}; page_count={}",
parser_version.0, page_count
)),
});
let mut blocks: Vec<Block> = Vec::with_capacity(pages.len());
for (&page_num, _) in pages.iter() {
let (text, warning) = match page_text::extract_one(&pdf_doc, page_num) {
Ok(t) if !t.trim().is_empty() => (t, None),
Ok(_) => (
String::new(),
Some(format!("page{page_num} empty (scanned candidate)")),
),
Err(e) => (
String::new(),
Some(format!(
"page{page_num} extract failed: {e} (scanned candidate)"
)),
),
};
let char_count = text.chars().count() as u32;
let span = SourceSpan::Page {
page: page_num,
char_start: Some(0),
char_end: Some(char_count),
};
// ordinal = page - 1; saturating_sub guards the (shouldn't-happen)
// case where lopdf hands back a 0-indexed page key.
let ordinal = page_num.saturating_sub(1);
let block_id = id_for_block(&doc_id, "paragraph", &[], ordinal, &span);
let common = CommonBlock {
block_id,
heading_path: Vec::new(),
source_span: span,
};
let inlines = if text.is_empty() {
Vec::new()
} else {
vec![Inline::Text { text: text.clone() }]
};
blocks.push(Block::Paragraph(TextBlock {
common,
text,
inlines,
}));
if let Some(note) = warning {
events.push(ProvenanceEvent {
at: now,
agent: "kb-parse-pdf".to_string(),
kind: ProvenanceKind::Warning,
note: Some(note),
});
}
}
let title = info
.title
.clone()
.filter(|t| !t.trim().is_empty())
.unwrap_or_else(|| {
let fname = filename_from_workspace_path(&asset.workspace_path.0);
strip_extension(&fname)
});
let mut user = Map::new();
let mut pdf_meta = Map::new();
pdf_meta.insert("page_count".into(), Value::Number(page_count.into()));
if let Some(p) = &info.producer {
pdf_meta.insert("producer".into(), Value::String(p.clone()));
}
if let Some(c) = &info.creator {
pdf_meta.insert("creator".into(), Value::String(c.clone()));
}
user.insert("pdf".into(), Value::Object(pdf_meta));
let metadata = Metadata {
aliases: Vec::new(),
tags: Vec::new(),
created_at: asset.discovered_at,
updated_at: asset.discovered_at,
source_type: SourceType::Paper,
trust_level: TrustLevel::Primary,
user_id_alias: None,
user,
};
tracing::debug!(
target: "kebab-parse-pdf",
"extracted PDF doc_id={} workspace_path={} pages={}",
doc_id.0,
asset.workspace_path.0,
page_count
);
Ok(CanonicalDocument {
doc_id,
source_asset_id: asset.asset_id.clone(),
workspace_path: asset.workspace_path.clone(),
title,
lang: Lang("und".to_string()),
blocks,
metadata,
provenance: Provenance { events },
parser_version,
schema_version: 1,
doc_version: 1,
})
}
}
fn filename_from_workspace_path(p: &str) -> String {
p.rsplit('/').next().unwrap_or(p).to_string()
}
fn strip_extension(filename: &str) -> String {
match filename.rfind('.') {
Some(0) => filename.to_string(),
Some(idx) => filename[..idx].to_string(),
None => filename.to_string(),
}
}