From ba9b47edf3ccb5a0b6ec283d94b4d32788686489 Mon Sep 17 00:00:00 2001 From: lrnselfreliance <33263708+lrnselfreliance@users.noreply.github.com> Date: Tue, 3 Nov 2026 15:01:37 -0700 Subject: [PATCH] API startup no longer imports pymupdf (#946) pymupdf 2.0 imports pymupdf.layout and onnxruntime, which every API start paid for (about 2.5 s on a Pi 4) although only PDF extraction uses it. It is now imported when a PDF is extracted. Co-authored-by: Claude Opus 5.5 --- modules/docs/extractors.py | 11 +++++------ modules/docs/test/test_extractors.py | 12 ++++++++++++ 2 files changed, 17 insertions(+), 6 deletions(-) diff --git a/modules/docs/extractors.py b/modules/docs/extractors.py index c2418bbe..d114147d 100644 --- a/modules/docs/extractors.py +++ b/modules/docs/extractors.py @@ -25,11 +25,6 @@ except ImportError: ebooklib = None epub = None -try: - import pymupdf -except ImportError: - pymupdf = None - try: import docx except ImportError: @@ -252,7 +247,11 @@ def _parse_pdf_date(date_str: str): def extract_pdf(path: pathlib.Path) -> DocMetadata: """Extract metadata from a PDF using pymupdf.""" - if pymupdf is None: + # Not imported at module level: pymupdf also imports pymupdf.layout and onnxruntime (~2s on a Pi 4), which must + # stay out of API startup. + try: + import pymupdf + except ImportError: logger.error(f'Cannot extract PDF metadata: pymupdf is not installed') return DocMetadata(title=Indexer.get_title(path)) diff --git a/modules/docs/test/test_extractors.py b/modules/docs/test/test_extractors.py index 908fc086..1b49e03c 100644 --- a/modules/docs/test/test_extractors.py +++ b/modules/docs/test/test_extractors.py @@ -1,3 +1,6 @@ +import subprocess +import sys + import pytest from modules.docs.extractors import extract_epub, extract_pdf, extract_metadata, extract_docx, extract_odt, \ @@ -284,3 +287,12 @@ async def test_extract_metadata_dispatches(test_session, test_directory, example metadata = extract_metadata(fg) assert metadata.title == 'WROLPi Test Book' assert metadata.author == 'roland' + + +def test_api_startup_does_not_import_pymupdf(): + """pymupdf imports onnxruntime (~2s on a Pi 4); only extracting a PDF may import it. A subprocess, because + importing `main` registers listeners on the shared app.""" + code = 'import sys, main; print(sorted(i for i in ("pymupdf", "onnxruntime") if i in sys.modules))' + result = subprocess.run([sys.executable, '-c', code], capture_output=True, text=True, timeout=120) + assert result.returncode == 0, result.stderr + assert result.stdout.strip().splitlines()[-1] == '[]'