HarborRAG

Parser smoke checks

parse_file.py parses one real local document through Harbor’s public parser interfaces. It verifies file access, automatic registry routing or an explicit PDF selection, and non-empty extracted content without pytest or test doubles.

Read the shared smoke-test safety and exit-code guidance before using sensitive documents or engines that download models.

Prerequisites

Run commands from the repository root with Python 3.12.
Use representative, non-sensitive files from the target environment.
Install the standard parser dependencies:

uv sync --package harborrag-adapters --extra parsers

For optional PDF profiles and exact backends, install the PDF engines as well:

uv sync --package harborrag-adapters --extra parsers --extra pdf

Some PDF engines require model downloads, native libraries, or suitable CPU/GPU resources. Installation alone does not guarantee that every backend is runnable on the current machine.

parse_file.py does not load dotenv files. If a backend needs credentials, model-cache controls, or OCR settings, use env-example/.env.parser.example as a reference and export the selected variables in the process environment before running the command.

Run automatic routing

Pass a real document path and let HarborParser select the parser:

python packages/harborrag-adapters/tests/parsers/smoke/parse_file.py samples/report.docx
python packages/harborrag-adapters/tests/parsers/smoke/parse_file.py samples/data.xlsx
python packages/harborrag-adapters/tests/parsers/smoke/parse_file.py samples/report.pdf

Maintain a representative corpus with at least one DOCX, PPTX, XLSX, CSV, HTML, EPUB, image, Markdown/text, JSON, text PDF, and scanned PDF.

Select PDF processing

For PDFs, choose either a Harbor profile or one exact backend. The options are mutually exclusive.

python packages/harborrag-adapters/tests/parsers/smoke/parse_file.py \
  samples/report.pdf --pdf-profile fast

python packages/harborrag-adapters/tests/parsers/smoke/parse_file.py \
  samples/scan.pdf --pdf-profile ocr

python packages/harborrag-adapters/tests/parsers/smoke/parse_file.py \
  samples/report.pdf --pdf-backend pymupdf

python packages/harborrag-adapters/tests/parsers/smoke/parse_file.py \
  samples/report.pdf --pdf-backend docling

Supported profiles come from PdfParserProfile. Exact backend choices are docling, liteparse, mineru, paddleocr, and pymupdf. PDF options used with a non-PDF input return exit code 1.

Success criteria and output

The check passes only when parsing completes and extracted content is non-empty. It prints only:

selected parser name;
extracted character count;
element count; and
warning count.

It does not print or persist extracted content.

Exit codes and troubleshooting

Code	Meaning
`0`	Parsing succeeded with non-empty content
`1`	Selection was invalid, the parser failed, or content was empty
`2`	The requested real input file does not exist

Routing failure: confirm the file suffix/content type is supported and the parsers extra is installed.
Backend import/model failure: install the pdf extra and check native, model-cache, memory, CPU/GPU, and network requirements for that backend.
Empty scanned PDF: retry with an OCR-capable profile or exact OCR backend.
Different result in deployment: run from the same Python environment and record the selected parser/backend and installed dependency versions.

This site is open source. Improve this page.