haystack/test/nodes/test_file_converter.py

import os
import subprocess

import pytest

from haystack.nodes import (
    MarkdownConverter,
    DocxToTextConverter,
    PDFToTextConverter,
    PDFToTextOCRConverter,
    TikaConverter,
    AzureConverter,
    ParsrConverter,
)

from ..conftest import SAMPLES_PATH


@pytest.mark.tika
@pytest.mark.parametrize("Converter", [PDFToTextConverter, TikaConverter, PDFToTextOCRConverter])
def test_convert(Converter):
    converter = Converter()
    document = converter.run(file_paths=SAMPLES_PATH / "pdf" / "sample_pdf_1.pdf")[0]["documents"][0]
    pages = document.content.split("\f")

    assert (
        len(pages) != 1 and pages[0] != ""
    ), f'{type(converter).__name__} did return a single empty page indicating a potential issue with your installed poppler version. Try installing via "conda install -c conda-forge poppler" and check test_pdftoppm_command_format()'

    assert len(pages) == 4  # the sample PDF file has four pages.
    assert pages[0] != ""  # the page 1 of PDF contains text.
    assert pages[2] == ""  # the page 3 of PDF file is empty.
    # assert text is retained from the document.
    # As whitespace can differ (\n," ", etc.), we standardize all to simple whitespace
    page_standard_whitespace = " ".join(pages[0].split())
    assert "Adobe Systems made the PDF specification available free of charge in 1993." in page_standard_whitespace


def test_pdftoppm_command_format():
    # Haystack's PDFToTextOCRConverter uses pdf2image, which calls pdftoppm internally.
    # Some installations of pdftoppm are incompatible with Haystack and won't raise an error but just return empty converted documents
    # This test runs pdftoppm directly to check whether pdftoppm accepts the command format that pdf2image uses in Haystack
    proc = subprocess.Popen(
        ["pdftoppm", f"{SAMPLES_PATH}/pdf/sample_pdf_1.pdf"], stdout=subprocess.PIPE, stderr=subprocess.PIPE
    )
    out, err = proc.communicate()
    # If usage info of pdftoppm is sent to stderr then it's because Haystack's pdf2image uses an incompatible command format
    assert (
        not err
    ), 'Your installation of poppler is incompatible with Haystack. Try installing via "conda install -c conda-forge poppler"'


@pytest.mark.parametrize("Converter", [PDFToTextConverter])
def test_pdf_encoding(Converter):
    converter = Converter()

    document = converter.run(file_paths=SAMPLES_PATH / "pdf" / "sample_pdf_2.pdf")[0]["documents"][0]
    assert "ɪ" in document.content

    document = converter.run(file_paths=SAMPLES_PATH / "pdf" / "sample_pdf_2.pdf", encoding="Latin1")[0]["documents"][0]
    assert "ɪ" not in document.content


@pytest.mark.parametrize("Converter", [PDFToTextConverter])
def test_pdf_ligatures(Converter):
    converter = Converter()

    document = converter.run(file_paths=SAMPLES_PATH / "pdf" / "sample_pdf_2.pdf")[0]["documents"][0]
    assert "ff" not in document.content
    assert "ɪ" in document.content

    document = converter.run(file_paths=SAMPLES_PATH / "pdf" / "sample_pdf_2.pdf", known_ligatures={})[0]["documents"][
        0
    ]
    assert "ff" in document.content
    assert "ɪ" in document.content

    document = converter.run(file_paths=SAMPLES_PATH / "pdf" / "sample_pdf_2.pdf", known_ligatures={"ɪ": "i"})[0][
        "documents"
    ][0]
    assert "ff" in document.content
    assert "ɪ" not in document.content


@pytest.mark.tika
@pytest.mark.parametrize("Converter", [PDFToTextConverter, TikaConverter])
def test_table_removal(Converter):
    converter = Converter(remove_numeric_tables=True)
    document = converter.convert(file_path=SAMPLES_PATH / "pdf" / "sample_pdf_1.pdf")[0]
    pages = document.content.split("\f")
    # assert numeric rows are removed from the table.
    assert "324" not in pages[0]
    assert "54x growth" not in pages[0]


@pytest.mark.tika
@pytest.mark.parametrize("Converter", [PDFToTextConverter, TikaConverter])
def test_language_validation(Converter, caplog):
    converter = Converter(valid_languages=["en"])
    converter.convert(file_path=SAMPLES_PATH / "pdf" / "sample_pdf_1.pdf")
    assert "samples/pdf/sample_pdf_1.pdf is not one of ['en']." not in caplog.text

    converter = Converter(valid_languages=["de"])
    converter.convert(file_path=SAMPLES_PATH / "pdf" / "sample_pdf_1.pdf")
    assert "samples/pdf/sample_pdf_1.pdf is not one of ['de']." in caplog.text


def test_docx_converter():
    converter = DocxToTextConverter()
    document = converter.convert(file_path=SAMPLES_PATH / "docx" / "sample_docx.docx")[0]
    assert document.content.startswith("Sample Docx File")


def test_markdown_converter():
    converter = MarkdownConverter()
    document = converter.convert(file_path=SAMPLES_PATH / "markdown" / "sample.md")[0]
    assert document.content.startswith("What to build with Haystack")


def test_azure_converter():
    # Check if Form Recognizer endpoint and credential key in environment variables
    if "AZURE_FORMRECOGNIZER_ENDPOINT" in os.environ and "AZURE_FORMRECOGNIZER_KEY" in os.environ:
        converter = AzureConverter(
            endpoint=os.environ["AZURE_FORMRECOGNIZER_ENDPOINT"],
            credential_key=os.environ["AZURE_FORMRECOGNIZER_KEY"],
            save_json=True,
        )

        docs = converter.convert(file_path=SAMPLES_PATH / "pdf" / "sample_pdf_1.pdf")
        assert len(docs) == 2
        assert docs[0].content_type == "table"
        assert docs[0].content.shape[0] == 4  # number of rows
        assert docs[0].content.shape[1] == 5  # number of columns, Form Recognizer assumes there are 5 columns
        assert list(docs[0].content.columns) == ["", "Column 1", "", "Column 2", "Column 3"]
        assert list(docs[0].content.iloc[3]) == ["D", "$54.35", "", "$6345.", ""]
        assert (
            docs[0].meta["preceding_context"] == "specification. These proprietary technologies are not "
            "standardized and their\nspecification is published only on "
            "Adobe's website. Many of them are also not\nsupported by "
            "popular third-party implementations of PDF."
        )
        assert docs[0].meta["following_context"] == ""

        assert docs[1].content_type == "text"
        assert docs[1].content.startswith("A sample PDF file")


def test_parsr_converter():
    converter = ParsrConverter()

    docs = converter.convert(file_path=str((SAMPLES_PATH / "pdf" / "sample_pdf_1.pdf").absolute()))
    assert len(docs) == 2
    assert docs[0].content_type == "table"
    assert docs[0].content.shape[0] == 4  # number of rows
    assert docs[0].content.shape[1] == 4
    assert list(docs[0].content.columns) == ["", "Column 1", "Column 2", "Column 3"]
    assert list(docs[0].content.iloc[3]) == ["D", "$54.35", "$6345.", ""]
    assert (
        docs[0].meta["preceding_context"] == "specification. These proprietary technologies are not "
        "standardized and their\nspecification is published only on "
        "Adobe's website. Many of them are also not\nsupported by popular "
        "third-party implementations of PDF."
    )
    assert docs[0].meta["following_context"] == ""

    assert docs[1].content_type == "text"
    assert docs[1].content.startswith("A sample PDF file")
    assert docs[1].content.endswith("Page 4 of Sample PDF\n… the page 3 is empty.")