Files
rendercv/scripts/update_entry_figures.py
2025-02-04 18:53:34 -05:00

154 lines
5.2 KiB
Python

"""This script generates the example entry figures and creates an environment for
documentation templates using `mkdocs-macros-plugin`. For example, the content of the
example entries found in
"[Structure of the YAML Input File](https://docs.rendercv.com/user_guide/structure_of_the_yaml_input_file/)"
are coming from this script.
"""
import pathlib
import shutil
import tempfile
import fitz
import pdfCropMargins
import pydantic
import rendercv.data as data
import rendercv.renderer as renderer
repository_root = pathlib.Path(__file__).parent.parent
rendercv_path = repository_root / "rendercv"
image_assets_directory = repository_root / "docs" / "assets" / "images"
class SampleEntries(pydantic.BaseModel):
education_entry: data.EducationEntry
experience_entry: data.ExperienceEntry
normal_entry: data.NormalEntry
publication_entry: data.PublicationEntry
one_line_entry: data.OneLineEntry
bullet_entry: data.BulletEntry
numbered_entry: data.NumberedEntry
reversed_numbered_entry: data.ReversedNumberedEntry
text_entry: str
def render_pngs_from_pdf(pdf_file_path: pathlib.Path) -> list[pathlib.Path]:
"""Render a PNG file for each page of the given PDF file.
Args:
pdf_file_path: The path to the PDF file.
Returns:
The paths to the rendered PNG files.
"""
# check if the file exists:
if not pdf_file_path.is_file():
message = f"The file {pdf_file_path} doesn't exist!"
raise FileNotFoundError(message)
# convert the PDF to PNG:
png_directory = pdf_file_path.parent
png_file_name = pdf_file_path.stem
png_files = []
pdf = fitz.open(pdf_file_path) # open the PDF file
for page in pdf: # iterate the pages
image = page.get_pixmap(dpi=300) # type: ignore
png_file_path = png_directory / f"{png_file_name}_{page.number + 1}.png" # type: ignore
image.save(png_file_path)
png_files.append(png_file_path)
return png_files
def generate_entry_figures():
"""Generate an image for each entry type and theme."""
# Generate PDF figures for each entry type and theme
entries = data.read_a_yaml_file(
repository_root / "docs" / "user_guide" / "sample_entries.yaml"
)
entry_types = entries.keys()
entries = SampleEntries(**entries)
themes = data.available_themes
with tempfile.TemporaryDirectory() as temporary_directory:
# Create temporary directory
temporary_directory_path = pathlib.Path(temporary_directory)
for theme in themes:
design_dictionary = {
"theme": theme,
"page": {
"show_page_numbering": False,
"show_last_updated_date": False,
},
}
for entry_type in entry_types:
# Create data model with only one section and one entry
data_model = data.RenderCVDataModel(
cv=data.CurriculumVitae(
sections={entry_type: [getattr(entries, entry_type)]}
),
design=design_dictionary,
)
# Render
typst_file_path = renderer.create_a_typst_file_and_copy_theme_files(
data_model, temporary_directory_path
)
pdf_file_path = renderer.render_a_pdf_from_typst(typst_file_path)
# Prepare output directory and file path
output_directory = image_assets_directory / theme
output_directory.mkdir(parents=True, exist_ok=True)
output_pdf_file_path = output_directory / f"{entry_type}.pdf"
# Remove file if it exists
if output_pdf_file_path.exists():
output_pdf_file_path.unlink()
# Crop margins
pdfCropMargins.crop(
argv_list=[
"-p4",
"100",
"0",
"100",
"0",
"-a4",
"0",
"-30",
"0",
"-30",
"-o",
str(output_pdf_file_path.absolute()),
str(pdf_file_path.absolute()),
]
)
# Convert PDF to image
png_file_path = render_pngs_from_pdf(output_pdf_file_path)[0]
desired_png_file_path = output_pdf_file_path.with_suffix(".png")
# If image exists, remove it
if desired_png_file_path.exists():
desired_png_file_path.unlink()
# Move image to desired location
png_file_path.rename(desired_png_file_path)
# Remove PDF file
output_pdf_file_path.unlink()
def update_index():
"""Update index.md file by copying README.md file."""
index_file_path = repository_root / "docs" / "index.md"
readme_file_path = repository_root / "README.md"
shutil.copy(readme_file_path, index_file_path)
if __name__ == "__main__":
generate_entry_figures()
print("Entry figures generated successfully.") # NOQA: T201