ArchiveBox/archivebox/index/csv.py

38 lines
934 B
Python

__package__ = 'archivebox.index'
from typing import List, Optional, Any
from archivebox.misc.util import enforce_types
from .schema import Link
@enforce_types
def links_to_csv(links: List[Link],
cols: Optional[List[str]]=None,
header: bool=True,
separator: str=',',
ljust: int=0) -> str:
cols = cols or ['timestamp', 'is_archived', 'url']
header_str = ''
if header:
header_str = separator.join(col.ljust(ljust) for col in cols)
row_strs = (
link.to_csv(cols=cols, ljust=ljust, separator=separator)
for link in links
)
return '\n'.join((header_str, *row_strs))
@enforce_types
def to_csv(obj: Any, cols: List[str], separator: str=',', ljust: int=0) -> str:
from .json import to_json
return separator.join(
to_json(getattr(obj, col), indent=None).ljust(ljust)
for col in cols
)