papis-extract/papis_extract/__init__.py

154 lines
4.2 KiB
Python
Raw Normal View History

2023-08-28 08:28:06 +00:00
import click
import papis.cli
import papis.config
import papis.document
import papis.logging
import papis.notes
import papis.strings
from papis.document import Document
2023-08-28 08:28:06 +00:00
from papis_extract.annotation import Annotation
from papis_extract import extraction
from papis_extract.exporter import Exporter
from papis_extract.extractors import all_extractors
from papis_extract.exporters import all_exporters
2023-10-17 20:03:35 +00:00
from papis_extract.formatter import Formatter, formatters
2023-08-28 08:28:06 +00:00
logger = papis.logging.get_logger(__name__)
DEFAULT_OPTIONS = {
"plugins.extract": {
"tags": {},
2023-08-28 08:28:06 +00:00
"on_import": False,
2023-08-28 10:55:01 +00:00
"minimum_similarity": 0.75, # for checking against existing annotations
"minimum_similarity_content": 0.9, # for checking if highlight or note
"minimum_similarity_color": 0.833, # for matching tag to color
2023-08-28 08:28:06 +00:00
}
}
papis.config.register_default_settings(DEFAULT_OPTIONS)
@click.command("extract")
@click.help_option("-h", "--help")
@papis.cli.query_argument()
@papis.cli.doc_folder_option()
@papis.cli.git_option(help="Commit changes made to the notes files.")
2023-08-28 08:28:06 +00:00
@papis.cli.all_option()
@click.option(
"--write/--no-write",
"-w",
help="Write extracted annotations into papis notes.",
show_default=True,
)
2023-08-28 08:28:06 +00:00
@click.option(
"--manual/--no-manual",
"-m",
help="Open note in editor for manual editing after annotation extraction.",
show_default=True,
2023-08-28 08:28:06 +00:00
)
@click.option(
"--template",
"-t",
type=click.Choice(
list(formatters.keys()),
case_sensitive=False,
),
help="Choose an output template to format annotations with.",
show_default=True,
2023-08-28 08:28:06 +00:00
)
@click.option(
"--extractor",
"-e",
"extractors",
type=click.Choice(
list(all_extractors.keys()),
case_sensitive=False,
),
default=list(all_extractors.keys()),
multiple=True,
help="Choose an extractor to apply to the selected documents. [default: all]",
)
@click.option(
"--force/--no-force",
"-f",
help="Do not drop any annotations because they already exist.",
show_default=True,
)
2023-08-28 08:28:06 +00:00
def main(
query: str,
# _papis_id: bool,
# _file: bool,
# _dir: bool,
_all: bool,
doc_folder: str,
manual: bool,
write: bool,
extractors: list[str],
template: str,
2023-08-28 08:28:06 +00:00
git: bool,
force: bool,
2023-08-28 08:28:06 +00:00
) -> None:
"""Extract annotations from any documents.
2023-08-28 08:28:06 +00:00
The extract plugin allows manual or automatic extraction of all annotations
contained in the documents belonging to entries of the papis library,
primarily targeting PDF documents currently.
2023-08-28 08:28:06 +00:00
It can write those changes to stdout or directly create and update notes
for papis documents.
It adds a `papis extract` subcommand through which it is invoked, but can
optionally run whenever a new document is imported for a papis entry,
if set in the plugin configuration.
2023-08-28 08:28:06 +00:00
"""
documents = papis.cli.handle_doc_folder_query_all_sort(
query, doc_folder, sort_field=None, sort_reverse=False, _all=_all
)
if not documents:
logger.warning(papis.strings.no_documents_retrieved_message)
return
formatter = formatters.get(template)
run(
documents,
edit=manual,
write=write,
git=git,
formatter=formatter,
extractors=[all_extractors.get(e) for e in extractors],
force=force,
)
2023-08-28 08:28:06 +00:00
def run(
2023-08-28 08:28:06 +00:00
documents: list[Document],
formatter: Formatter | None,
extractors: list[extraction.Extractor | None],
edit: bool = False,
write: bool = False,
git: bool = False,
force: bool = False,
) -> None:
if write:
exporter: Exporter = all_exporters["notes"](
formatter=formatter or formatters["markdown-atx"],
edit=edit,
git=git,
force=force,
)
else:
exporter: Exporter = all_exporters["stdout"](
formatter=formatter or formatters["markdown"]
)
doc_annots: list[tuple[Document, list[Annotation]]] = []
for doc in documents:
annotations: list[Annotation] = []
for ext in extractors:
if not ext:
continue
annotations.extend(extraction.start(ext, doc))
doc_annots.append((doc, annotations))
exporter.run(doc_annots)