aboutsummaryrefslogtreecommitdiffstats
path: root/python
diff options
context:
space:
mode:
authorAaron LI <aaronly.me@outlook.com>2017-01-19 19:20:04 +0800
committerAaron LI <aaronly.me@outlook.com>2017-01-19 19:20:04 +0800
commit025cdd3ab110c7f06bcd76271e4fcafd7f77c21d (patch)
treea89b584171e242121059a22e48ebc36d731165db /python
parenta1815079ad4ff8076eb33bbd3f6b50e5c1a66763 (diff)
downloadatoolbox-025cdd3ab110c7f06bcd76271e4fcafd7f77c21d.tar.bz2
Add python/adsbib.py
Diffstat (limited to 'python')
-rwxr-xr-xpython/adsbib.py739
-rwxr-xr-x[-rw-r--r--]python/xkeywordsync.py0
2 files changed, 739 insertions, 0 deletions
diff --git a/python/adsbib.py b/python/adsbib.py
new file mode 100755
index 0000000..823a2e8
--- /dev/null
+++ b/python/adsbib.py
@@ -0,0 +1,739 @@
+#!/usr/bin/env python3
+#
+# Copyright (c) 2016 Aaron LI
+# MIT License
+#
+
+"""
+Build and maintain the BibTeX database for a literature of PDF files.
+
+This tool try to extract the metadata from the PDF file, then query
+the SAO/NASA Astrophysics Data System (ADS) for details, and finally
+build the BibTeX entries.
+
+
+References/Credits:
+[1] SAO/NASA Astrophysics Data System
+ https://ui.adsabs.harvard.edu/
+[2] ADS API Description
+ https://github.com/adsabs/adsabs-dev-api
+[3] ADS Search Syntax
+ https://adsabs.github.io/help/search/search-syntax
+[4] PDFtk: https://www.pdflabs.com/docs/pdftk-man-page/
+[5] Poppler utilities: pdftotext
+[6] Pybtex: https://pybtex.org/
+[7] Requests: HTTP for Humans
+ https://github.com/kennethreitz/requests
+"""
+
+
+import os
+import re
+import json
+import fnmatch
+import argparse
+import logging
+import subprocess
+from collections import OrderedDict
+
+import requests
+import pybtex.database
+from pybtex.database import Entry, Person
+from unidecode import unidecode
+
+
+logger = logging.getLogger("adsbib")
+
+
+class AdsRecord:
+ """
+ Record data retrieved from ADS service.
+ """
+ # Regular expression to match the "bibstem" of a bibcode
+ BIBSTEM_RE = re.compile(r"^\d{4}([a-z&]+)[\d.]+[a-z.\d]+$", re.IGNORECASE)
+ # ADS URL pattern to record
+ ADS_URL = "http://adsabs.harvard.edu/abs/%(bibcode)s"
+
+ def __init__(self, data):
+ self.data = data
+ # Supplement "bibstem" if missing
+ self.data["bibstem"] = self.get_bibstem()
+ self.bibcode = data["bibcode"]
+ self.keys = sorted(data.keys())
+
+ def __str__(self):
+ return "<{bibcode} with {keys}>".format(
+ bibcode=self.bibcode, keys=",".join(self.keys))
+
+ def pprint(self, indent=4):
+ """
+ Pretty print the record data
+ """
+ for key in self.keys:
+ print("{indent}{key}: {value}".format(
+ indent=" "*indent, key=key, value=self.data[key]))
+
+ def get_bibstem(self):
+ """
+ Get the "bibstem" from the bibcode, which is the abbreviated name
+ of the journal/publication (e.g., ApJ, A&A)
+ """
+ if "bibstem" in self.data.keys():
+ return self.data["bibstem"]
+ bibcode = self.data["bibcode"]
+ bibstem = self.BIBSTEM_RE.search(bibcode).group(1)
+ return bibstem
+
+ def to_bib_data(self):
+ """
+ Convert to the BibTeX data
+ """
+ bibtype = self.data["doctype"]
+ persons = OrderedDict([
+ ("author", [Person(p) for p in self.data["author"]]),
+ ])
+ fields = OrderedDict([
+ ("title", self.data["title"][0]),
+ ("adsbibcode", self.data["bibcode"]),
+ ("url", self.ADS_URL % {"bibcode": self.data["bibcode"]}),
+ ("journal", self.data["pub"]),
+ ("year", self.data["year"]),
+ ])
+ if self.data["bibstem"].lower() == "arxiv":
+ fields.update([
+ ("archiveprefix", "arXiv"),
+ ("eprint", self.data["identifier"][0]),
+ ])
+ else:
+ fields.update([
+ ("volume", self.data["volume"]),
+ ("page", self.data["page"][0]),
+ ])
+ if self.data.get("doi"):
+ fields["doi"] = self.data["doi"][0]
+ if self.data.get("keyword"):
+ fields["keywords"] = ", ".join(self.data["keyword"])
+ if self.data.get("abstract"):
+ fields["abstract"] = self.data["abstract"]
+ #
+ return (bibtype, persons, fields)
+
+
+class AdsQuery:
+ """
+ SAO/NASA ADS query and access class
+ """
+ # ADS API URLs
+ API_URL = "https://api.adsabs.harvard.edu/v1"
+ QUERY_URL = API_URL + "/search/query"
+
+ # List of fields to return by the query
+ fields = set([
+ "abstract", # abstract of the record
+ "ack", # acknowledgments
+ "aff", # array of the author's affiliations
+ "alternate_bibcode", # list of alternate bibcodes
+ "arxiv", # arXiv ID
+ "arxiv_class", # arXiv class the pre-print was submitted to
+ "author", # array of the author names
+ "bibcode", # canonical ADS bibcode ID
+ "bibgroup", # bibliographic groups the bibcode been associated with
+ "bibstem", # abbreviated name of the journal/publication (e.g., ApJ)
+ "copyright", # copyright
+ "data", # list of sources that have data related
+ "database", # which database the record is associated with
+ "doi", # digital object identifier (DOI)
+ "doctype", # document type, extension to the BibTeX entry types
+ "first_author", # first author of the article
+ "grant", # list of grant IDs and agencies noted
+ "id", # a *non-persistent* ID, for fast look-up of a document
+ "identifier", # array of alternative IDs for the record
+ "issue", # issue number
+ "keyword", # array of normalized and un-normalized keyword values
+ "orcid", # ORCiD IDs supplied by publishers
+ "page", # starting page
+ "property", # array of miscellaneous flags associated
+ "pub", # canonical name of the publication
+ "pubdate", # publication date in the form "YYYY-MM-DD"
+ "title", # title of the record
+ "volume", # volume number
+ "year", # year the record was published
+ ])
+
+ def __init__(self, tokenfile):
+ self.token = open(tokenfile).read().strip()
+ logger.info("Read in ADS API token from: {0}".format(tokenfile))
+
+ def query(self, q, **params):
+ headers = {
+ "Authorization": "Bearer:{0}".format(self.token),
+ }
+ params["q"] = q
+ if "fl" not in params:
+ params["fl"] = ",".join(self.fields)
+ r = requests.get(self.QUERY_URL, params=params, headers=headers)
+ r.raise_for_status()
+ response = json.loads(r.text)
+ logger.debug("ADS response header:\n{0}".format(
+ response["responseHeader"]))
+ logger.debug("ADS response body:\n{0}".format(response["response"]))
+ records = [AdsRecord(data) for data in response["response"]["docs"]]
+ logger.info("ADS found {0} records".format(len(records)))
+ return records
+
+
+class PDF:
+ """
+ Class to deal with the PDF file
+ """
+ # Regular expression to match a DOI
+ DOI_RE = re.compile(r"\b(?:doi[:\s]*|)(10\.\d{4,9}/[^\s\|\]\}\?\,\'\"]+)",
+ re.IGNORECASE)
+ # Regular expression to match a ADS Bibcode
+ BIBCODE_RE = re.compile(r"\b(\d{4}[a-zA-Z0-9&.]{14}[A-Z])")
+ # Regular expression to match a arXiv ID
+ ARXIV_RE = re.compile(r"\b(?:arxiv:?\s*)([a-z/-]*\d{4}\.?\d{3,5}v?\d+)",
+ re.IGNORECASE)
+ # Flag indicating whether the file/metadata is modified
+ modified = False
+
+ def __init__(self, filepath):
+ self.filepath = filepath
+ #
+ self.text_metadata = subprocess.check_output(args=[
+ "pdftk", self.filepath, "dump_data_utf8"
+ ]).decode("utf-8")
+ self.text_firstpage = subprocess.check_output(args=[
+ "pdftotext", "-f", "1", "-l", "1", self.filepath, "-"
+ ]).decode("utf-8")
+ #
+ self.doi = self.extract(self.DOI_RE)
+ self.bibcode = self.extract(self.BIBCODE_RE)
+ self.arxiv = self.extract(self.ARXIV_RE)
+
+ @staticmethod
+ def is_annotated(path):
+ return path.endswith(".annot.pdf")
+
+ @staticmethod
+ def is_annotated_of(path1, path2):
+ """
+ Test whether PDF of path1 is the annotated version of PDF of path2?
+ """
+ basepath1 = os.path.splitext(path1)[0]
+ basepath2 = os.path.splitext(path2)[0]
+ return (basepath1 == basepath2 + ".annot")
+
+ @property
+ def annotated(self):
+ return self.is_annotated(self.filepath)
+
+ def extract(self, regex, grpnum=1):
+ value = None
+ for text in [self.text_metadata, self.text_firstpage]:
+ match = regex.search(text)
+ if match:
+ value = match.group(grpnum)
+ break
+ return value
+
+ def add_metadata(self, key, value, overwrite=False):
+ metadata = self.text_metadata
+ target = "InfoKey: {0}".format(key)
+ idx0 = metadata.upper().find(target.upper())
+ if idx0 >= 0:
+ # Key already exists
+ if overwrite:
+ # Update existing value
+ idx1 = metadata.find("\n", idx0) + 1 # InfoValue's head
+ idx2 = metadata.find("\n", idx1) # InfoValue's tail
+ infoval_new = "InfoValue: {0}".format(value)
+ self.text_metadata = (metadata[:idx1] + infoval_new +
+ metadata[idx2:])
+ self.modified = True
+ logger.info("Updated metadata: {0} = {1}".format(key, value))
+ else:
+ logger.warning("PDF metadata: key already exists!")
+ else:
+ # Append the new key
+ idx0 = metadata.upper().rfind("InfoValue:".upper())
+ idx1 = metadata.find("\n", idx0) + 1
+ keyval_new = "".join([
+ "InfoBegin\n",
+ "InfoKey: {0}\n".format(key),
+ "InfoValue: {0}\n".format(value),
+ ])
+ self.text_metadata = metadata[:idx1] + keyval_new + metadata[idx1:]
+ self.modified = True
+ logger.info("Added new metadata: {0} = {1}".format(key, value))
+
+ def write(self, outfile=None, overwrite=False):
+ """
+ Write PDF with updated metadata.
+ """
+ if not self.modified:
+ logger.info("PDF not modified; skip")
+ return
+ #
+ infile = self.filepath
+ if (outfile is None) or (outfile == infile):
+ # Backup original file
+ infile = self.filepath + ".orig"
+ os.rename(self.filepath, infile)
+ outfile = self.filepath
+ if os.path.exists(outfile):
+ if overwrite:
+ os.remove(outfile)
+ logger.warning("Overwriting file: {0}".format(outfile))
+ else:
+ raise OSError("File already exists: {0}".format(outfile))
+ subprocess.check_output(args=[
+ "pdftk", infile, "update_info_utf8", "-", "output", outfile,
+ ], input=self.text_metadata, universal_newlines=True)
+ logger.info("Saved PDF file: {0}".format(outfile))
+
+
+class BibEntry:
+ """
+ BibTeX database entries.
+ """
+ def __init__(self, filepath, ads):
+ self.filepath = filepath
+ self.ads = ads # ADS querier
+ self.pdf = PDF(self.filepath)
+ self.doi = self.pdf.doi
+ self.bibcode = self.pdf.bibcode
+ self.arxiv = self.pdf.arxiv
+
+ def query_ads(self):
+ """
+ Query the ADS to retrieve the bibliographic data
+ """
+ if self.pdf.bibcode:
+ q = "bibcode:{0}".format(self.pdf.bibcode)
+ elif self.pdf.doi:
+ q = "doi:{0}".format(self.pdf.doi)
+ elif self.pdf.arxiv:
+ q = "arxiv:{0}".format(self.pdf.arxiv)
+ else:
+ raise ValueError("No available metadata in PDF for ADS query")
+ logger.info("ADS query expression: q={0}".format(q))
+ records = self.ads.query(q=q)
+ if len(records) != 1:
+ raise ValueError("Queried NO or MULTIPLE records")
+ self.ads_record = records[0]
+ logger.info("Queried record: {0}".format(str(self.ads_record)))
+
+ def make_bib_data(self):
+ """
+ Make the BibTeX data for this entry from queried ADS record data
+ """
+ bibtype, persons, fields = self.ads_record.to_bib_data()
+ fields["file"] = self.filepath
+ self.bibtype = bibtype
+ self.persons = persons
+ self.fields = fields
+ # Update DOI, BibCode and arXiv ID
+ self.doi = fields.get("doi")
+ self.bibcode = fields.get("adsbibcode")
+ self.arxiv = fields.get("eprint")
+
+ def make_bib_key(self, db=None):
+ """
+ Generate the BibTeX key for this entry from BibTeX data
+ """
+ first_author = self.persons["author"][0]
+ last_name = "".join(first_author.last_names)
+ last_name = unidecode(last_name)
+ last_name = re.sub(r"[ {}`'\"\\]", "", last_name)
+ year = self.fields["year"]
+ journal = self.ads_record.get_bibstem()
+ bibkey = "".join([last_name, year, journal])
+ if db and db.exists_key(bibkey):
+ num = 2
+ while db.exists_key(bibkey+str(num)):
+ num += 1
+ bibkey += str(num)
+ logger.info("Generated BibTeX key: {0}".format(bibkey))
+ self.bibkey = bibkey
+
+ def load_db_data(self, dbentry):
+ self.bibkey = dbentry.key
+ self.bibtype = dbentry.type
+ self.persons = dbentry.persons
+ self.fields = dbentry.fields
+ # Update DOI, BibCode and arXiv ID
+ self.doi = self.fields.get("doi")
+ self.bibcode = self.fields.get("adsbibcode")
+ self.arxiv = self.fields.get("eprint")
+
+ def update_pdf(self):
+ if self.doi:
+ self.pdf.add_metadata("DOI", self.doi)
+ if self.bibcode:
+ self.pdf.add_metadata("AdsBibCode", self.bibcode)
+ if self.arxiv:
+ self.pdf.add_metadata("arXiv", self.arxiv)
+ self.pdf.write()
+
+
+class BibDB:
+ """
+ BibTeX database
+ """
+ modified = False
+
+ def __init__(self, filepath):
+ self.filepath = filepath
+ self.db = pybtex.database.parse_file(filepath)
+ logger.info("Loaded BibTeX database from: {0}".format(filepath))
+ self.doi_bibkeys = self._make_bibkey_invmap("doi")
+ self.bibcode_bibkeys = self._make_bibkey_invmap("adsbibcode")
+ self.arxiv_bibkeys = self._make_bibkey_invmap("eprint")
+
+ def _make_bibkey_invmap(self, key):
+ """
+ Make an inverse map of bibkey to the specified key,
+ e.g., doi, adsbibcode, epring.
+ """
+ invmap = {
+ entry.fields[key]: entry.key
+ for entry in self.db.entries if entry.fields.get(key)
+ }
+ return invmap
+
+ def exists_key(self, bibkey):
+ return (bibkey in self.db.entries)
+
+ def check_file(self):
+ """
+ Check the existence of entry-related file.
+ """
+ raise NotImplementedError
+
+ def contains(self, entry):
+ """
+ Whether the given entry is contained in the database?
+ Return the key of the entry if exists.
+ """
+ for key, invmap in [(entry.bibcode, self.bibcode_bibkeys),
+ (entry.doi, self.doi_bibkeys),
+ (entry.arxiv, self.arxiv_bibkeys)]:
+ bibkey = invmap.get(key)
+ if bibkey:
+ return bibkey
+ return None
+
+ def get_entry(self, bibkey):
+ return self.db.entries[bibkey]
+
+ def update_entry(self, bibkey, entry):
+ """
+ Update the specified entry (e.g., the file path to PDF) if necessary.
+ """
+ dbentry = self.get_entry(bibkey)
+ file_orig = dbentry.fields.get("file", "")
+ if not PDF.is_annotated_of(file_orig, entry.filepath):
+ dbentry.fields["file"] = entry.filepath
+ self.modified = True
+ logger.info("BibTeX DB: updated entry: <{0}>".format(bibkey))
+ else:
+ logger.info("Skip to update BibTeX entry: <{0}>".format(bibkey))
+
+ def add_entry(self, entry):
+ """
+ Add new entry to the database
+ """
+ bibkey = entry.bibkey
+ self.db.add_entry(bibkey,
+ Entry(type_=entry.bibtype, fields=entry.fields,
+ persons=entry.persons))
+ self.modified = True
+ if entry.bibcode:
+ self.bibcode_bibkeys[entry.bibcode] = bibkey
+ if entry.doi:
+ self.doi_bibkeys[entry.doi] = bibkey
+ if entry.arxiv:
+ self.arxiv_bibkeys[entry.arxiv] = bibkey
+ logger.info("BibTeX DB: added new entry: <{0}>".format(bibkey))
+
+ def write(self, outfile=None, overwrite=False):
+ """
+ Write the database if modified.
+ """
+ if not self.modified:
+ logger.info("Database not modified; skip")
+ return
+ #
+ if outfile is None:
+ outfile = self.filepath
+ # Backup original file
+ backfile = outfile + ".orig"
+ os.rename(outfile, backfile)
+ if os.path.exists(outfile):
+ if overwrite:
+ os.remove(outfile)
+ logger.warning("Overwriting file: {0}".format(outfile))
+ else:
+ raise OSError("File already exists: {0}".format(outfile))
+ self.db.to_file(outfile)
+ logger.info("Saved database to file: {0}".format(outfile))
+
+
+class FileIgnore:
+ """
+ Manage the files to be ignored.
+ """
+ def __init__(self, specfile):
+ self.specfile = specfile
+ spec = [line.strip() for line in open(specfile).readlines()]
+ spec = [l for l in spec if (not l.startswith("#")) and (l != "")]
+ # Remove prefix "./"
+ self.spec = [s[2:] if s.startswith("./") else s for s in spec]
+ logger.debug("Ignored specifications:\n{0}".format(
+ "\n".join(self.spec)))
+
+ def is_ignored(self, filepath):
+ for s in self.spec:
+ if fnmatch.fnmatch(filepath, s):
+ logger.debug("Ignored file: {0}".format(filepath))
+ return True
+ return False
+
+
+class FileCollection:
+ """
+ Build the collection of files to be processed.
+ """
+ def __init__(self, inputs, ignorespec=None):
+ self.inputs = inputs
+ paths = self._normalize(self._build(inputs))
+ paths = self._filter(paths, ignorespec)
+ self.paths = sorted(paths)
+ logger.debug("Collection of {0} files:\n{1}".format(
+ len(self.paths), "\n".join(self.paths)))
+
+ @staticmethod
+ def _build(inputs):
+ paths = []
+ for f in inputs:
+ if os.path.isdir(f):
+ # Walk through the directory and get all PDF files
+ # Credit: https://stackoverflow.com/a/2186565/4856091
+ for root, dirnames, filenames in os.walk(f):
+ for filename in fnmatch.filter(filenames, "*.pdf"):
+ paths.append(os.path.join(root, filename))
+ elif f.endswith(".pdf"):
+ paths.append(f)
+ else:
+ # Get the contents as list of files
+ _files = [line.strip() for line in open(f).readlines()]
+ _files = [_f for _f in _files
+ if (not _f.startswith("#")) and (_f != "")]
+ paths += _files
+ return paths
+
+ @staticmethod
+ def _normalize(paths):
+ """
+ Remove the prefix "./" from file paths
+ """
+ paths = [p[2:] if p.startswith("./") else p for p in paths]
+ return paths
+
+ @staticmethod
+ def _filter(paths, ignorespec=None):
+ if ignorespec is None:
+ return paths
+ ignore = FileIgnore(ignorespec)
+ paths = [p for p in paths if not ignore.is_ignored(p)]
+ return paths
+
+
+def cmd_query(args):
+ """
+ Sub-command "query".
+ """
+ ads = AdsQuery(args.token)
+
+ print("ADS query: q={0}".format(args.query))
+ records = ads.query(q=args.query)
+ if len(records) == 0:
+ print("No records found on ADS!")
+ else:
+ for i, rec in enumerate(records):
+ print("\n[%d/%d] %s" % (i+1, len(records), str(rec)))
+ rec.pprint()
+
+
+def cmd_pdf(args):
+ """
+ Sub-command "pdf".
+ """
+ pdf_missing_arxiv = set()
+ pdf_missing_bibcode = set()
+ pdf_missing_doi = set()
+ # Find the PDF files that missing the specified metadata
+ filepaths = FileCollection(args.inputs, args.ignore).paths
+ # Update PDF metadata
+ if (args.set_arxiv or args.set_bibcode or args.set_doi):
+ if len(filepaths) != 1:
+ raise ValueError("only allow ONE input PDF; " +
+ "but gave %d" % len(filepaths))
+ pdf = PDF(filepaths[0])
+ if args.set_arxiv:
+ pdf.add_metadata("arXiv", args.set_arxiv)
+ if args.set_bibcode:
+ pdf.add_metadata("AdsBibCode", args.set_bibcode)
+ if args.set_doi:
+ pdf.add_metadata("DOI", args.set_doi)
+ pdf.write()
+ # Default action: show PDF information
+ for fp in filepaths:
+ print("\n==> PDF: {0}".format(fp))
+ pdf = PDF(fp)
+ print("Annotated: {0}".format(pdf.annotated))
+ if pdf.doi:
+ print("DOI: {0}".format(pdf.doi))
+ else:
+ logger.warning("DOI: missing!")
+ pdf_missing_doi.add(fp)
+ if pdf.bibcode:
+ print("ADS BibCode: {0}".format(pdf.bibcode))
+ else:
+ logger.warning("ADS BibCode: missing!")
+ pdf_missing_bibcode.add(fp)
+ if pdf.arxiv:
+ print("arXiv ID: {0}".format(pdf.arxiv))
+ else:
+ logger.info("arXiv ID: missing!")
+ pdf_missing_arxiv.add(fp)
+ # Show files missing the specified metadata
+ if (args.find_missing_arxiv or
+ args.find_missing_bibcode or
+ args.find_missing_doi):
+ label = [
+ l for f, l in [(args.find_missing_arxiv, "arXiv"),
+ (args.find_missing_bibcode, "BibCode"),
+ (args.find_missing_doi, "DOI")]
+ if f is True
+ ]
+ setlist = [
+ s for f, s in [(args.find_missing_arxiv, pdf_missing_arxiv),
+ (args.find_missing_bibcode, pdf_missing_bibcode),
+ (args.find_missing_doi, pdf_missing_doi)]
+ if f is True
+ ]
+ pdf_missing_metadata = set.intersection(*setlist)
+ print("\n-----------------------------------------------------------")
+ print("{0} PDF files missing metadata: {1}".format(
+ len(pdf_missing_metadata), " && ".join(label)))
+ print("\n".join(pdf_missing_metadata))
+
+
+def cmd_db(args):
+ """
+ Sub-command "db".
+ """
+ db = BibDB(args.bibtex)
+ ads = AdsQuery(args.token)
+ #
+ for filepath in args.input:
+ logger.info("\n==> PDF: {0}".format(filepath))
+ entry = BibEntry(filepath, ads=ads)
+ bibkey = db.contains(entry)
+ if bibkey:
+ logger.info("Already in database: <{0}>".format(bibkey))
+ db.update_entry(bibkey, entry)
+ entry.load_db_data(db.get_entry(bibkey))
+ entry.update_pdf()
+ else:
+ entry.query_ads()
+ entry.make_bib_data()
+ entry.make_bib_key(db)
+ entry.update_pdf()
+ db.add_entry(entry)
+ #
+ db.write()
+
+
+def main():
+ # Default arguments
+ token_dft = ".adstoken"
+ bibtex_dft = "references.bib"
+ ignore_dft = ".adsignore"
+
+ parser = argparse.ArgumentParser(
+ description=("Maintain the BibTeX database of a collection of PDF " +
+ "papers by querying the ADS database"))
+ # Common arguments
+ parser.add_argument("-b", "--bibtex", dest="bibtex", default=bibtex_dft,
+ help="BibTeX database (default: %s)" % bibtex_dft)
+ parser.add_argument("-d", "--debug", dest="debug", action="store_true",
+ help="report verbose debug information")
+ parser.add_argument("-t", "--token", dest="token", default=token_dft,
+ help="file containing the ADS API token " +
+ "(default: %s)" % token_dft)
+ parser.add_argument("-I", "--ignore", dest="ignore", default=ignore_dft,
+ help="ignore specification file " +
+ "(default: %s)" % ignore_dft)
+ subparsers = parser.add_subparsers(dest="subparser_name",
+ title="sub-commands",
+ help="additional help")
+ # Sub-command: "query"
+ parser_query = subparsers.add_parser("query", help="Query the ADS service")
+ parser_query.add_argument("query", help="query expression")
+ parser_query.set_defaults(func=cmd_query)
+ # Sub-command: "pdf"
+ parser_pdf = subparsers.add_parser("pdf", help="Manipulate PDF files")
+ parser_pdf.add_argument("--find-missing-arxiv",
+ dest="find_missing_arxiv",
+ action="store_true",
+ help="find PDF files missing arXiv ID")
+ parser_pdf.add_argument("--find-missing-bibcode",
+ dest="find_missing_bibcode",
+ action="store_true",
+ help="find PDF files missing ADS BibCode")
+ parser_pdf.add_argument("--find-missing-doi",
+ dest="find_missing_doi",
+ action="store_true",
+ help="find PDF files missing DOI")
+ parser_pdf.add_argument("-a", "--set-arxiv", dest="set_arxiv",
+ help="set arXiv ID in PDF metadata")
+ parser_pdf.add_argument("-b", "--set-bibcode", dest="set_bibcode",
+ help="set ADS BibCode in PDF metadata")
+ parser_pdf.add_argument("-d", "--set-doi", dest="set_doi",
+ help="set DOI in PDF metadata")
+ parser_pdf.add_argument("inputs", nargs="+",
+ help=("PDF files, directory, or text file of " +
+ "filenames. NOTE: only one PDF file " +
+ "when to update its metadata."))
+ parser_pdf.set_defaults(func=cmd_pdf)
+ # Sub-command: "db"
+ parser_db = subparsers.add_parser("db", help="Maintain the BibTeX DB")
+ parser_db.add_argument("-s", "--show", dest="show", action="store_true",
+ help="show database status")
+ parser_db.add_argument("-c", "--check-pdf", dest="check_pdf",
+ action="store_true",
+ help="check the database against local PDF files")
+ parser_db.add_argument("-a", "--add", dest="add", nargs="+",
+ help="add PDF files to the database")
+ parser_db.set_defaults(func=cmd_db)
+ #
+ args = parser.parse_args()
+ if (args.ignore == ignore_dft) and (not os.path.exists(ignore_dft)):
+ args.ignore = None
+
+ # Setup logging
+ loglevel = "DEBUG" if args.debug else "INFO"
+ logging.basicConfig(level=loglevel,
+ format="[%(levelname)s] <%(name)s> %(message)s")
+
+ logger.debug("Parsed arguments:\n{0}".format(args))
+
+ # Dispatch sub-commands to call its associated default function
+ args.func(args)
+
+
+if __name__ == "__main__":
+ main()
diff --git a/python/xkeywordsync.py b/python/xkeywordsync.py
index 9a7cd79..9a7cd79 100644..100755
--- a/python/xkeywordsync.py
+++ b/python/xkeywordsync.py