blob: a1d6372d051e043e61f0902fa523482bf6dd5442 (
plain) (
tree)
|
|
#!/bin/python
"""
usage: ./convert_papermode_to_metadata.py
This script converts the .paperinfo CSV file in the current directory to an
equivalent .metadata.json file.
ranger used to store metadata in .paperinfo files, but that format was rather
limited, so .metadata.json files were introduced.
"""
import csv
import json
import os
import sys
if sys.version < '3.':
getuserinput = raw_input
else:
getuserinput = input
FIELDS = ["name", "year", "title", "authors", "url"]
def replace(source, target):
if not os.path.exists(source):
print("Source file `%s' doesn't exist, skipping." % source)
return
# Ask for user confirmation if the target file already exists
if os.path.exists(target):
sys.stdout.write("Warning: target file `%s' exists! Overwrite? [y/N]")
userinput = getuserinput()
if not (userinput.startswith("y") or userinput.startswith("Y")):
print("Skipping file `%s'" % source)
return
result = dict()
# Read the input file and convert it to a dictionary
with open(".paperinfo", "r") as infile:
reader = csv.reader(infile, skipinitialspace=True)
for lineno, row in enumerate(reader):
if len(row) != len(FIELDS):
print("skipping invalid row `%s' on line %d" % (row, lineno))
continue
name = row[0]
entry = {}
# Filling up the resulting entry dict
for i, column in enumerate(row[1:]):
if column:
entry[FIELDS[i + 1]] = column
# Adding the dict if it isn't empty
if entry:
result[name] = entry
# Write the obtained dictionary into the target file
if result:
with open(".metadata.json", "w") as outfile:
json.dump(result, outfile, indent=2)
else:
print("Skipping writing `%s' due to a lack of data" % target)
if __name__ == "__main__":
if set(['--help', '-h']) & set(sys.argv[1:]):
print(__doc__.strip())
else:
replace(".paperinfo", ".metadata.json")
|