import pytest from sandcrawler import gen_file_metadata, b32_hex, parse_cdx_line def test_gen_file_metadata(): # valid (but very small) PDF file with open('tests/files/dummy.pdf', 'rb') as f: file_meta = gen_file_metadata(f.read()) assert file_meta == { 'mimetype': 'application/pdf', 'md5hex': '2942bfabb3d05332b66eb128e0842cff', 'sha1hex': '90ffd2359008d82298821d16b21778c5c39aec36', 'sha256hex': '3df79d34abbca99308e79cb94461c1893582604d68329a41fd4bec1885e6adb4', 'size_bytes': 13264, } # valid HTML fm = gen_file_metadata( b"""dummyhtml document""") assert fm['mimetype'] == 'text/html' # bogus text fm = gen_file_metadata(b"asdf1234") assert fm['mimetype'] == 'text/plain' assert fm['size_bytes'] == 8 def test_b32_hex(): # valid b32 assert b32_hex('sha1:TZCYZ2ULEHYGESS4L3RNH75I23KKFSMC') == '9e458cea8b21f0624a5c5ee2d3ffa8d6d4a2c982' assert b32_hex('TZCYZ2ULEHYGESS4L3RNH75I23KKFSMC') == '9e458cea8b21f0624a5c5ee2d3ffa8d6d4a2c982' # sha1hex pass-through s = 'bda3c1017d52e826bbd1da51efad877272d300f9' assert b32_hex(s) == s # invalid with pytest.raises(ValueError): assert b32_hex('blah') == 'blah' def test_parse_cdx_line(): raw = "edu,upenn,ldc)/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf 20170828233154 https://www.ldc.upenn.edu/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf application/pdf 200 WL3FEA62TEU4F52Y5DOVQ62VET4QJW7G - - 210251 931661233 SEMSCHOLAR-PDF-CRAWL-2017-08-04-20170828231135742-00000-00009-wbgrp-svc284/SEMSCHOLAR-PDF-CRAWL-2017-08-04-20170828232253025-00005-3480~wbgrp-svc284.us.archive.org~8443.warc.gz" correct = { 'sha1b32': "WL3FEA62TEU4F52Y5DOVQ62VET4QJW7G", 'sha1hex': "b2f65203da9929c2f758e8dd587b5524f904dbe6", 'mimetype': "application/pdf", 'surt': "edu,upenn,ldc)/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf", 'url': "https://www.ldc.upenn.edu/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf", 'datetime': "20170828233154", 'warc_path': "SEMSCHOLAR-PDF-CRAWL-2017-08-04-20170828231135742-00000-00009-wbgrp-svc284/SEMSCHOLAR-PDF-CRAWL-2017-08-04-20170828232253025-00005-3480~wbgrp-svc284.us.archive.org~8443.warc.gz", 'warc_offset': 931661233, 'warc_csize': 210251, 'http_status': 200, } assert parse_cdx_line(raw) == correct assert parse_cdx_line(raw + "\n") == correct assert parse_cdx_line(raw + " extra_field") == correct def test_invalid_cdx(): print("missing warc") raw = "edu,upenn,ldc)/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf 20170828233154 https://www.ldc.upenn.edu/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf application/pdf 200 WL3FEA62TEU4F52Y5DOVQ62VET4QJW7G - - 210251 931661233 -" assert parse_cdx_line(raw) == None print("bad datetime") raw = "edu,upenn,ldc)/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf 2070828233154 https://www.ldc.upenn.edu/sites/www.ldc.upenn.edu/files/medar2009-large-arabic-broadcast-collection.pdf application/pdf 200 WL3FEA62TEU4F52Y5DOVQ62VET4QJW7G - - 210251 931661233i SEMSCHOLAR-PDF-CRAWL-2017-08-04-20170828231135742-00000-00009-wbgrp-svc284/SEMSCHOLAR-PDF-CRAWL-2017-08-04-20170828232253025-00005-3480~wbgrp-svc284.us.archive.org~8443.warc.gz" assert parse_cdx_line(raw) == None