mirror of
https://github.com/ArchiveBox/ArchiveBox
synced 2024-11-22 04:03:06 +00:00
93 lines
3.7 KiB
Python
93 lines
3.7 KiB
Python
import subprocess
|
|
import json
|
|
import sqlite3
|
|
|
|
from .fixtures import *
|
|
|
|
def test_depth_flag_is_accepted(process, disable_extractors_dict):
|
|
arg_process = subprocess.run(["archivebox", "add", "http://127.0.0.1:8080/static/example.com.html", "--depth=0"],
|
|
capture_output=True, env=disable_extractors_dict)
|
|
assert 'unrecognized arguments: --depth' not in arg_process.stderr.decode("utf-8")
|
|
|
|
|
|
def test_depth_flag_fails_if_it_is_not_0_or_1(process, disable_extractors_dict):
|
|
arg_process = subprocess.run(
|
|
["archivebox", "add", "--depth=5", "http://127.0.0.1:8080/static/example.com.html"],
|
|
capture_output=True,
|
|
env=disable_extractors_dict,
|
|
)
|
|
assert 'invalid choice' in arg_process.stderr.decode("utf-8")
|
|
arg_process = subprocess.run(
|
|
["archivebox", "add", "--depth=-1", "http://127.0.0.1:8080/static/example.com.html"],
|
|
capture_output=True,
|
|
env=disable_extractors_dict,
|
|
)
|
|
assert 'invalid choice' in arg_process.stderr.decode("utf-8")
|
|
|
|
|
|
def test_depth_flag_0_crawls_only_the_arg_page(tmp_path, process, disable_extractors_dict):
|
|
arg_process = subprocess.run(
|
|
["archivebox", "add", "--depth=0", "http://127.0.0.1:8080/static/example.com.html"],
|
|
capture_output=True,
|
|
env=disable_extractors_dict,
|
|
)
|
|
|
|
archived_item_path = list(tmp_path.glob('archive/**/*'))[0]
|
|
with open(archived_item_path / "index.json", "r", encoding='utf-8') as f:
|
|
output_json = json.load(f)
|
|
assert output_json["base_url"] == "127.0.0.1:8080/static/example.com.html"
|
|
|
|
|
|
def test_depth_flag_1_crawls_the_page_AND_links(tmp_path, process, disable_extractors_dict):
|
|
arg_process = subprocess.run(
|
|
["archivebox", "add", "--depth=1", "http://127.0.0.1:8080/static/example.com.html"],
|
|
capture_output=True,
|
|
env=disable_extractors_dict,
|
|
)
|
|
|
|
conn = sqlite3.connect("index.sqlite3")
|
|
c = conn.cursor()
|
|
urls = c.execute("SELECT url from core_snapshot").fetchall()
|
|
conn.commit()
|
|
conn.close()
|
|
|
|
urls = list(map(lambda x: x[0], urls))
|
|
assert "http://127.0.0.1:8080/static/example.com.html" in urls
|
|
assert "http://127.0.0.1:8080/static/iana.org.html" in urls
|
|
|
|
|
|
def test_overwrite_flag_is_accepted(process, disable_extractors_dict):
|
|
subprocess.run(
|
|
["archivebox", "add", "--depth=0", "http://127.0.0.1:8080/static/example.com.html"],
|
|
capture_output=True,
|
|
env=disable_extractors_dict,
|
|
)
|
|
arg_process = subprocess.run(
|
|
["archivebox", "add", "--overwrite", "http://127.0.0.1:8080/static/example.com.html"],
|
|
capture_output=True,
|
|
env=disable_extractors_dict,
|
|
)
|
|
assert 'unrecognized arguments: --overwrite' not in arg_process.stderr.decode("utf-8")
|
|
assert 'favicon' in arg_process.stdout.decode('utf-8'), 'archive methods probably didnt run, did overwrite work?'
|
|
|
|
def test_add_updates_history_json_index(tmp_path, process, disable_extractors_dict):
|
|
subprocess.run(
|
|
["archivebox", "add", "--depth=0", "http://127.0.0.1:8080/static/example.com.html"],
|
|
capture_output=True,
|
|
env=disable_extractors_dict,
|
|
)
|
|
|
|
archived_item_path = list(tmp_path.glob('archive/**/*'))[0]
|
|
|
|
with open(archived_item_path / "index.json", "r", encoding="utf-8") as f:
|
|
output_json = json.load(f)
|
|
assert output_json["history"] != {}
|
|
|
|
def test_extract_input_uses_only_passed_extractors(tmp_path, process):
|
|
subprocess.run(["archivebox", "add", "http://127.0.0.1:8080/static/example.com.html", "--extract", "wget"],
|
|
capture_output=True)
|
|
|
|
archived_item_path = list(tmp_path.glob('archive/**/*'))[0]
|
|
|
|
assert (archived_item_path / "warc").exists()
|
|
assert not (archived_item_path / "singlefile.html").exists()
|