introduce MAX_REDIRECTS config setting and fix urllib3 redirect han…

…dling

Fixes issue adbar#450

After setting `MAX_REDIRECTS` to 5, I could fetch the original URL from the issue:
`trafilatura -u https://www.hydrogeninsight.com/production/breaking-us-reveals-the-seven-regional-hydrogen-hubs-to-receive-7bn-of-government-funding/2-1-1534596`

I also fixed this old issue: adbar#128
The underlying urllib3 bug has not been fixed: urllib3/urllib3#2475

I had to pass the retry strategy to the actual request method: it doesn't propagate from the pool maanger

docs/settings.rst

@@ -43,6 +43,7 @@ The default file included in the package is `settings.cfg <https://github.com/ad
    * ``EXTENSIVE_DATE_SEARCH = on`` set to ``off`` to deactivate ``htmldate``'s opportunistic search (lower recall, higher precision)
 - Navigation
    * ``EXTERNAL_URLS = off`` do not take URLs from other websites in feeds and sitemaps (CLI mode)
+   * ``MAX_REDIRECTS = 2``: maximum number of `URL redirections <https://en.wikipedia.org/wiki/URL_redirection>`_ to be followed. Set to 0 to not follow any redirection.
 
 
 Using a custom file on the command-line

tests/downloads_tests.py

@@ -33,7 +33,7 @@
                                    _send_pycurl_request, _send_request,
                                    _urllib3_is_live_page,
                                    add_to_compressed_dict, fetch_url,
-                                   is_live_page, load_download_buffer)
+                                   is_live_page, load_download_buffer, _reset_global_objects)
 from trafilatura.settings import DEFAULT_CONFIG, use_config
 from trafilatura.utils import decode_response, load_html
 
@@ -94,7 +94,17 @@ def test_fetch():
     assert load_html(response) is not None
     # nothing to see here
     assert extract(response, url=response.url, config=ZERO_CONFIG) is None
-
+    # test handling redirects
+    res = fetch_url('https://httpbun.com/redirect/2')
+    assert len(res) > 100  # We followed redirects and downloaded something in the end
+    new_config = use_config()  # get a new config instance to avoid mutating the default one
+    # Patch max directs: limit to 0. We won't fetch any page as a result
+    new_config.set('DEFAULT', 'MAX_REDIRECTS', '0')
+    _reset_global_objects()  # force Retry strategy and PoolManager to be recreated with the new config value
+    res = fetch_url('http://httpbin.org/redirect/1', config=new_config)
+    assert res is None
+    # Also test max redir implementation on pycurl
+    assert _send_pycurl_request('http://httpbin.org/redirect/1', True, new_config) is None
 
 def test_config():
     '''Test how configuration options are read and stored.'''

trafilatura/downloads.py

@@ -43,7 +43,6 @@
 PKG_VERSION = version("trafilatura")
 
 NUM_CONNECTIONS = 50
-MAX_REDIRECTS = 2
 
 urllib3.disable_warnings(urllib3.exceptions.InsecureRequestWarning)
 HTTP_POOL = None
@@ -90,8 +89,8 @@ def _send_request(url, no_ssl, config):
     global HTTP_POOL, NO_CERT_POOL, RETRY_STRATEGY
     if not RETRY_STRATEGY:
         RETRY_STRATEGY = urllib3.util.Retry(
-            total=0,
-            redirect=MAX_REDIRECTS, # raise_on_redirect=False,
+            total=config.getint("DEFAULT", "MAX_REDIRECTS"),
+            redirect=config.getint("DEFAULT", "MAX_REDIRECTS"), # raise_on_redirect=False,
             connect=0,
             backoff_factor=config.getint('DEFAULT', 'DOWNLOAD_TIMEOUT')/2,
             status_forcelist=[
@@ -107,13 +106,13 @@ def _send_request(url, no_ssl, config):
             if not HTTP_POOL:
                 HTTP_POOL = urllib3.PoolManager(retries=RETRY_STRATEGY, timeout=config.getint('DEFAULT', 'DOWNLOAD_TIMEOUT'), ca_certs=certifi.where(), num_pools=NUM_CONNECTIONS)  # cert_reqs='CERT_REQUIRED'
             # execute request
-            response = HTTP_POOL.request('GET', url, headers=_determine_headers(config))
+            response = HTTP_POOL.request('GET', url, headers=_determine_headers(config), retries=RETRY_STRATEGY)
         else:
             # define pool
             if not NO_CERT_POOL:
                 NO_CERT_POOL = urllib3.PoolManager(retries=RETRY_STRATEGY, timeout=config.getint('DEFAULT', 'DOWNLOAD_TIMEOUT'), cert_reqs='CERT_NONE', num_pools=NUM_CONNECTIONS)
             # execute request
-            response = NO_CERT_POOL.request('GET', url, headers=_determine_headers(config))
+            response = NO_CERT_POOL.request('GET', url, headers=_determine_headers(config), retries=RETRY_STRATEGY)
     except urllib3.exceptions.SSLError:
         LOGGER.warning('retrying after SSLError: %s', url)
         return _send_request(url, True, config)
@@ -275,7 +274,7 @@ def _send_pycurl_request(url, no_ssl, config):
     curl.setopt(pycurl.HTTPHEADER, headerlist)
     # curl.setopt(pycurl.USERAGENT, '')
     curl.setopt(pycurl.FOLLOWLOCATION, 1)
-    curl.setopt(pycurl.MAXREDIRS, MAX_REDIRECTS)
+    curl.setopt(pycurl.MAXREDIRS, config.getint('DEFAULT', 'MAX_REDIRECTS'))
     curl.setopt(pycurl.CONNECTTIMEOUT, config.getint('DEFAULT', 'DOWNLOAD_TIMEOUT'))
     curl.setopt(pycurl.TIMEOUT, config.getint('DEFAULT', 'DOWNLOAD_TIMEOUT'))
     curl.setopt(pycurl.NOSIGNAL, 1)
@@ -327,3 +326,12 @@ def _send_pycurl_request(url, no_ssl, config):
     # tidy up
     curl.close()
     return RawResponse(bufferbytes, respcode, effective_url)
+
+
+def _reset_global_objects():
+    """
+    Force global objects to be re-created
+    Currently only used during tests
+    """
+    global HTTP_POOL, NO_CERT_POOL, RETRY_STRATEGY
+    HTTP_POOL, NO_CERT_POOL, RETRY_STRATEGY = None, None, None

trafilatura/settings.cfg

@@ -12,6 +12,8 @@ SLEEP_TIME = 5
 USER_AGENTS =
 # cookie for HTTP requests
 COOKIE =
+# Maximum number of redirects that we will follow
+MAX_REDIRECTS = 5
 
 # Extraction
 MIN_EXTRACTED_SIZE = 250