No more weird caching issues.

README.md

@@ -1,6 +1,33 @@
 # tldextract
 
-## Python Module [![PyPI version](https://badge.fury.io/py/tldextract.svg)](https://badge.fury.io/py/tldextract) [![Build Status](https://travis-ci.org/john-kurkowski/tldextract.svg?branch=master)](https://travis-ci.org/john-kurkowski/tldextract)
+## CircleUp Fork
+
+This fork fixes an issue with how the original version caches the Public Suffix List (see [PR](https://github.com/john-kurkowski/tldextract/pull/144)).
+
+The upstream version caches one particular version of the list (either with or without private suffixes),
+depending on the settings you use the first time you call it.   All subsequent calls, return using that originally cached list.
+Any calls with different settings than the original silently return incorrect results.
+
+This fork caches the entire list and filters at query time based on the specified arguments.
+
+This change is not backwards compatible.
+
+Changes
+- Moves `include_psl_private_domains` to the `__call__` method.  This is now something you choose on a per-call basis.
+- The entire dataset from publicsuffix.org is saved to cache
+- Added 'source' attribute to named tuple which tells you which suffix list the url was matched against
+- Ensured no weird cache issues happen when using with different `suffix_list_urls` by using different filenames per `suffix_list_urls`
+- Deletes the bundled snapshot
+
+**How to use**
+```python
+import tldextract
+tldextract.extract('http://forums.news.cnn.com/', include_psl_private_domains=True)
+```
+
+The documentation below has been unmodified from upstream.
+
+## Python Module
 
 `tldextract` accurately separates the gTLD or ccTLD (generic or country code
 top-level domain) from the registered domain and subdomains of a URL. For

pypi_deploy.sh

@@ -0,0 +1,4 @@
+devpi use http://pypi.cu
+devpi login root --password=
+devpi use root/circleup
+devpi upload

setup.py

@@ -34,13 +34,13 @@
 LONG_DESCRIPTION_MD = __doc__
 LONG_DESCRIPTION = re.sub(r'(?s)\[(.*?)\]\((http.*?)\)', r'\1', LONG_DESCRIPTION_MD)
 
-INSTALL_REQUIRES = ["setuptools", "idna", "requests>=2.1.0", "requests-file>=1.4"]
+INSTALL_REQUIRES = ["setuptools", "idna", "requests>=2.1.0", "requests-file>=1.4", 'filelock>=3.0.8']
 if (2, 7) > sys.version_info:
     INSTALL_REQUIRES.append("argparse>=1.2.1")
 
 setup(
     name="tldextract",
-    version="2.2.0",
+    version="3.0.1.circleup",
     author="John Kurkowski",
     author_email="john.kurkowski@gmail.com",
     description=("Accurately separate the TLD from the registered domain and "

tldextract/utils.py

@@ -4,6 +4,7 @@
 import os.path
 from hashlib import md5
 
+from filelock import FileLock
 from six import wraps
 
 
@@ -22,15 +23,19 @@ def return_cache(*args, **kwargs):
 
             cache_path = path + '/' + key + '.json'
             cache_path = cache_path.replace('//', '/')
+            lock_path = cache_path + '.lock'
 
-            if not os.path.isfile(cache_path):
-                result = func(*args, **kwargs)
-                make_dir(cache_path)
-                with open(cache_path, 'w') as cache_file:
-                    json.dump(result, cache_file)
+            make_dir(cache_path)
 
-            with open(cache_path) as cache_file:
-                return json.load(cache_file)
+            # without locking, Concurrency could lead to multiple writers, or readers of a partial write
+            with FileLock(lock_path, timeout=20):
+                if not os.path.isfile(cache_path):
+                    result = func(*args, **kwargs)
+                    with open(cache_path, 'w') as cache_file:
+                        json.dump(result, cache_file)
+
+                with open(cache_path) as cache_file:
+                    return json.load(cache_file)
 
         return return_cache