Florian Hatat authored Nov 18, 2018 and hydrargyrum committed Feb 24, 2019

This commit explains in particular how to save the fact that login has
succeeded once by using the browser's logged attribute, so that do_login
isn't called multiple times afterwards.

This reverts commit ead90ebf.

ntome authored Jan 25, 2019 and Romain Bignon committed Feb 05, 2019

Hex-digest was required for some buggy site fixed since then

ntome authored Feb 28, 2018 and Romain Bignon committed Mar 03, 2018

Browser.responses_count is not modified atomically as if there were only
one thread, but the AsyncLoad filter will use multiple threads and
access it concurrently. As a result, multiple responses were written to
the same file.
Add a lock to prevent concurrency on the responses_count field.

Some bad sites have links to upper-cased domains, or with explicit port
(while it's the default port), which forces URL instances to handle
explicitly those cases. Now, the URLs passed to open() are normalized
before being used.

This prevents unsuspecting users giving their logs from being
impersonated.

ntome authored Jul 06, 2017 and Romain Bignon committed Jul 08, 2017

object.__new__ only takes one argument: the class to construct.
It accepts more arguments only if __init__ is not overloaded, but it
is overloaded in Abstract*'s parent classes, so don't pass more args.

In Python 2, it was just deprecated, in Python 3, it doesn't work.

ntome authored Jul 06, 2017 and Romain Bignon committed Jul 08, 2017

When using dir() to find URL fields, we also checked @properties, which
can access other fields that are not defined yet.
Exclude them since they probably won't be URLs anyway.

ntome authored Jul 05, 2017 and Romain Bignon committed Jul 08, 2017

Creating a new type on-the-fly prevents from using isinstance, super and
more.
We can get rid of this by changing the class hierarchy dynamically.

Also, the modules should now pass the Ouiboube instance as a
keyword-argument to the Browser, so the arguments do not have to be
modified between __new__ and __init__.

ntome authored Jul 05, 2017 and Romain Bignon committed Jul 08, 2017

The use of metaclass is not required for constructing the dict of URLs.

Plus, it prevents changing the class hierarchy at runtime, which could
be an improvement for AbstractBrowser.

baptiste delpey authored Jun 21, 2017 and Romain Bignon committed Jul 08, 2017

This reverts commit d0a2a6bbe3054a4cb34cf8d5b0bac7e93ac3d509.

Since those classes store some class-level declared fields (Field and
URL) in a dictionary and are not present as attributes, those fields are
not present in dir() even though __getattr__ will find them.
By reimplementing __dir__, they can be discovered as ordinary
attributes.

raise_from wraps the Python3 syntax "raise exc from cause" which is useful
when an exception is raised to wrap a lower-level exception.

The syntax for using a metaclass is different in python2 and python3,
and there are existing helpers.

Add six as a dependency in setup.py.

TODO this is less efficient on python2

Some functions changed return type from python2 to python3

TODO: not everything is tested, a full review is needed + run is needed