[skeleton] Ship a tox.ini file upon new cube

[web] fix headers on 304 responses (closes #11466875) The response MUST include the following header fields: [...] - ETag and/or Content-Location, if the header would have been sent in a 200 response to the same request https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html#sec10.3.5 Unfortunately we weren't copying headers from what would have been the 200 response, but from the request.

[web/httpcache] GMTOFFSET no longer exists

[web] fix headers on 304 responses (closes #11466875) The response MUST include the following header fields: [...] - ETag and/or Content-Location, if the header would have been sent in a 200 response to the same request https://www.w3.org/Protocols/rfc2616/rfc2616-sec10.html#sec10.3.5 Unfortunately we weren't copying headers from what would have been the 200 response, but from the request.

[rset] fix infinite recursion introduced in c1eb5a676c80 The mechanism to avoid loops in c1eb5a676c80 "[rset] Always complete attribute/relation caches in ResultSet.get_entity" breaks down: we would loop forever if two entities that were already in the cache were linked by a relation with ? or 1 cardinality in both directions. To avoid that, keep a set of already-considered columns. Related to #9942503.

pep8 unittest_application and unittest_formfields

[rset] Always complete attribute/relation caches in ResultSet.get_entity RQL queries are often designed to fill up the ORM's caches when fetching entities out of the result set. Until now, if an entry already existed in the entity cache, ResultSet.get_entity would return it unchanged, not using the new ResultSet's contents to update the attribute cache, breaking expectations (if the attributes are needed, they'd then be fetched later one at a time, one entity at a time), resulting in loads of DB accesses. So we change ResultSet.get_entity so that: * if the entity is already cached and has been instantiated from the same rset, it is returned as-is (to avoid loops) * if the entity is not yet cached, it is instantiated * if the entity is cached via another rset, its attribute/relation caches are completed Closes #9942503

[rset] refactoring of _build_entity method define a method for entity instantiation and storage in cache (related to #9942503)

[entity] comparing two entities returns NotImplemented Fixes a test when using some versions of unittest2, where assertItemsEqual tries a sort and expects TypeError, which it doesn't get if we directly raise NotImplementedError. The NotImplemented value is treated specially by the interpreter: https://docs.python.org/3.4/library/constants.html#NotImplemented

[test] replace one use of lgc.testlib.unittest_main with unittest.main

[test] stop using TestCase.set_description It's a lgc-ism.

[test] convert a lgc-style generative test into unittest2-style

[web/application] fix query log handling on python3 Writing bytes to a file opened in text mode doesn't work.

[devtools/test] duplicate CubicWebServerTC tests for CubicWebWsgiTC

[devtools/test] fix bytes vs text in test_base_url Needed for python3 with the next changeset

[devtools] skip CubicWebServerTC on python3 People should migrate to wsgi if they're going to use CubicWeb on python3.

[web/test] switch unittest_web from twisted to wsgi And make it pass on python3 while at it.

[devtools] add CubicWebWsgiTC Similar to CubicWebServerTC, except using wsgiref instead of twisted as an http server.

[web/test] return the hash of uploaded files in FileUploadTC instead of their contents json must be unicode, which doesn't go well with arbitrary file contents. Compute the file hash instead, as we know the hexdigest is (ascii) text.

[web/test] open() as a context manager

[web/test] json.loads expects text, not bytes

[wsgi] fix handling of multiple posted files We were only keeping the last received file.

[web/test] Reword doc string for fileupload ajaxfunc

[skeleton] do not import the __pkginfo__.py module to make cubes installable via setuptools When using a simple "import" statement, running 'python setup.py install' for a cube fails in a fresh virtualenv due to the fact the install of the dependencies (here cubicweb iteself) is done by the python process doing the cube's installation, so the __pkginfo__ module from the cube is already loaded (thus used) when installing cubicweb... So we apply the third technique from: https://packaging.python.org/en/latest/single_source_version.html

merge changes from 3.22.2

Added tag 3.22.2, debian/3.22.2-1, centos/3.22.2-1 for changeset 1b93ff37755b

[pkg] 3.22.2

[pkg] fix centos install

[datafeed] do NOT call commit in DataFeedXMLParser.process() (closes #10928592) and so, remove more useless code. Doing a commit in DataFeedXMLParser.process() is a bad idea because this method is called for every imported/updated entity, before handling deletions. So if we commit here, odds are quite high that we're violating some integrity constraint and end up with a ValidationError.

[schema] restore constraint checking when running on old sqlite Old sqlite3 doesn't provide CHECK constraint names in error messages, preventing us from translating a backend integrity error into a ValidationError. This was added in 2012, but the sqlite3 version in RHEL6 is older; so if we run on old sqlite, keep checking the constraints in python rather than only in SQL. Closes #10927494