This includes turning generative test into subtest, using standard api for
doctest and some use of bare unittest instead of lgc.testlib as bonus.

else we may run into problem when pkg_resources is used, as it will detect that
a dependancy in the python package isn't installed.

Fix it this way rather than using extras_require because we want the
package to be usable with a simple pip install.

Closes #9343431

Closes #9441457

closes #10041087

Since the binary will disappear in logilab-common 1.3.0.

Before this 'Machin 12' is transformed to 'Any 12' since it expect a 'set type
shortcut', but since the selection doesn't hold any variable, the type
information is lost and no type checking is ever done.

We can not expect to have the schema at this point, which would allow to check
if the type exists. Even if we had, we would then need dedicated structures to
hold the information that we expect entity with eid '12' to be a 'Machin'...

Thus, it seems simpler to disallow usage of this simplified form without any
selected variable.

Closes #9234282.

[__pkginfo__] use yapp2 in requires.txt since that is the one being installed by debian package (and also fits in pip install scenario)

our spec files conflict with filesystem package in centos7: we set permission
755 for /usr/bin and /usr/lib whereas filesystem set it to 555.

So to be compatible with both centos7 and others we simply ignore theses
directories.

pybuild doesn't seem to find any tests.  We should fix that, but, later.

when types are actually yams schema instead of bare string, one ended up with:

  File "/usr/lib/python2.7/dist-packages/rql/nodes.py", line 639, in get_type
    if lhstype == rhstype and not 'Date' in lhstype:
  TypeError: argument of type 'CubicWebEntitySchema' is not iterable

* add support for TZDatetime
* properly raise on unsupported addition (e.g. date+date)
* dates can be added to Interval, not Time

Closes #3248236

Capturing the latter only results in an empty StringIO.

The original patch for bug #1190458 broke tests in CubicWeb. The patch
has now been amended to fix the broken paths (namely, replacing the 'in'
tests with appropriate 'is_equivalent' loops).

These tests are adapted from CubicWeb's own test suite, especially those
that test FTIRANK() on sqlite (the function exists but is replaced by a
no-op).

Related to #1167312

Mostly in comments and defunct tools.
Related to #1167312

WARNING: This is an API break

Among other things we no longer use repr() for String constants, which
means they consistently get quoted with double quotes instead of a mix
of single and double quotes.

Closes #1190507

Related to #1167312

- there is no distinction between PyInt and PyLong anymore
- module initialization changed a bit

Related to #1167312

The existing implementation relies on hash returning different values
for objects that compared equal.  This is horribly wrong.  Instead, stop
implementing comparison, and use the is_equivalent method explicitly.

Closes #298337

Fix regression introduced by switch to dh_python2: it reads our
egg-info/requires.txt to try and guess package dependencies.  This
generates a broken dependency on nonexistent "python-yapps" (instead of
yapps2-runtime) and an unnecessary one on python-logilab-constraint (we
use gecode).