builder.fill: Make starting sites that are not in the template a warning, and improve the warning message in this case.

See merge request kwant/kwant!244

We raise a warning and remove the sites from 'start', so as to avoid
cryptic 'KeyError's later.

Also take an early exit if there are no starting sites in the template,
to avoid a spurious warning about all starting sites already being
present in the target.

Closes #224.

This is the standard way to re-raise errors and makes the messages
easier to follow.

So it's no longer necessary to provide the REFNAME environment variable.
Also the link template string format is now more general to support at
least gitlab github and cgit.

The former creates a numpy.matrixlib.matrix object, while the latter a regular
numpy array.

Closes #218

'box-forced' is deprecated as of matplotlib 2.2.

Gitlab now has built-in mirroring, so we don't need to have a CI
step for this now.

Closes #207.

pyplot "attaches" a figure to a backend on calling pyplot.figure. Previously we
implemented a hack that worked around this limitation in case the user did not
import pyplot. However this hack is fragile and breaks new things with newer
matplotlib versions. Now we detect whether the figure is intended for pyplot
and import pyplot directly if it is.

This change also allows us to simplify the interface and the implementation of
output_fig, but this is permitted because it is an internal function.

clarify usage of pseudoinverse, and factor out
rank checks for primitive lattice vectors

Closes #209.

See merge request kwant/kwant!231

Rafal Skolasinski authored 6 years ago and Joseph Weston committed 6 years ago

Currently code that do constraints check on "prim_vecs" during lattice
creation is doubled. In addition to factoring it out into separate
function this commit also adds check if "prim_vecs" are linearly independent.

This was always what was meant semantically; newer versions
of Scipy are stricter about what they accept for row/column
indices for COO matrices.

Closes #211.

In the future numpy will not allow multidimensional slices that
are specified as sequences other than tuples.

Closes #210.

Fix compatibility issues with Sympy 1.2.

Closes #203.

See merge request kwant/kwant!228

Rafal Skolasinski authored 6 years ago and Joseph Weston committed 6 years ago

SymPy 1.2 brought changes to LambdaPrinter that broke kwant.continuum
for inputs containing functions like "sin", "cos", or constant "pi".
By this fix we assert that "kwant.continuum.sympify" represents in
its output all functions as AppliedUndef in favour to built-in SymPy
functions. We also put a hard constaint that all functions and constants
in the input expression (with except to pi) must be provided through params.

Backport of commit 499f3d91 to stable branch.
Closes #208.

Pablo Piskunow authored 7 years ago and Joseph Weston committed 6 years ago

  When passing `energy_resolution`, the number of moments is set using
  the yet undefined parameter `self._a`. Now the number of moments is
  set after defining `self._a`.

If 'mplot3d' is not available then plotting 2D systems fails because
'plot' calls 'set_colors', which references 'mplot3d'.

The check is meant to make sure we don't use a hopping length that
is smaller than a certain threshold (10E-5 previously). Here we
correct the check and increase the threshold so as not to have
too fine-grained gridding.

Thanks to Jörg Behrmann for reporting this problem!

Add RELEASE.rst that documents the process of making
a Kwant release.

See merge request kwant/kwant!184

Also include grammatical/stylistic changes, and a re-ordering
of some sections.