The previous implementation of tests/run-unittests.sh had two bugs:

First, when multiple modules (*.py) were given on the command line, only
the first entry was being used.  Force list context on the $modules list
variable to access all given elements.

Second, when multiple packages were given (test/package/), the last
module of the first had its name concatenated with the first of the
second into a module that did not exist.  Iterate over package modules
to securely form the list of modules to test.

The current implementation works in both cases, and also with a mix of
modules and packages given on the command line.

The previous implementation was using an untracked configuration file
that contained a valid Cloudy path.  Replacement of that file caused
test failures.

Guard against both the configuration file being unedited and containing
an invalid Cloudy path.  Parametrize the Cloudy path used in the example
config file as ui.config.cloudy_path_mock and use consistently in bin/
and tests/test/.

Also modify test suite to work with adm.conf instead of example.conf.
The latter is only created by test/ui/test_config.py, the former can be
used by all other modules, and is expected to be edited.

The function was repeated in several test modules.  Refactor into module
that all can use.

'j:' (joined) is confusing, unless you are reading the source script.
'h:' (hybrid) should be more intuitive to users.

This affects text files produced by visualization.  The strings appear
in comments at the end of data lines.

With the previous implementation, the plots comparing wavelengths with
experimental values treated all entries the same.

Now, wavelengths are distinguished into those that have no theoretical
counterparts (labeled 'Exp'), from theoretical ones that arise between
valence levels ('Valence'), and the rest ('AI').

The former are shown as a sanity check -- the final dataset should not
mess them up.

Also add description of the command adm-create-config.py

Also improve output of command, so that the registration column looks
the same for levels and terms on output to screen.  This is already
the case on LaTeX output.

With the previous implementation, level indices were guaranteed to be
ascending, but they were not updated to the new dataset.  After a cut,
say to exclude isolated levels, each level would retain its index, so
on output, e.g., in a .nrg file, level index 40 could be followed by 43,
assuming levels 41 and 42 were isolated, and therefore excluded.

Now level indices are consecutive integers, under any applied cut.

One implication of this is that the same level may have different
indices in different cuts of the same initial dataset.

Also eliminated infrastructure in adm-export-stout.py for printing out
excluded datalines to output files.  They were previously disabled.

Infrastructure that could be removed remains in
adm.stout.files.FileSections.

Add lower level and upper level indices to the method interface.

In the previous implementation, the indices and the collider type were
drawn from the object that was passed to the method.  The collider is
not added to the interface because it cannot change; it is still drawn
from the object.

The aim is to simplify updating the level indices.  It will prove useful
with exporting scripts.

Now this interface is similar to that of TranList.insert_transition.

In the previous implementation, both export scripts were updating the
indices of the final dataset before producing their output.  It makes
more sense to move these operations to adm-finalize.py.

Update manual to explain the purpose of these updates in the description
of adm-finalize.py

With the previous implementation, all NIST notes were reported in the
.nrg, even for excluded levels (--noai and/or --noisol).  Now excluded
levels are accounted for when compiling list of NIST notes.

Printing the call stack on an abort initiated by ADM is useful during
development, but could be a nuisance to users. Switch it off by default.

Make use of the SHOW_STACK_ON_ABORT option in the configuration file to
control whether the call stack is printed on an ADM abort.  Unhandled
exceptions are not affected by this (the call stack is still reported).

Update all scripts to use this facility.  If the config file does not
exist, it is not parsed and the default option for the stack is used
(off).

Turn all keywords to all caps, and use '=' as a data separator.  This
should inhibit errors like not picking up the entire input, as it did
with the user name.

Remove tests/example.conf -- instead, create it on the fly, and use it
with unit testing.

The previous implementation used only the first word of data in the
'user' field of the configuration file as the name.  Now the entire
input is used.

It is unnecassary to require a file name for the configuration file; ADM
should operate only on the default.  Parameter 'cfile' turned to keyword
argument (made optional).

Also simplified logic of post-processing Cloudy path, and renamed
'resolve' keyword to 'validate_cloudy'.  Useful only for unit testing.

Add default configuration file tests/example.conf, or the unit test
will fail.

With the previous infrastructure, levels were marked as 'connected' if
they participated in any transitions.  Marking them as isolated is far
more intuitive.

These extra levels, enclosed within two 'fields of stars' lines should
be confusing to users of the data, and they are anyway useless.

Disable them in adm-export-stout.py for now.

Consider deprecating and then removing the relevant infrastructure in
stout.files.FileSections.

Output files used strings such as 'matched_levels', while all other
files use 'registered_levels' instead.  Updated these labels in function
adm.register.trans.prepare_WL_scatter_data, where they are set.

Also updated the manual, both TeX and names of files used.

Make it easier to tell if adjacent levels/terms have flipped.

ADAS omits the ancestry terms from its configurations.  It so happens
(e.g., Be+) that duplicate levels may exist, having identical names;
these usually show up as overpopulated terms.

With the previous implementation, Energies.index_for_level reported the
index of the first matching name.  Now it returns a list of indices for
all levels with matching names.

This change cascades, and affects the behavior of other functions and
scripts:

- convert.theory_level_to_experm_index now returns a list.  Called from
  adm-inspect.py, option 'ld' now shows the details of all matching
  levels, not just the first one matching a given name.

- merge.copy_experm_levels reported the first level in the hybrid
  dataset that matched an experimental name.  Now it returns the first
  that is not registered (with either level or term), so adm-merge.py
  reports the right level when it announces that an experimental level
  already exists in the hybrid set.  Obviously, this is innocuous,
  since the experimental level would not be entered anyway, but it is
  good to keep the log sane.

  merge.indices_for_experm_term is likewise affected.

- shifts.get_transition_bounds now returns a list of tuples, instead of
  a single tuple.  These are assembled from all the transitions an
  unregistered level may be involved in.  These are anyway compiled into
  a list by shifts.get_transition_bounds, so there is no cascade of
  changes.  The results of adm-shift-unreg-terms.py invoked with the
  --tnick option are influenced.  (NB NB this has not been tested.)

This changeset was conceived as a means of comparing the properties of
different shifting methods in the registration of Be+, but obviously it
can be used with any atomic models.

It allows calculating the RMS of residuals for various datasets.  TP and
branching ratios are currently implemented, and various cuts of these
datasets are considered.

Comparison of the RMS of homologous datasets should aid with selecting
one model out of those considered.

For each term, the code shows a number of terms both below and above the
one to be shifted, each with their original and hybrid energies.

As it is, users have to do some calculations to compare the suggested
shifts with those of the neighboring terms.

Add to that output the energy difference, and the fractional change.
Show in bold.

Also show in the same format the suggested energy shift for each metric.

Bash function check_style uses wc to count the length of the header line
and the longest line in the file.  However, wc includes newlines in its
counts, while the hardcoded limits do not.  Adjusted here.

With the previous implementation, adm-shift-unreg-terms.py options 't'
and 'r' led to crashes when the term to be shifted was not bracketed
from above.

The strategy now used in such cases is:

- Option 't' minimizes changes to the total Einstein A a term's levels
  are involved in, both to lower and from upper levels.  Absent
  registered upper terms, it is reasonable to use only transitions to
  lower levels instead, option 'l' (for level lifetimes).

- Option 'r' maintains the relative spacing with the bracketing terms.
  Absent registered upper terms, it is sufficient to use instead option
  'c' (for constant spacing to the nearest lower term).

I am moving toward a workflow where the best atomic model is chosen
among a number produced with different shifting methods applied to all
terms consistently.

This is now done via a command-line switch to adm-shift-unreg-terms.py.

Automating this process guarantees that all terms be treated the same in
each run, and also makes execution much more efficient.

The key implementation choice is that adm.ui.upr.__get_user_input is
modified to allow overriding user input acquisition with a preselected
text.

This design choice guarantees that the output produced be identical to
that produced by an equivalent run of manual input entry.

In the previous implementation, shifted terms were identified using
their indices in addition to their configuration and term, but the
retrieval of the experimental levels they were shifted did not use their
indices.

In cases such as in Be+, where some ADAS terms had to be split in two
groups of terms (because it does not include ancestry terms), this
resulted in the first group being reported instead of the second, i.e.,
in unregistered terms being reported as registered.

Supply indices in call to ConfTerms.shifted_with to disambiguate groups.

The script checks if the file exists and offers to skip its creation,
since it can take a while.  However, it was not updated to work with the
synthesis iterations infrastructure, and always pointed to the merged
energy dataset created in the first iteration.

Update file look-up to use the adm.io.synth module, and always point to
the right file.

The footnotes were not updated when the ShiftMethods class was
introduced in adm.idata.nrg, and were still referring to unregistered
terms shown as having no registration info, when they do.

Refactor the code for writing the footnote into a function, and make
it show explanations for all registration modes shown in table.  Use it
from both tex_level_registration and tex_term_registration.

In the previous implementation, Nicks was being accessed directly, even
though it had not been explicitly imported from adm.io.  Prepend all
instances with the 'io.'.

In the previous implementation, the __all_nicks list stored the names of
class attributes, not their values.  This was detected with the first
attribute whose name differs from its value: sort.