tricky...
- the only way I found to express the concept of the "associated type"
(in Rust lingo) model_type was by making Type generic over its value
and null types.
- in addition, the class hierarchy of Integer and Float types had to be
modified, since previously some of them would have conflicting null
types relative to their super class (this required a change to the
edit plugin; hopefully no more breakage is caused by these changes)
- don't import the query module, but only the relevant Query's to avoid
confusing the module query and the class variable query
This works around a bug that does not exist in Python 3.x, and the
workaround (by calling the underlying shlex.split function with bytes)
was causing crashes on some versions of Python 3. Seemed to work fine on
3.10-dev, though, oddly.
Previously, if continuing to edit (i.e. invoking the $EDITOR) multiple
times in one invocation of EditPlugin.edit_objects, the plugin would
reload the old state from the file tags. The initial 'old state' is
usually only loaded from the database, though. As a consequence, if
database and tags were not in sync, the diffs from first and all
subsequent edits could differ unexpectedly.
* Revert the changes related to allowing the album- and item-level
fields to be edited at the same time, as the increase in complexity
was deemed excesive during review.
* Modify the interactive execution so temporary Item.id's are used,
removing the extra functionality needed for dealing with both id and
path as reference fields.
* Docstrings and comments cleanup.
* Add support for editing both the item fields and the album fields in
a single YAML file, by appending an Album-like object as the front of
the objects to be edited.
* The FakeAlbum class provides that object, mimicking the original
Album behaviour and including an _apply_changes() method that propagates
the changes read from the yaml onto the Items.
* Modify edit_objects() so the flattening of the objects takes into
account the type of object, using different fields for Albums and for
Items.
* Renamed apply() to apply_() to prevent an IDE warning about reusing a
reserved built-in symbol.
* Make the edit plugin return action.RETAG when invoked during an
interactive import session, making the importer handle the writing of
the tags to the files (if needed) properly.
* Move the logic relative to the "reference field" to
_set_reference_field(), simplifying a bit the functions that depend on
this field.
* Hide the "edit Candidates" choice if no candidates are found.
* Initial draft for invoking the edit plugin during an importer session.
* Add prompt choices for editing the original file tags ("eDit") and
apply a candidate and then edit ("edit Candidates").
* Modify plugin (_get_fields, apply_data, edit_objects) so "path" can be
used as a reference field instead of "id", as the Items are not still on
the database when the plugin is invoked via the importer.
* Modify ImportTask.manipulate_files() with a temporary flag for writing
the item tags even if ASIS was selected.
This avoids some round-tripping problems with types (such as ScaledInt) that
are not represented in strings with 100% fidelity. It also makes the syntax
nicer when editing numbers and booleans: they no longer appear to be
needlessly surrounded by quotes in the YAML.
I hadn't quite realized before that the user could also change the *keys* to
be non-strings too! This also prevents against that by just reinterpreting
everything as strings.
We now ask for a trinary edit/apply/cancel confirmation *after* showing the
updates. This lets you decide whether you're done based on a "preview" of the
changes and keep editing if they don't look right.
