GitLab

We add a --format option to "skos-import" ccplugin command with values
being "rdf" or "lcsv".

The lcsv format import behaves a bit differently than RDF import:

* only on file is accepted,
* a concept scheme is created beforehand to hold concepts to be imported
  from the LCSV file,
* this scheme is eventually dropped in case of import failure or
  interruption
* since lcsv_extentities() accepts a file object, we have to handle file
  opening and closing ourselves, hence the extentities_generator()
  wrapper function.

Closes extranet #37463080.

We handle a keyword argument 'stdout_check' in run_import_skos() test
helper and use its value to compare with process stdout. This will be
used in the next changeset.

We handle errors from import code as well as user interruption. This is
mostly in preparation of the next changeset in which we'll add some
cleanup logic.

We'll need to use another file in the next changeset.

Thus making the "delimiter" parameter of LCSV2RDF and lcsv_extentities()
optional.

Flake8 complains about that.

It seems that on form validation error when using the pyramid front-end, the
form values are stored by pyramid in the session cookie. If we don't pop
'stream' first, this silently fails because StringIO are not pickleable (and
_set_cookie fails silently), leading to a redirect to the original form without
any information.

Closes #17106505

it was crashing because of junk inserted in the posted data with an existing
entity (eid, __maineid, etc.)

Closes #17106504

by adding them to the MANIFEST.in file

by first testing it's not an empty rset.

but was randomly defined previously

cnx.session is gone with CubicWeb 3.25, use vreg to get the schema instead.

since this attribute is explicitly displayed for those entity types, and cwuri
use as reference uri rather than 'internal uri' (the default translation).

Related patch has been integrated in 3.19.

File isn't deleted, we'll need it in next cset.

This overrides of abstract class property "uri" in each derived class's
__init__ method is cryptic and there does not appear to be a good reason for
this. So simplify and make this property access a private `_uri_cls` attribute
to be defined in derived classes as any other implementation-dependent
mechanism. The name `_uri_cls` closely matches what's in lcsv module so this
also improves consistency.

Those fails because we ignore it's dev-requirements.

This may cause problem because absolute_url may embed changing part such as
entity's name or title. So, add a `canonical_uri` class method to the
RDFGraphGenerator which by default returns `<base url>/eid` (this is internally
used in cubicweb, it sounds reasonnable that this url scheme will always be
available).

Closes #15922107

In case where we import several files or alike, we may not want to rebuild it
at each round, because it may be costly if eg. the massive store is used (all
indexes may then be removed).