backport stable

3.6 api update

[vreg] object_by_id should not have been deprecated. Fix facets. it was *designed* to be able to force an object retreival when selection is potentially not possible. For instance using select instead of object_by_id in get_facet totally breaks the implementation since in get_facet we've not context result set, and we ends up with a NoSelectableObject exception.

fix stupid typo ...

fix condition to paginate a result set

make rollback available in shell (as commit is)

add a reminder

fix indentation / tab pbs

fix name error

add approximated version number to deprecation warning

oops, missing return

we need extrapath when trying to import site_cubicweb files

move extra path computing as a configuration class property so we can reused it

first draft for a simple hooks based custom attribute storage, with a BytesFileSystemStorage POC implementation. Basically: * a dictionary contains maps from which attribute of which entity types are mapped to which custom storage * hooks check for one of these entity type being added/modified/deleted * read is based on the sql generator callback mecanism (used in vcsfile for instance) * all storages have the same basic interface (read, add, update, delete), and should be pluggable in a transparent way (except at migration time when one want to change from a storage to another) * the sample BytesFileSystemStorage: * may store Bytes attributes content of any entity type as file on the file system * is based on one FSPATH rql/sql function and another _fsopen only available in sql * has a dumb file name allocation algorithm

d-t-w

closes #345410: two_lines_rset, two_cols_rset, two_etypes_rset are badly named replaced by multi_lines_rset, multi_columns_rset, multi_etypes_rset, all taking optional 'nb' argument allowing to specify and exact count. Bw compat is kept, usage of deprecated function updated in cw itself. Also drop custom at_least_two_columns selector (in views.plots) since it was the same as multi_columns_rset. Hopefuly the renaming will help in avoid such pb... We should now decide if we want to deprecate one_* selector in favor of multi_*(nb=1). See note on this at then end of selectors.py

fix #636936: read_permission relation should not be mandatory

fix #615848: undefined 'new' variable was refering to the newly introduced subquery / remove_subquery doesn't take an undefine argument (probable copy/paste from the remove_node call)

some docstring

[doc] reformulate dependencies

[views] FinalView should display an empty string if value is None

merge

move base form renderer label handling into a function to reuse it in CompositeFormRenderer and avoid seeing a tuple in column headers...

minor schema deserialization optimizations

oopb, widget has been renamed

specify formvid=edition to get multiple edition working w/ 3.6

reminder

this clutter startup screen for little benefit, remove it

update copyright

move bookmark related uicfg stuff into bookmark specific module, use the EditableURLWidget widget for Bookmark.path

new field's responsibility POC: EditableURLWidget allow to edit Bookmark.path in two separated fields, displaying unquoted values which are requoted on form post processing

refactor field's value retreiving from the widget (eg 'display value' concept): * kill field's display_value method, since getting value specified in a previous post or in req.form should be the widget responsibility (the widget nows what inputs to display, not the field) * new typed_value / format_value on widgets, mainly to ease overriding

dom_id now accepts a suffix argument as input_name

cleaner input_name implementation

when the vregistry is initialized, no need to call vreg.set_schema (else we trigger appobjects registration twice. This allows significant startup speedup.

new url_parse_qs method on base request to parse unicode url formatted query strings

merge

enable cubicweb-ctl db-dump and db-restore on Windows with SQL Server os.system has some issues on Windows, especially regarding whitespace handling, which subprocess handles correctly.

3.6 api update

merge

allow control of generic relations field display through display_fields

no self._cw on widget objects while '_' is defined locally

delete remaining stuff w/ handling of the generic relations

test update

clear vreg's parent_classes cache to fix etype cache / auto-reloading bug EtypeRegistry.parent_classes()'s is cached. When the registry is reloaded, if the cache is not cleared, we might get the old list of parent classes and that leads to strange bugs. Typical scenario : 1/ start application in debug mode 2/ go to add/Bookmark -> everything's fine 3/ change the source 4/ reload add/Bookmark -> 'creation' not selectable because : a/ cw detects a change in the source and reloads vreg b/ cw tries to select 'creation' view for 'Bookmark' - 'creation' view uses specified_etype_implements('Any') - cw compares Bookmark.__bases__[0] (i.e. new version of AnyEntity) with the old version of AnyEntity cached in the vreg. They don't match => view is not applicable.

merge

fix bad indent

backport stable

cleanup

fix bug: the 'table' argument is overwritten by the loop variable, causing latter crash due to this missing information

don't record deambiguifier when it has no type restriction using 'is', so we don't miss later some available constraints

fix warning stack level

cw 3.6 api update

merge

merge

more input/button fixes

backport stable

cleanup

turn all the stuff used to handle 'generic relations' in forms into proper field / widget. -> regroup code from web.request, web.controller, web.views.autoform, web.views.formrenderers, web.views.editcontroller (!) into GenericRelationsField, GenericRelationsWidget in the editviews module (together with the UnrelatedDiv view). So: * almost everything in one place * no more specific behaviour in the form renderer * almost no custom behaviour in autoform (simply add the field when it think it should) Also, the form renderer now display field's value with colspan=2 when field.label is None.

fix tests