0.22.1

0.22.1 (2026-06-04)

• Refine the "Did you mean" field suggestions so a clear match is no longer
  diluted by unrelated fields that merely share a __suffix — e.g.
  autorzy__cnt now suggests only autorzy__count, not utworzono__month.
  Among candidates above the cutoff, only those within a small similarity margin
  of the best match are kept. New suggest_cutoff / suggest_margin /
  suggest_limit schema attributes tune the behavior.
• Translations: complete and correct the catalogs for all eleven locales —
  translate the messages introduced in 0.22.0 (the "Did you mean" hint, the
  empty-result breakdown strings, the hidden derived-field hints) and fix a few
  stale machine-merged strings (e.g. the numeric numbers type and the time
  "HH:MM[:SS]" format/validation messages).

0.22.0

0.22.0 (2026-06-04)

• The "Unknown field" error now suggests close matches for likely typos:
  Unknown field: autho. Did you mean: author?. Suggestions are matched
  case-insensitively against every field name (including hidden derived fields,
  so book__coun can point at book__count); when nothing is close it
  falls back to the full Possible choices list as before. The matching is
  backed by a new overridable DjangoQLSchema.suggest_field_names() hook.
• Add an empty-result breakdown: when a valid DjangoQL query returns zero
  rows, explain where in the query the data runs out. The new
  djangoql.breakdown.explain_empty(queryset, search, schema=None, *, max_nodes=50) helper walks the validated AST for an arbitrary boolean
  structure (and / or / parentheses), runs one count() per
  sub-expression against the base queryset, and returns a tree of
  {text, count, role, children} nodes flagging the killer AND
  (killer_and) and dead OR branches (dead_or_branch). The admin
  surfaces it automatically on empty results (lazy; disable with
  djangoql_explain_empty = False, tune the cost guard with
  djangoql_explain_empty_max_nodes).
• Add pluggable autocomplete value fields in djangoql.extras
  (AutocompleteField + AutocompleteSchemaMixin). A field's value
  suggestions can come from an arbitrary source — most usefully an existing
  django-autocomplete-light endpoint — letting a user pick an object and filter
  by its primary key. Suggestions are formatted "<label> [<id>]" and the
  field filters <field> = pk (=, != and in supported), with an
  icontains free-text fallback over search_fields. Three providers are
  supported (priority high→low): a url (a DAL endpoint resolved and called
  in-process with the current request), a queryset / callable, or a
  subclass override. SuggestionsAPIView threads the current request into the
  field via a non-breaking set_request hook. AutocompleteSchemaMixin is
  also included in ExtrasSchema. The change is server-side only (no
  JavaScript changes), additive and non-breaking.
• Breaking (derived fields): numeric relation aggregates now use dot
  syntax — <rel>.<numfield>__{sum,avg,min,max} (e.g. book.rating__sum)
  instead of the flat <rel>__<numfield>__sum. This reads consistently with
  DjangoQL's dot navigation and stops the parent model's field list from
  exploding. Relation count keeps its flat name <rel>__count.
• Derived fields are hidden from autocomplete. All generated derived fields
  ship with suggested=False (relation count, date/time parts), and numeric
  aggregates are synthesized on demand so they are never listed at all. They
  remain fully usable in queries.
• The "Unknown field" error no longer dumps the hidden derived fields into
  its "Possible choices" list. It lists only the suggested fields and appends a
  short hint describing the derived-field syntax with a couple of real examples
  for the current model. New overridable schema hooks resolve_unknown() and
  unknown_field_hint() back this behavior.