Repository navigation
Add ORM query autocomplete inside .filter() / .exclude() / .annotate() #3
Description
This is the moat feature identified in competitor analysis: no shipped Django extension does ORM query autocomplete well.
Goal: when the user types inside Post.objects.filter(...), complete field names based on the parsed model index, and resolve lookups like author__name through FK edges.
Design sketch:
- Detect
.filter(/.exclude(/.annotate(call inside a Python file via lightweight regex or textDocument context. - Resolve the QuerySet's base model from the assignment chain (best-effort).
- Suggest field names +
__lookuptraversal from the parsedWorkspaceIndex. - Warn on typos with a diagnostic.
Prior art: the archived django-intellisense extension had a partial implementation worth studying.
Bigger feature — likely v0.5+. Design discussion welcome on Discussions first.
Activity
feat(codeactions): v0.8 Inline QuickFixes for Django ORM anti-patterns
feat(query): v0.8 WOW-feature #2 — Interactive Query Builder
feat(completions): field and lookup completion inside .filter() (#3)
release: 0.13.1 — the listing leads with autocomplete and is searchab…
FROWNINGdev commentedon Aug 20, 2026
Shipped in v0.13.0, live on the Marketplace and Open VSX. Open since day two of the project.
Against the design sketch above
Detect the call — done, and stricter than "lightweight regex". The innermost unclosed
(is found by walking the line with a quote-aware scanner, solabel = 'filter(' + xdoes not open a completion context. Scope isfilter,exclude,get,get_or_create,update_or_create,annotate,aggregate,order_by,values,values_list,only,defer,distinct,create,update. AQ(...)/F(...)/Count(...)wrapper is stepped out of, sofilter(Q(authcompletesauth— notQ.Resolve the base model — best-effort, and deliberately narrow. The literal
<Model>.objects.<method>(shape resolves, including down a chain:Post.objects.filter(published=True).exclude(tiresolves toPost. Anything else returns nothing rather than a guess. An empty list costs the user nothing; a list of fields from the wrong model costs them a debugging session. Assignment-chain inference (qs = Post.objects.all()on an earlier line, thenqs.filter() is the obvious next increment and is not in this release.Fields plus
__traversal — done.author__walks the FK and offers the related model's fields; two edges compose (parent__author__name);ForeignKey('self')resolves back to its own model;pkis offered everywhere even though it is not a declared field. Lookups are typed rather than a flat list —title__offersicontainsbut notgt,views__offersgteandrangebut noticontains,created__addsyear/month/hour/date, andBooleanFieldstays atexact/in/isnull.Warn on typos with a diagnostic — not in this release. A completion that is silent when unsure is safe; a diagnostic that is wrong when unsure is an error squiggle on correct code. That wants the model resolution from (2) to be stronger first. Worth its own issue rather than holding this one open.
Two details that were invisible until they were wrong, both now under test:
_is a trigger character alongside(and,, because a lookup path never leaves a word and VS Code would not re-query afterauthor__on its own; and the replaced range is the whole lookup path rather than VS Code's default word range, which stops at the last__and would landauthor__nameasauthor__author__name.The decision is
completionsAt()insrc/ormCompletions.ts— a pure function over one line of text and the parsed index, novscodeimport, 30 tests, verified against real parser output rather than only a fixture. Suite is at 176, up from 146.New setting
djangoOrmLens.completions.enabled(defaulttrue) for anyone who already has a Python completion source they prefer here.feat(rules): DOL008 — a lookup name that looks misspelled (#3)