Repository navigation
bug: magic sets rewriting rejects intermediate variables when first rule arg is bound by literal keyword #297
Description
Description
In v1.2.0, the magic sets rewriting introduced in #289 incorrectly rejects queries where a rule's first argument is bound by a literal keyword at the call site and the rule body contains an intermediate variable (one that appears only in the body, not in the rule head).
Reproduction
(transact [[:a :edge :b]])
(transact [[:b :edge :c]])
(rule [(reach ?x ?y) [?x :edge ?y]])
(rule [(reach ?x ?y) [?x :edge ?mid] (reach ?mid ?y)])
; This works (all variables free):
(query [:find ?y :where (reach ?a ?y)])
; This FAILS with "Unbound variable in rule head: ?mid":
(query [:find ?y :where (reach :a ?y)])
The error Unbound variable in rule head: ?mid is incorrect — ?mid is a valid intermediate (existential) variable in the rule body; it does not need to appear in the head.
Affected versions
Introduced in v1.2.0 (magic sets rewriting, #289). v1.1.1 is not affected.
Workaround
Look up the target entity via a fact pattern and pass the bound variable into the rule:
[?a :some/attr "value"]
(reach ?a ?y)
Impact
Breaks all recursive rule queries where the starting entity is given as a literal keyword, including the reports-to, transfer-chain, and reachable examples in minigraf-playground (steps o3-s2, o3-s3, sl3-s3, t3-s3).
Activity
Merge pull request #300 from project-minigraf/worktree-fix+magic-sets…
chore: bump version to 1.2.1
adityamukho commentedon Sep 26, 2026
Reopening. This still fails in
@minigraf/browser2.0.1.The fix in #300 covers the repro above, where the recursive rule body uses a plain fact pattern. That case now works. It still fails when the recursive rule body calls another user-defined rule instead of a fact pattern.
Repro (2.0.1)
I get the same error on 1.2.0 and on 2.0.1.
Impact
minigraf-playground steps sl3-s3 (
transfer-chainovertransfer-link) and t3-s3 (reachableoversame-line) still need the pattern-bind workaround. The org-chart steps (o3-s2, o3-s3) recurse over a plain fact pattern, and those now work with a literal keyword. See project-minigraf/minigraf-playground#40.Generated by Claude Code
fix(magic-sets): keep preceding rule calls in magic propagation rules (…
Merge pull request #412 from project-minigraf/merge/main-into-v3
release: v2.0.2 — final planned v2.x release