Repository navigation
v2.x: a rule call with a float literal returns -0.0's row in some runs (float_constant_test flaky) #750
Description
Summary
On main (v2.0.10 line), the answer to a rule call with a float literal argument differs from run to run. tests/float_constant_test.rs::float_constant_matches_equal_floats_only fails in about 6–10% of runs (19/300, 24/300 and 29/300 in three local Linux loops; also failed once in Windows CI on #749, passed on rerun).
Repro
(transact [[:tiny :m/x 0.00000000000000001] [:zero :m/x 0.0] [:neg :m/x -0.0]])
(rule [(xval ?e ?v) [?e :m/x ?v]])
(query [:find ?e :where (xval ?e 0.0)])Most runs return :zero only (the rule call's literal joins by Value's hash on v2.x, which tells -0.0 from 0.0, as the test's comment says). Some runs return :zero and :neg. The same process gives the same answer within one run, so the choice looks seeded by RandomState (map iteration order picking a path).
cargo test -q --test float_constant_test --no-run
B=$(ls -t target/debug/deps/float_constant_test-* | grep -v '\.d$' | head -1)
f=0; for i in $(seq 1 300); do $B -q float_constant_matches_equal_floats_only >/dev/null 2>&1 || f=$((f+1)); done; echo $f/300Failure: assertion left == right failed: a rule call's float argument; left: [":neg", ":zero"], right: [":zero"] at tests/float_constant_test.rs:59.
Notes
- Not caused by fix(query): expression = / != equate a keyword with a ref to its entity (#744) — backport #749 (same rate at its parent commit).
- v3 pairs
0.0with-0.0everywhere (streaming engine 3/9: IndexNestedLoopJoin and HashJoin (#473) #545), so it returns both rows deterministically; this is v2.x only. - Fix direction: find the path that compares the literal by IEEE
==and make the choice deterministic (either answer, always). Candidates: the magic-sets seed for the constant argument (Recursive rule calling another rule fails with INT-022 when its second argument is bound #608) vs the plain rule evaluation.
Found in passing while running CI for #749 (#744 backport).
🤖 Generated with Claude Code
Activity
fix(types): -0.0 and 0.0 hash alike, so a rule call's 0.0 returns -0.…
release: v2.0.11
adityamukho commentedon Oct 26, 2026
Fixed in v2.0.11, now published: https://github.com/project-minigraf/minigraf/releases/tag/v2.0.11. The fix shipped in #752: equal zero floats hash alike, so a rule call returns -0.0 rows in every run. Updated the pinned known-issues list (#421).