You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
constructs.lua short-circuit harness fails at level=4 deep composition #281
The dynamic short-circuit harness at test/lua53_tests/constructs.lua:287-298 builds programs of the form ((((a op b) op c) op d) op e) for every combination of and/or and the basiccases values, compiles each via load(...), runs it, and asserts the result matches a precomputed truth value.
At level=3 (4105 cases): all pass.
At level=4 (the suite default since _soft is unset): the harness eventually fails with a bare assert(false) ("assertion failed!") at roughly the 120000-iteration mark.
The level=3 standalone repro at /tmp/triage_short_circuit.exs (lifted from constructs.lua) is the starting point.
Reclassification from A42
A42 noted this range as "load()-driven short-circuit harness; pending separate triage" but the load() call itself is fine:
load() is installed by the suite runner at tasks/suite_runner.ex:141-156.
A minimal single-case repro (/tmp/triage_constructs_load.exs) passes end-to-end.
The actual failure is in evaluating one of the deeply-composed and/or expressions — an executor short-circuit semantics edge case at depth 4. Possibly register-aliasing under the conditional jump bytecode, possibly a not precedence wrinkle, possibly something else. Not load().
Next steps (not in scope here)
A bisection pass: at level=4, identify the smallest failing cases[n][k] expression and reduce it to a 1-line repro. Then classify and fix or skip.
Activity
test(vm): add on-demand level-4 short-circuit harness
chore(suite): recategorize short-circuit skip and split checkload range