The test case here exhibits a problem relating to the handling of Rc_close_at_apply in Lambda_to_flambda. It was being misinterpreted as "close all regions up to those required by the return continuation" prior to the call -- whereas it should only affect the current region in this regard. I've added some explanations in lambda.mli which are almost entirely due to @stedolan (thanks also to Stephen for other assistance with this bug).
I'm going to think about how we can adjust the test case so it actually reliably fails prior to this PR, and succeeds after.
This patch needs looking over very carefully, as this is a subtle area.
The test case here exhibits a problem relating to the handling of
Rc_close_at_applyinLambda_to_flambda. It was being misinterpreted as "close all regions up to those required by the return continuation" prior to the call -- whereas it should only affect the current region in this regard. I've added some explanations inlambda.mliwhich are almost entirely due to @stedolan (thanks also to Stephen for other assistance with this bug).I'm going to think about how we can adjust the test case so it actually reliably fails prior to this PR, and succeeds after.
This patch needs looking over very carefully, as this is a subtle area.