Cannot use set_value_at2 when block is a dynamically allocated integer?

[fsaad]

Here is a sample trace, noting that y has dynamic integer scope and string block, whereas z has string scope and dynamic integer block.

$ cat case.vnts
assume x = mem((i) -> tag(i, "x", normal(0, 1)));
assume y = mem((i) -> tag(i, "y", normal(x(i), 1)));
assume z = mem((i) -> tag("z", i, normal(x(i), 1)));
predict y(1);
predict z(1);

Load up the trace, and try set_value_at2 on the random choices in 1:"y" and "z";1 (both of which exist due to the predict statements in case.vnts.

$ venture -if case.vnts
probcomp-1:/scratch/fsaad/cgpm% venture -if case.vnts 
Tracing models with the Lite backend
Initial random seed is 139951036 (to reproduce run, use -s 139951036)
Venture Script, version 0.5.1a20170816.post69+g9304f69.dirty  http://probcomp.csail.mit.edu/venture/  `help' for help
Venture is *alpha quality* software.
Expect occasional crashes that wipe out your console session.
venture[script] case.vnts > set_value_at2(1, "y", 10)
-62.5691647371
venture[script] case.vnts > sample y(1);
10.0
venture[script] case.vnts > set_value_at2("z", 1, 10)
Warning: skipping annotating did 24, assumed to be from the inference prelude
Warning: skipping annotating did 24, assumed to be from the inference prelude
Warning: skipping annotating did 23, assumed to be from the inference prelude
Warning: skipping annotating did 10, assumed to be from the inference prelude
Warning: skipping annotating did 10, assumed to be from the inference prelude
Warning: skipping annotating did 11, assumed to be from the inference prelude
Warning: skipping annotating did 11, assumed to be from the inference prelude
*** evaluation: Tried to propose 1 values, but subproblem accepts 2 values
(autorun (set_value_at2 "z" 1 10))
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
Caused by
Tried to propose 1 values, but subproblem accepts 2 values
venture[script] case.vnts > 

Hypothesis: The set_value_at2(<scope>, <block>, <value>) procedure will fail if the scope:block identifies more than one random choice. In the test case both 1:"y" and "z";1 actually have two random choices since the invocation of x(i) simulates a normal(0,1), so the semantics imply that set_value_at2 should crash in both cases.

However, set_value_at2 seems to work just fine for the case 1:"y" by setting the "last" random choice, but it fails (as expected, perhaps?) for "z":1.