|
|
| Reporter | zs | |
|---|
| Assigned To | | |
|---|
| Priority | normal | Severity | minor | Reproducibility | always |
|---|
| Status | new | Resolution | open | |
|---|
| Product Version | | |
|---|
| Target Version | | Fixed in Version | | |
|---|
|
|
| Summary | 0000032: loop invariant optimization generates invalid HLDS |
|---|
| Description | The bug can be reproduced by
mmc -Cv check_typeclass.m -Dmost -d145 -d150 -O5
The stage 150 HLDS dump shows that the loop invariant predicate for the is_valid_instance_type predicate has the argument HeadVar__6 in the argument list *twice*, in both cases with the mode unique>>unique, even though the right thing for loop_inv to do would have been not to pass it at all (it is computed *before* the loop is entered).
I think loop_inv might be confused by the fact that the sixth position in the "loop" call is an unwanted output.
|
|---|
| Additional Information | This bug occurs with version 1.117 of check_typeclass.m. I have a diff to make is_valid_instance_type look nicer, but the version with the change does not tickle this bug. |
|---|
| Tags | No tags attached. |
|---|
|
|
| Attached Files |
|
|---|
Relationships 
Relationships