| Title | Documentation (and test cases) are wrong about mps_collections |
| Status | closed |
| Priority | essential |
| Assigned user | Gareth Rees |
| Organization | Ravenbrook |
| Description | On platform W3I3MV, in the cool variety, finalcv gets stuck with seed 489167363 [1]. This is also reproducible on XCI3LL. |
| Analysis | The documentation for mps_collections [2] says, "Return the number of flips that have taken place in an arena since it was created." But this isn't correct -- it actually returns the epoch, which counts the number of flips *in which some objects might have moved*. But this means that if a trace condemned only segments in non-moving pools, then the epoch was not changed. This causes finalcv to go wrong, because it tests mps_collections to ensure that it terminates, but this is ineffective when testing the non-moving pools like LO. |
| How found | automated_test |
| Evidence | [1] Jenkins build #381. [2] http://www.ravenbrook.com/project/mps/...html/topic/arena.html#c.mps_collections |
| Created by | Gareth Rees |
| Created on | 2016-04-21 11:25:46 |
| Last modified by | Gareth Rees |
| Last modified on | 2016-04-21 12:34:26 |
| History | 2016-04-21 GDR Created. |
| Change | Effect | Date | User | Description |
|---|---|---|---|---|
| 191442 | closed | 2016-04-21 12:34:26 | Gareth Rees | Correct the documentation for mps_collections. In finalcv, don't rely on mps_collections to terminate the test case: use garbage collection messages. |