| Title | No validation tests for client formats. |
| Status | open |
| Priority | nice |
| Assigned user | Richard Brooksby |
| Organization | Ravenbrook |
| Description | We rely on clients writing formats that obey quite a few rules, but the MPS does not validate them (for reasons of speed), nor do we provide any tests to validate them. Recently, Christian Schafmeister spent several days debugging Clasp because he assumed padding objects in his AMC cons pool (holding only two-word cons cells) would also all be two-word objects. The pool therefore skipped into dead objects, causing obscure and hard-to-reproduce errors. |
| Analysis | In cool builds, the MPS could do more to check that formats do the right thing. We could provide a test that checks format behaviour, and instructions for running it. The MPS could perhaps have a validating pool for that purpose. DL suggests that we could scan objects naively (looking at all words) and compare this to whatever the client does — this might instantly reveal underscanning errors. |
| How found | unknown |
| Evidence | Related to RB by DL on 2016-03-19. |
| Created by | Richard Brooksby |
| Created on | 2016-03-19 19:48:05 |
| Last modified by | Gareth Rees |
| Last modified on | 2018-06-27 13:05:05 |
| History | 2016-03-19 RB Created. |