Bugzilla – Full Text Bug Listing
| Summary: | [planner] Encoding stability | ||
|---|---|---|---|
| Product: | [Eclipse Project] Equinox | Reporter: | Pascal Rapicault <pascal> |
| Component: | p2 | Assignee: | Daniel Le Berre <leberre> |
| Status: | RESOLVED FIXED | QA Contact: | |
| Severity: | normal | ||
| Priority: | P3 | CC: | john.arthorne, leberre |
| Version: | 3.5 | ||
| Target Milestone: | 3.5 M7 | ||
| Hardware: | PC | ||
| OS: | Mac OS X - Carbon (unsup.) | ||
| Whiteboard: | |||
|
Description
Pascal Rapicault
We have code in the Projector (~ line 147) that sorts the IUs when in debug mode. I have commented out this code and run again with this and see even worst number because we now have to sort... I'm currently not comfortable releasing this because everyone takes a hit. Daniel, could you describe how much time does the solver spend finding that there is a conflict, over the time spent computing the explanation. I think this would help shed some lights. Detection that there is a conflict is immediate (it is done without search in those examples, just by unit propagation). The time spent is on applying the explanation algorithm: Real explanation time:800 Time to compute conflict: 1329 Real explanation time:2343 Time to compute conflict: 2874 Real explanation time:1217 Time to compute conflict: 1691 Real explanation time:13908 Time to compute conflict: 14387 It takes roughly 500ms to translate the computed explanation into Status, etc. I agree that sorting the IUs might not be the right solution. What is important in the explanation algorithm is the order of the extra variables used for the explanation computation. I will investigate tonight if there is a way to have a better order than what we have now. BTW, I will need for that purpose an order on the possible explanations. (that way, I could for the order of the explanation analysis, independently of the constraints order). There are only two reasons for a conflict: - I ) a missing requirement - II) a violated singleton constraint For the second case, there are two possible situations: - II.1) it appears between one already installed IU and one that the user would like to install - II.2) it appears between two IU to install In case I and II.1, it means that the IU to install involved in the conflict cannot be installed while in II.2 one of the two IU cannot be installed. So I consider that the preferences in the explanations should be: I/II.1/II.2 I prefer to gather first the missing requirements because the explanation is often simple. Then I prefer the singleton violation between an installed IU and an IU to install because I know that the IU cannot be installed. Finally, I am interested in top conflicting IUs, but I cannot decide which one should be dropped. Does it makes sense? With the new set of SAT4J jars (v20090308) I have the following results: Time to compute conflict: 888 Time to compute conflict: 1009 Time to compute conflict: 6815 Time to compute conflict: 4087 I don't think we have more to do on this. Do you agree Daniel? If so, please close. The current behavior has been stabilized on SAT4J side. |