| Age | Commit message (Collapse) | Author | Files | Lines | |
|---|---|---|---|---|---|
| 2009-06-03 | Fix a compile error in process_promotion (s.step_tier vs s.get_tier()). | Daniel Burrows | 1 | -1/+1 | |
| 2009-06-03 | Make process_promotion non-const. | Daniel Burrows | 1 | -1/+1 | |
| Necessary since it changes the state of the search graph. | |||||
| 2009-06-03 | Better logging of score changes when a new step is created. | Daniel Burrows | 1 | -3/+13 | |
| 2009-06-03 | Un-reverse step comparison by goodness. | Daniel Burrows | 1 | -5/+8 | |
| Previously I was using a priority queue, where the highest element was easiest to access. Now I'm using a std::set, so the lowest element is where we pull from ... but I forgot to adjust the comparison operation accordingly, so we would always examine the worst step, not the best one. | |||||
| 2009-06-03 | Avoid referencing uninitialized memory. | Daniel Burrows | 1 | -0/+11 | |
| 2009-06-03 | Compute the step's new score after updating it. | Daniel Burrows | 1 | -3/+5 | |
| This should allow the full solution bonus to be properly applied. | |||||
| 2009-06-03 | Also log when we generate a new step. | Daniel Burrows | 1 | -0/+4 | |
| 2009-06-03 | Log more information about the contents of steps as we process them and ↵ | Daniel Burrows | 1 | -2/+6 | |
| generate solutions. | |||||
| 2009-06-03 | Use choice_indexed_maps instead of choice_sets to represent the output ↵ | Daniel Burrows | 1 | -6/+6 | |
| domains of promotion sets, avoiding the problem with conflicting choices. choice_sets are only designed for situations where there will be at most one action on a given package; e.g., promotions and action sets. When we're looking for an incipient promotion, we might care about several different versions of a single package (e.g., different solvers of a dependency), so a choice_indexed_map is more appropriate. The map is given a value-type of "bool" mainly because the value type isn't needed; the code supports disabling some elements of the domain because it was easy to implement, but that shouldn't ever happen in practice. | |||||
| 2009-06-03 | Initialize the root's action score to 0. | Daniel Burrows | 1 | -0/+1 | |
| Since the action score is updated incrementally, a wrong score at the root means a wrong score elsewhere. | |||||
| 2009-06-03 | Remove a stray space in a log message. | Daniel Burrows | 1 | -1/+1 | |
| 2009-06-03 | Add more logging around scores. | Daniel Burrows | 1 | -1/+4 | |
| 2009-06-03 | Actually, just make the "process the step now" block the "else" clause of ↵ | Daniel Burrows | 1 | -7/+0 | |
| the cascading tests. | |||||
| 2009-06-03 | Actually enable processing steps that aren't irrelevant in some way. | Daniel Burrows | 1 | -0/+2 | |
| When I reworked the cascading series of tests that threw out irrelevant solutions, to take into account the fact that many cases are handled up-front, I forgot to keep the "else" condition where we actually do work. | |||||
| 2009-06-03 | Log the root step number and its tier. | Daniel Burrows | 1 | -0/+2 | |
| 2009-06-03 | Make set_step_tier do nothing if the step tier is being set to its current ↵ | Daniel Burrows | 1 | -0/+6 | |
| value, and log when the step tier changes. | |||||
| 2009-06-03 | Add more logging to the new problem resolver. | Daniel Burrows | 1 | -0/+10 | |
| 2009-06-03 | Actually set up the unresolved dependency map, solver lists, etc in the root ↵ | Daniel Burrows | 1 | -0/+6 | |
| step. Now instead of doing nothing, the resolver segfaults. Progress! :-D | |||||
| 2009-06-02 | Fix the rest of the compile errors in the problem resolver. | Daniel Burrows | 1 | -22/+22 | |
| There are still some link errors. | |||||
| 2009-06-02 | Fix enough compile errors that the problem resolver subdirectory compiles. | Daniel Burrows | 1 | -70/+114 | |
| The rest of the code still needs to be overhauled to take the changes into account, and it looks like there are bugs in code that's only instantiated from the client (e.g., reject_version()). | |||||
| 2009-06-01 | More compile fixes. | Daniel Burrows | 1 | -34/+45 | |
| 2009-06-01 | Use set_id() to make sure that choices added to solutions get IDs as expected. | Daniel Burrows | 1 | -1/+4 | |
| 2009-06-01 | More compile fixes. | Daniel Burrows | 1 | -12/+22 | |
| 2009-06-01 | Fix a bunch of compile errors. | Daniel Burrows | 1 | -112/+159 | |
| 2009-06-01 | Yank out the old abortive attempt at minimizing solutions. | Daniel Burrows | 1 | -642/+0 | |
| 2009-06-01 | Finish rewriting the main resolver loop to take the new structure into account. | Daniel Burrows | 1 | -62/+28 | |
| From here out it should mainly just be compile fixes. That, and hooking visited_packages up, but that's a relatively minor feature and can wait until I've got everything else working. | |||||
| 2009-06-01 | Fix another type error. | Daniel Burrows | 1 | -1/+1 | |
| 2009-06-01 | Fix a variable name typo. | Daniel Burrows | 1 | -1/+1 | |
| 2009-06-01 | Fix up various type names. | Daniel Burrows | 1 | -8/+8 | |
| 2009-06-01 | Initial draft of the new main loop code. | Daniel Burrows | 1 | -404/+176 | |
| Still incomplete, but fleshes out the major areas of the code that are needed. | |||||
| 2009-05-30 | Rewrite the rejection/frontend approval functions for the new way of storing ↵ | Daniel Burrows | 1 | -39/+52 | |
| and tracking rejections and approvals. | |||||
| 2009-05-30 | Now that deferral information is updated immediately, we don't need a ↵ | Daniel Burrows | 1 | -29/+3 | |
| variable to track whether it's dirty. | |||||
| 2009-05-30 | Use a choice-set, not a step, in the constructor's sanity check. | Daniel Burrows | 1 | -8/+6 | |
| 2009-05-30 | Rename step_installation to choice_set_installation and lift it to the top ↵ | Daniel Burrows | 1 | -25/+25 | |
| of the class definition. No need for this to be quite so specific. | |||||
| 2009-05-30 | Remove the old code to check whether a step was deferred. | Daniel Burrows | 1 | -137/+0 | |
| 2009-05-30 | Redesign how the closed set is stored. | Daniel Burrows | 1 | -61/+59 | |
| Instead of storing solutions, now we store a minimal set of information about a step: its choices and its scores. The scores aren't strictly necessary, but they're used to accelerate the process of checking the closed set for a step. I'm not currently planning to do any sort of forward searching for closed steps: I don't believe that they're all that common, the current code doesn't try to do this, eliminating them would add more computation to each step, and it would require more book-keeping to ensure that having successors in the closed set didn't result in a parent step being pushed to the conflict tier. | |||||
| 2009-05-29 | Various compilation fixes for the problem resolver. | Daniel Burrows | 1 | -18/+18 | |
| 2009-05-28 | Finish up everything we need to update step tiers correctly (I hope). | Daniel Burrows | 1 | -4/+52 | |
| 2009-05-28 | Consistently use set_step_tier instead of set_tier (oops). | Daniel Burrows | 1 | -2/+2 | |
| Turns out it was implemented, just under the wrong name. | |||||
| 2009-05-28 | First pass at rewiring the promotion/tier/index stuff to handle deferrals ↵ | Daniel Burrows | 1 | -10/+70 | |
| better. Deferral expressions now trigger immediate deferral of all related steps when they become "true", in addition to canceling deferrals when they become "false". Still to do: need to implement increase_step_tier and set_tier, so already-generated steps can be adjusted. | |||||
| 2009-05-27 | Explicitly make the is-deferred expression a listener with a sub-expression ↵ | Daniel Burrows | 1 | -33/+30 | |
| that's the actual condition. This gives us the separation I want between the active expression and the passive expression, although not in the most pretty way. There are still some bits of this that need to be filled in, mainly in how the tier validity expressions are computed and updated. Also, choices in the action set of a step need to impact its tier; I need to somehow incorporate that into the recomputation routines. | |||||
| 2009-05-26 | Log the validity condition when applying a promotion to a solver. | Daniel Burrows | 1 | -1/+2 | |
| 2009-05-26 | When applying a promotion to a solver, copy over the validity condition. | Daniel Burrows | 1 | -4/+13 | |
| 2009-05-26 | Build validity conditions when creating promotions. | Daniel Burrows | 1 | -4/+34 | |
| 2009-05-26 | When creating a new choice, add all the choices it contains to the global ↵ | Daniel Burrows | 1 | -0/+73 | |
| reverse index. | |||||
| 2009-05-25 | Implement a first draft of the rest of the code to handle deferral and ↵ | Daniel Burrows | 1 | -2/+135 | |
| retraction. This isn't quite right. The biggest thing is that deferral and tier validity are conflated too much; they are closely related, but I think that they need to be stored seprately (but this needs more thought). | |||||
| 2009-05-25 | Another iteration of how the reverse indices are stored and managed. | Daniel Burrows | 1 | -30/+153 | |
| This one incorporates the needs of the new system for retracting deferrals into the search graph structure, and updates the problem resolver to accomodate the new search graph data structures. Still not complete: the code to actually recompute a solver's tier needs to be written. In addition, various incorrect pieces of code that I came across while working on these changes have been fixed. | |||||
| 2009-05-25 | Use the new indexed set/map types to overhaul how promotions are processed ↵ | Daniel Burrows | 1 | -176/+314 | |
| in the resolver. The new code (as with the last few changesets, it's untested and not compilable) should be more correct than what we were doing before. It might also be a bit more efficient. | |||||
| 2009-05-23 | Fix some logging statements that didn't have a logger listed. | Daniel Burrows | 1 | -2/+2 | |
| 2009-05-23 | Use set_tier() when applying a promotion to existing steps. | Daniel Burrows | 1 | -1/+1 | |
 |
index : aptitude | |
| [no description] | Dyson |
| summaryrefslogtreecommitdiff |