| 1 |
Hi all, |
| 2 |
|
| 3 |
this is the next step into getting rid of backtracking. This patch adds |
| 4 |
a new function that allows _solve_non_slot_operator_slot_conflicts from |
| 5 |
the last patch series to solve conflicts caused by built slot operator |
| 6 |
dependencies. It does this by pulling in alternative packages for the |
| 7 |
parents of the conflict packages. |
| 8 |
|
| 9 |
As always pyflakes, pylint and the tests don't show any error. |
| 10 |
|
| 11 |
You may find this patch in my github repo [1]. |
| 12 |
|
| 13 |
Here are some numbers for -uDN world. |
| 14 |
|
| 15 |
Before the first patch series: |
| 16 |
* 11 slot conflicts |
| 17 |
* 5 backtracking steps and fails |
| 18 |
|
| 19 |
Current master: |
| 20 |
* no conflicts |
| 21 |
* 7 backtracking step and succeeds |
| 22 |
|
| 23 |
This patch: |
| 24 |
* no conflicts |
| 25 |
* no backtracking required |
| 26 |
|
| 27 |
Some times on my system (core2 2 GHz, 4 GB RAM) with hot file system caches. |
| 28 |
|
| 29 |
Current master: |
| 30 |
9:28 |
| 31 |
|
| 32 |
Current master with -backtrack=0 (fails to find a solution): |
| 33 |
1:38 |
| 34 |
|
| 35 |
This patch: |
| 36 |
1:58 |
| 37 |
|
| 38 |
|
| 39 |
As you can see by comparing with the --backtrack=0 case, the new conflict |
| 40 |
solving abilities cost some time (+18%). This is to be expected as it now |
| 41 |
has to resolve the dependencies of more packages. |
| 42 |
|
| 43 |
Compared to the normal case this is a decrease by 79%. |
| 44 |
|
| 45 |
Note that if your system doesn't have any conflict, there shouldn't be any |
| 46 |
increase in run time. |
| 47 |
|
| 48 |
|
| 49 |
[1] https://github.com/few/fews-portage-branch/tree/no-backtracking-for-rebuilds |
Archives