[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: [p2-dev] New explanation algorithm
- From: Daniel Le Berre <leberre@xxxxxxxxxxxxxxxxxxx>
- Date: Fri, 26 Mar 2010 18:01:53 +0100
- Delivered-to: email@example.com
- User-agent: Thunderbird 126.96.36.199 (X11/20100301)
Pascal Rapicault a écrit :
> Sorry for the late reply, most of us were at eclipsecon promoting, p2, p2CUDF and SAT4J :)
> What is your level of confidence of stability of this new code. What sort of test suite do you have on the SAT4J side?
I am relatively confident, because I broke many test cases in the SAT4J
library when designing that new algorithm (the ones I created when I
introduced the explanation support last year).
Then I broke many Eclipse test cases.
Now all the tests pass :)
> Is the old code still in SAT4J? I'm asking because having the two code side by side could help us verify the explanations being produced (maybe by setting a sys property) and thus gain confidence.
More or less. The new algorithm reduces the set of candidates to
consider for the explanation in the quickxplain algorithm.
Previously, we were considering all the dependencies.
I removed from SAT4J solver some code that was there just to order the
dependencies to focus the search for an improved explanation time.
It became useless with the new approach.
I can easily add a flag that enables/disables the reduction of
candidates. However, I would prefer not to include back the trick to
order the dependencies.
> On 2010-03-24, at 3:00 PM, Daniel Le Berre wrote:
>> Daniel Le Berre a écrit :
>>> Dear all,
>>> I have been working on a new explanation algorithm in SAT4J this week end.
>>> I am looking for feedback on that new algorithm.
>>> You can grab an updated version of SAT4J with that new algorithm here:
>>> Please report any issue there.
>> Well, I just noticed that my message was maybe too cryptic for the
>> majority of the list :)
>> The explanation algorithm is used when the resolver cannot find a plan
>> to explain why the installation cannot proceed.
>> I decided to write a new algorithm after noticing last week that the
>> explanation of a failure to upgrade from a 3.6M5 with JEE packages to
>> 3.6M6 could take more than one minute.
>> With the updated algorithm, it should take only a few seconds.
>> If you meet such kind of slow explanations, I would like to have your
>> feedback with the new algorithm.
>> Daniel Le Berre mailto:leberre@xxxxxxxxxxxxxxxxxxx
>> MCF, CRIL-CNRS UMR 8188, Universite d'Artois
>> p2-dev mailing list
> p2-dev mailing list
Daniel Le Berre mailto:leberre@xxxxxxxxxxxxxxxxxxx
MCF, CRIL-CNRS UMR 8188, Universite d'Artois