On Tue, Apr 10, 2007, Daniel Burrows wrote:
>   Hrm, the resolver has a cutoff to stop thinking after "a while", but
> it looks like you've overridden that (-y).  Maybe I should add some
> additional logic that aborts the search regardless of -y if the
> resolver's data structures get too big (this means we have a problem
> that's "too hard" to solve).

 I don't know; I used "-y" for aptitude to work alone in non-interactive
 mode, but perhaps there's something lighter which has the same effect?
 If not, I suppose it would be nice to stop the resolver after a while
 despite -y indeed.

>   If there isn't a solution at all (which it sounds like is probably
> happening in your case), I'd hope the resolver would be able to
> accumulate enough conflicts to toss the whole problem out.

 I think there's a solution; the build succeeded with the hackish shell
 script which I added to pbuilder to resolve build-deps in experimental.
 I'm attaching a successful build log which has each build-dep choice
 listed. (epiphany-extensions_2.18.1-1_i386.build.gz)

>                                                             I bet that
> you have a cache with several copies of a large dependency tree (e.g.
> in testing, unstable & experimental) and where some dependencies in
> the tree can be satisfied from any repository but others conflict
> across repositories.  That's the hardest case for the resolver, and I
> don't think it handles it very gracefully.

 Yes, I have the full sid and experimental repositories, and a local
 almost empty one, and the situation probably looks awful.

>   Could you add "-o Aptitude::CmdLine::Resolver-Debug=true" to the
> aptitude command-line and send me the output?

 This rapidly produces huge amounts of debug, so I've only kept the
 first 100000 lines of build log and bziped them.  It's still 60 MB
 uncompressed but tiny when compressed.
 (epiphany-extensions_2.18.1-1_i386.build.bz2)
 
<http://people.dooz.org/~lool/debian/epiphany-extensions_2.18.1-1_i386.build.bz2>
 (I can't attach it, master.d.o and bugs.d.o reject large bz2.)

 It seems aspell appears a lot in the log.

-- 
Loïc Minier
"For subalterns, saying something intelligent is as risky as saying something
 stupid."

Attachment: epiphany-extensions_2.18.1-1_i386.build.gz
Description: Binary data

Reply via email to