Mailinglist Archive: zypp-devel (75 mails)

[Michael Matz <matz@xxxxxxx>]

Hi,

On Fri, 16 May 2008, Michael Matz wrote:

> Nah.  The problem seems to be that right now PoolQuery wants to match 
> strings on it's own instead of letting it be done by the DataIterator, 
> so it has to iterate over all solvables itself, and somewhere there must 
> be a bug that leads to actually viting all solvables again leading to 
> quadraticness.  There's clearly something going very wrong in PoolQuerys 
> use of the DataIterator, I see it in gdb.

I know coolos problem I think.  He had a lock file written with the old 
attribute names still, like:

kind: package
solvable_name: build
string_type: glob

When parsing this in, "kind" and "string_type" are parsed as normal 
attributes to match, because they meanwhile are called "type" and 
"match_type".  This makes it a three-attribute query (where two of them 
never will match anything because they are special attributes that aren't 
recognized anymore).  And queries with more than one attribute are not 
matched by the dataiterator but by libzypp itself, which is _very_ 
inefficient (it basically retrieves _all_ attributes of all solvables and 
tries to match them against the query).

To show you the difference, with my locks file using the old attribute 
names:

% time zypper in fate
real    0m42.440s
user    0m41.127s
sys     0m1.204s

When I edit the locks file by hand to use the meanwhile correct attribute 
names (but otherwise the same):

% time zypper in fate
real    0m2.161s
user    0m1.312s
sys     0m0.252s

So, coolo: do this on your locks file:
sed -e 's/^kind:/type:/' -e 's/^string_type:/match_type:/' 

Josef: it would be worthwhile I think to still recognize and parse the old 
names (without exposing them in the API) to avoid this problem with lock 
files written in the intermediate form.


Ciao,
Michael.
-- 
To unsubscribe, e-mail: zypp-devel+unsubscribe@xxxxxxxxxxxx
For additional commands, e-mail: zypp-devel+help@xxxxxxxxxxxx