[Rpm-maint] rpmtsOrder failed,

[Panu Matilainen]

On Mon, 23 Jun 2008, Johnson, Richard wrote:

> Folks--
>
> I've been having a bear of a time installing a suite of rpms where all 
> dependencies are satisfied, only to fail in tsOrder.  I've tracked the 
> error down to this snippet from lib/depends.c (nrescans is initially 10)
>
> 1388         /* If a relation was eliminated, then continue sorting. */
> 1389         /* XXX TODO: add control bit. */
> 1390         if (nzaps && nrescans-- > 0) {
> 1391             rpmlog(RPMLOG_DEBUG, "========== continuing tsort ...\n");
> 1392             goto rescan;
> 1393         }
>
> Increasing the allowable rescans enables install to proceed.  This makes 
> me wonder why rescans are limited in the first place.  My reading is 
> that as long as nzaps!=0 progress was made and a rescan is appropriate.
>
> Or could someone enlighten me otherwise?

Ah, that... Indeed the limit of 10 is artificial, no real reason to stop 
when progress can be made. Hitting that limit says you have a lot of 
dependency loops in your package set (and probably a large set of packages 
too). Now, dependency loops aren't exactly a good thing but failing 
unnecessarily is just silly.

This was recently encountered in Fedora too and the artificial limit 
removed there, only I forgot to apply it upstream (done now). Thanks for 
reminding me :)


 	- Panu -