[Rpm-maint] How to get RPM requires quickly

[Panu Matilainen]

On 09/13/2014 04:13 PM, Richard W.M. Jones wrote:
>
> We have a program[1] that needs to take a list of installed RPM
> package names, and quickly generate all of the installed dependencies
> (recursively).
>
> *Note* this is NOT a question about yum/dnf/zypper/etc depsolving.
> It's about doing 'rpm -qR' quickly on installed packages only.
>
> At the moment we run the following command:
>
>    rpm -qR <current set of packages> |
>      awk '{print $1}' |
>      xargs rpm -q --qf '%%{name}\n' --whatprovides |
>      grep -v 'no package provides' |
>      sort -u
>
> The output is added to the current set, and we repeat the process
> until the set of packages reaches a fixpoint.
>
> To give you an idea, on a typical starting set this takes around 20 seconds,
> which is over half the total running time of the program.
>
> Does anyone have any suggestions about how to do this more quickly
> and/or cleverly?  Maybe using rpmlib (which unfortunately seems to
> lack any documentation)?

That kind of usage is pretty much worst case behavior for rpm as it 
defies almost all caching etc. For immediate speed gain with no other 
changes, add --nosignature to stop rpm from re-re-re-re-revalidating 
header signatures over and over again. In the average 'rpm -qa' time 
--nodigest doesn't help much but the cumulative effect might be worth it 
here.

To get rpm do a bit more work for you, use 'rpm -e --test' to make rpm 
resolve the deps. Basically

rpm -e --test <current set of packages> 2>&1 |
     awk '/needed by/{print $NF}' |
     sort -u

...and feed the output to the next cycle. --nosignature and --nodigest 
will help here too, but to lesser degree.

For more speed, you'd need to use the rpmlib API, pointers to 
documentation can be found at http://rpm.org/wiki/Docs

	- Panu -