On Wed, Feb 14, 2018 at 4:14 PM, Matthew Miller
<mat...@fedoraproject.org> wrote:
> I'd be super-interested in benchmarks comparing before and after
> install times. I guess since the plan is to do this _after_ the mass
> rebuild, we'll need to wait until after the *next* rebuild to see how
> much impact this has.

I don't think this single change will make a huge difference within
the existing ecosystem, but I think it's an important step in a larger
shift toward make package installation & image composition a)
introspectable and b) deterministic, so that we _can_ make
installs/composes faster and more reliable.

I talked about all the reasons eliminating/reducing scriptlets is a
good idea (and how we can actually do it) at Flock and at DevConf[1],
but since you're asking about speed specifically, I can say that with
the weldr[2] proof-of-concept tooling (which skips scriptlets) we can
build a system image in about 6 seconds[3], where the current tools
took 5+ minutes to build an image with the same contents[4].

So: before-and-after install times within Fedora probably won't change
much immediately, but these kinds of changes are important to enable
future work that will eventually give us dramatically faster and more
reliable tooling.


[1] If you want to know more about the proposed Scriptlet Reforms you
could watch the recording from DevConf:
(Or: just ask me to explain more and I'll be all too happy to talk about it!)
[2] http://weldr.io/
[3] I played a short recorded demo showing this during the talk,
10m52s in: https://www.youtube.com/watch?v=kE-8ZRISFqA#t=10m52
[4] To be fair, those numbers are from comparisons I ran by hand 6
months ago. But we'll start doing regular benchmarks after we've
finished a few missing bits and landed the code in Fedora.
devel mailing list -- devel@lists.fedoraproject.org
To unsubscribe send an email to devel-le...@lists.fedoraproject.org

Reply via email to