Jack Hill <jackh...@jackhill.us> writes: > On Tue, 8 Aug 2017, Ricardo Wurmus wrote: > >> >> Paul Dufresne <dufres...@gmail.com> writes: >> >>> It takes 45 mins. on my relatively old dual core to guix-pull >> >> Yeah, this isn’t great. Since 0.13.0 compilation is slower and requires >> a whole lot more memory. That’s a known problem. >> >> There are some ideas to reduce the amount of compilation that has to >> happen locally, but it isn’t quite as simple as a first look may >> suggest. > > Could building be avoided entirely with substitutes? Are substitutes not > appropriate for some reason, or is it just that substitutes are not > produced on Hydra for every guix commit?
Before a substitute can be requested, Guix will have to compute a derivation locally. In the case of Guix itself this is rather expensive. Once we have the derivation we can ask substitute servers if they have a binary substitute for performing the work the derivation describes. So, local *computation* cannot be avoided, but local building should be avoidable for most packages — here the problem is that hydra isn’t fast enough yet. I’ve been preparing an alternative build farm at the institute where I work, which will hopefully soon be powerful enough to build packages more quickly than our current Hydra does. -- Ricardo GPG: BCA6 89B6 3655 3801 C3C6 2150 197A 5888 235F ACAC https://elephly.net
