+1 on merging it and also agreed with Uwe that we will need to deal with LLVM version conflicts. In addition it would be good to come up with a plan on how it can be useful for other DataFrame open source projects. Having end-to-end applications that let people profit from this code will help adoption/development a lot I think.
On Wed, Jul 25, 2018 at 6:56 AM, Uwe L. Korn <uw...@xhochy.com> wrote: > Having it in Arrow also will enable us to better promote it on the Python > side with the pyarrow package. It will be a great addition but we will need > to figure out then the problem of dealing with in-memory LLVM versions so > that we don't conflict with packages like Numba. This is a problem on the > Python packaging side we should address early but won't be a blocker in > getting Gandiva into the Arrow repositories and infrastructure. > > Uwe > > On Tue, Jul 24, 2018, at 9:39 PM, Phillip Cloud wrote: > > While I'm not a level 83 LLVM wizard like Antoine :) I have a small > amount > > of experience with it and would also be happy to review/merge patches. > > > > Having Gandiva in arrow will simplify packaging and building the library, > > which IME has always been annoying with large cross-platform and > > cross-language projects. > > > > I think this would be a great addition to Arrow. > > > > On Tue, Jul 24, 2018 at 8:46 AM Uwe L. Korn <uw...@xhochy.com> wrote: > > > > > Small update: > > > > * CentOS 6 and later (we use the RedHat devtoolset-2 compiler > toolchain) > > > > > > We still build the Python wheels on CentOS 5. This should be no > problem as > > > llvmlite can also be build on that OS. > > > > > > It looks like there will be a new Linux Python wheel standard soon > where > > > we will upgrade to CentOS 6. > > > > > > Cheers, > > > > > > Uwe > > > >
