On 2020-03-15 02:28, Craig Ringer wrote: > On Fri, 13 Mar 2020 at 15:04, Andres Freund <andres@anarazel.de > <mailto:andres@anarazel.de>> wrote: > > On 2020-03-13 14:08:12 +0800, Craig Ringer wrote: > > The alternative would be to detect a missing clang and emit a > much more > > informative error than the current one that explicitly suggests > retrying > > with > > > > make with_llvm=no > > > > or setting with_llvm=no in the environment. > > That, that, that's what I suggested upthread? > > > Yes, and I still don't like it. "with_llvm" is the actual > already-working option. I'd rather make this not randomly explode for > users, but failing that we can just hack the Makefile in the rpms for > EL-7 (where it's a particular mess) and rely on an error message for > other cases.
I don't really get the problem. with_llvm=no works, so it can be used.
Options that automatically disable things based on what is installed in the build environment are bad ideas. For instance, we on purpose don't have configure decide anything based on whether readline is installed. Either you select it or you don't, there is no "auto" mode.
Fine with me. I wrote it before identifying that with_llvm=no was a viable workaround.
The whole thing is a bit ugly, but if the fix isn't clearly better than the problem the fix shouldn't go in. This way it'll be searchable-for at least.
I think we'll be adopting some kind of ugly workaround like this for the CentOS 7 packages in PGDG yum because they're a bit of a special case, since the llvm support requires an additional 3rd party that isn't declared as a build-depend on the devel package. But that can be done at packaging level + a small patch applied during package builds for CentOS 7 only.