Greg Troxel <gdt%lexort.com@localhost> writes: >> Possibly related thought - rather than hand curating sets of PKGPATH >> checks, would something of the form >> - 'set _BUILDING_COMPILER' when building the compiler >> - Only set GCC_REQD if _BUILDING_COMPILER is unset >> work? > > I don't think so, because then we get different binaries for a package > depending on whether it was asked for normally or bootstrapped. That's > the central hard point, and it could lead to an entirely different > pkgsrc instance to build the bootstrap compiler. That seems like a lot > of CPU time, but I think it's not that much, and it's fairly clearly > sound. > > The straw proposal says to encode essentially as Jason has done, but in > the package Makefiles, not another place, and to reduce dependencies. > However, that is a NetBSD-centric approach as the set above is huge, > compared to what one needs on NetBSD 7, where far more tools from base > are acceptable. I thought about this more, and it seems that on Linux, we need to build a lot. So the choice is between the recursive variable you suggest, encoding that same behavior explicitly, or a separate prefix. I lean to your approach until we're sure it hurts more than the alternatives.
Attachment:
signature.asc
Description: PGP signature