On Mon, Jan 06, 2003 at 10:39:30PM -0500, Colin Walters wrote: > > I think it's important that the 'C++ compiler' build-essential > > requirement *not* be satisfied by a compiler that builds to an older, > > broken ABI. If someone needs that ABI, they have special build > > dependencies beyond the scope of build-essential, IMHO. > I initially agreed with this, but Colin Watson convinced me otherwise. > If you look at the definition of build-essential, it is defined so that > a 'Hello World!' type program will build. And it is quite reasonable > for a simple 'Hello World!' type program which does not use any > libraries other than libstdc++ to be compiled with 2.95, and still > work. It being compiled with an older compiler would not affect > anything else. So this makes gcc 3.2 not build-essential. This would be fine if the build-essential package declared 'c-compiler' or 'c++-compiler'. It doesn't: it declares a dependency on 'gcc' and 'g++', which always point to a specific version of gcc. There's no reason why now, of all times, we would allow two different, quite incompatible versions of g++ to satisfy the build-essential requirements. -- Steve Langasek postmodern programmer
Attachment:
pgpxpekDRIajc.pgp
Description: PGP signature