Re: infer GCC_REQD from USE_LANGUAGES

[Greg Troxel <gdt%lexort.com@localhost>]

Edgar Fuß <ef%math.uni-bonn.de@localhost> writes:

>> Mixing GCC versions doesn't work well.
> Yes. Therefore, that's exactly what my proposes solution/patch avoids.
>
> To repeat my proposal:
> 1. Choose /the one single GCC version/ to build everything with (besides 
>    that GCC itself, of course. Provide defaults for OS/version combinations, 
>    let user override (once for the whole build).
> 2. Infer (additional) GCC_REQD values from USE_LANGUAGES.
> 3. Error out in case GCC_REQD sums up to a higher GCC that what was chosen.
> 4. Build with the GCC version chosen.
>
> Open questions:
> 1. Building GCC may build libraries with a lower GCC than the rest.
>    Can we re-use them or do we need to build them into another PREFIX?
> 2. The base GCC version may be unable to build the target GCC, we may need to
>    build an intermediate version. This may build dependencies we need to 
>    discard/put into an interim PREFIX.
>
> Open details:
> 1. Does it work to use a lower GCC for C-only packages?
>    Benefit: don't need to build GCC if not building C++ packages.
> 2. Which GCC version do we regard as C++nn-capable?
> 3. Determine exact set of (recursive) GCC dependencies (varies with OS).
>
> The current situation is a mixture of some local works-for-me patches plus 
> an attitude to ignore the problem and wait until NetBSD versions using a 
> "too old" base gcc becomes unsupported an as such can be openly ignored. 
> Which doesn't work well because upstreams will demand newer GCCs/C++ levels 
> faster than NetBSD versions age out. Unless pkgsrc support is cut to 
> NetBSD-current only, something needs to be done.

It would be helpful to read the wiki page

  http://wiki.netbsd.org/pkgsrc/gcc/

and make suggestions relative to that, and make sure not to introduce
regressions from those designs.

Briefly, the notion is

  set a gcc version to be always used (at least for C++)

  bootstrap the chosen version without causing trouble

  fail if GCC_REQ exceeds the chosen version

  fail if the declared language is too new for the chosen version

This is almost what Jason is doing, except that I think on his builds
pkgsrc will go to a higher version if package GCC_REQ exceeds his system
GCC_REQ.

This leads to building a new compiler for pkgsrc, essentially almost
always.   That seems unfortunate, but that's just the velocity of the
C++ language specification these days.