GCC versions to keep

[vcunat]

I've been fixing some older gcc versions on stdenv-updates, but do we really need so many of them? Most don't have any explicit reference anywhere (also holds for gcj, gnat, gfortran).

The default is 4.8 there and I'm sure some others do have use cases, but it seems to me that it's unlikely to be so for all of them.

[edolstra]

I removed 2.95, 4.0 and 4.1 a while back. Probably some others can be removed as well. But please don't remove 4.4 since we need it at LB :-)

All those unused gcj/gnat/gfortran instances can go IMHO.

[edolstra]

I'd say that 4.7 can certainly be removed because it was never used in a "stable" Nixpkgs.

[viric]

On Tue, Jul 02, 2013 at 08:43:24AM -0700, Eelco Dolstra wrote:

I removed 2.95, 4.0 and 4.1 a while back. Probably some others can be removed as well. But please don't remove 4.4 since we need it at LB :-)

All those unused gcj/gnat/gfortran instances can go IMHO.

Attention with gcc 4.3 and 4.4; ghdl depends on them.

[falsifian]

On a related note, gcc 4.6 doesn't build on stdenv-updates. I don't know if anything uses it. (I only noticed because I was trying to see if "stdenv = overrideGCC stdenv gcc46" fixes #707.)

[jcumming]

I have some local changes to make gcc46_multi work with stdenv updates (keep cloog-0.16, ppl-0.11.2, isl-0.07), as I haven't got gcc48_multi to work.

[vcunat]

I knew about that... recent gcc versions seem to depend on exact versions of these libraries. Thanks for fixing, @jcumming.

As for 4.7, currently I didn't make fortran-4.8 compile (some strange c++ linking errors), but I hope we will solve that in time. EDIT: same for jdk.

[mornfall]

Hmm, please keep 4.7 for now, as 4.8 has some unfortunate performance regressions, but pre-4.7 have substantial C++11 problems.

[shlevy]

Is there anything left to do here?

[vcunat]

Probably not. Maybe resurrect gcc-4.7 (if @mornfall or someone else still want it; but probably not build it on Hydra).

[shlevy]

OK, closing for now then.