GCC 6 is still intended to be the default GCC for the upcoming stretch release.
It doesn't yet build on mips and mipsel (gfortran and gnat), however GCC 5 has
non-addressed RC issues on these architectures, so maybe better regress on
release architectures than toolchain progress.
Debian QA didn't yet find the time for a comparative test rebuild on the amazon
hardware, so until now I only have data for a test rebuild of the current Ubuntu
development series, otoh with a wider coverage on architectures. Lucas Nussbaum
assured me that an amd64 test rebuild could be done in early Feb.
Now, I'd like to file the issues I found for the Ubuntu test rebuild on the
Debian packages. Of course there will be some false positives. Just not having
any bugs filed at all is bad, maybe having bugs filed for diverging
Debian/Ubuntu versions could be bad too (although it looks like that the version
is the least important thing for new warnings, more strictness or symbols file
changes). So at least I'd like to file issues for matching Debian/Ubuntu
versions, but would strive to file issues for all packages.
For regressions build packages please see
http://people.canonical.com/~doko/tmp/gcc6-regr/
Issues for all known GCC ICEs were already filed and most of them already fixed
upstream (see https://gcc.gnu.org/ml/gcc/2016-01/msg00101.html).
Full list of build failures (including those which are not regressions):
http://people.ubuntuwire.org/~wgrant/rebuild-ftbfs-test/test-rebuild-20151218.1-gcc6-xenial.html
gcc-6 packages are available in experimental.
Matthias