Switch Travis to use Ubuntu 16.04 "xenial", and other CI tweaks

[fingolfin]

This should allow us to build semigroups and digraphs again (alas, it seems digraphs still runs into these weird atexit troubles...)

[codecov]

Codecov Report

Merging #2759 into master will increase coverage by 1.43%.
The diff coverage is n/a.

@@             Coverage Diff             @@
##           master    #2759       +/-   ##
===========================================
+ Coverage   83.81%   85.24%    +1.43%     
===========================================
  Files         683      110      -573     
  Lines      345977    56994   -288983     
===========================================
- Hits       289967    48586   -241381     
+ Misses      56010     8408    -47602

Impacted Files	Coverage Δ
src/hpc/aobjects.h	50% <0%> (-20%)	⬇️
src/streams.c	70.93% <0%> (-3.99%)	⬇️
src/hpc/misc.c	26.92% <0%> (-2.71%)	⬇️
src/saveload.c	66.57% <0%> (-2.39%)	⬇️
src/profile.c	41.27% <0%> (-1.99%)	⬇️
src/objects.c	80.48% <0%> (-1.66%)	⬇️
src/objects.h	89.55% <0%> (-1.5%)	⬇️
src/io.c	58.76% <0%> (-1.33%)	⬇️
src/dt.c	50.54% <0%> (-1.25%)	⬇️
src/gasman.c	87.15% <0%> (-1.21%)	⬇️
... and 633 more

[markuspf]

Not sure whether it makes sense; for meataxe64 I am now also using binutils-2.26: (https://github.com/gap-packages/meataxe64/blob/master/.travis.yml)

I spent quite a bit of time trying to debug a linker problem with meataxe64 which boiled down to "outdated binutils".

[markuspf]

sorry, I merged #2702 which caused a conflict with this PR...

[olexandr-konovalov]

gcc at least 5.0 would be very useful. The list of hacks to disable packages might now be extended with Digraphs (#2768). Hope with this PR all those hacks will be gone. But test of this PR currently fail to build packages:

checking whether GMP C++ library is available... configure: error: GMP C++ library not found. Make sure it was compiled with --enable-cxx

[olexandr-konovalov]

@fingolfin did you manage to get at least the gcc-5 of this PR working? I'd really love to have it because otherwise Semigroups can not be compiled on Travis, which severely reduces ability to identify potential problems.

[olexandr-konovalov]

Oh, again this

Loading digraphs ... 
#W dlopen() error: /home/travis/build/gap-system/gap/pkg/digraphs-0.12.2/bin/x\
86_64-pc-linux-gnu-default64/digraphs.so: undefined symbol: atexit
Error, module '/home/travis/build/gap-system/gap/pkg/digraphs-0.12.2/bin/x86_64-pc-li\
nux-gnu-default64/digraphs.so' not found in
  if not LOAD_DYN( arg[1], false ) then
    Error( "no support for dynamic loading" );
fi; at /home/travis/build/gap-system/gap/lib/files.gd:583 called from 

[fingolfin]

This seems to be working now, at least with reduced tests. I've now cleaned up the PR, and in particular re-enabled all tests, so let's see if it still passes.

[fingolfin]

To document the digraphs issue in more place: this happened because we added --coverage to CFLAGS, and then kernel extensions with C++ code like digraphs, semigroups and float ended up getting those flags; then GCC/G++/the GNU C++ library or something ends up including the header profiler_trace.h, which in turn inserts an atexit call. But with glibc, atexit is not in libc.so. but rather in a separate library libc_nonshared.a. So the .so has an unresolved symbol, as that symbol is not available in any shared library. Hence the error we were seeing.

[markuspf]

Suggested change

	- libmpfi0-dev:i386 # for float
	- libmpfi0-dev:i386 # for float