[lmi] Upgrading gcc [Was: callback.hpp vs boost::function]

[Greg Chicares]

On 2005-10-30 23:02 UTC, Vadim Zeitlin wrote:
> On Sun, 30 Oct 2005 21:28:44 +0000 Greg Chicares <address@hidden> wrote:
> 
> GC> On 2005-10-30 20:52 UTC, Vadim Zeitlin wrote:
> GC> > 
> GC> >  while fixing this file to compile with g++ 4.0
> GC> 
> GC> BTW, I'm working on getting a clean compile with g++-3.4.4 now. The

I've committed those changes, but tested them only with gcc-3.4.2 . I need
to figure out the mpatrol problem before I upgrade to gcc-3.4.4 . MinGW
doesn't distribute any gcc-4.x yet; that's why I'm not using version 4.

I don't think my changes should affect your autotools work. Should we test
that now, or are you still refining it?

> GC> worst problem I've found is described here
> GC>   http://gcc.gnu.org/bugzilla/show_bug.cgi?id=22207
> GC> and can be worked around by specifying '-Wno-uninitialized'. Was that
> GC> a problem with 4.0 as well? If not, I'll use the workaround only for
> GC> 3.4.4 <= version < 4.0 .
> 
>  I think you should use it for 3.4.0 <= version <= 3.4.4 because I don't
> get the warning neither wiht 3.4.5 nor 4.0.2. Unfortunately I don't have
> 3.4.4 any more to test with it.

The bugzilla report (supra) specifies 'i686-pc-cygwin', and a followup
says it's reproducible with that target but not with 'i686-pc-linux-gnu'.
In 'workhorse.make' I now have:

  ifeq (3.4.4,$(gcc_version))
    gcc_version_specific_warnings := -Wno-uninitialized
  endif

Do you think it's worth making that conditional on the platform? It's a
shame to lose a potentially useful warning, but if you're building the
system with a version later than gcc-3.4.4 anyway, I guess you'll notice
any '-Wuninitialized' diagnostics (especially now that I've specified
'-pedantic-errors -Werror').

> GC> Have you found many other problems with 4.0 ? Please let me know so
> GC> that we can avoid duplicating each others' work. I'm fixing some expm1()
> GC> warnings right now. Wendy pointed out a 3.4.4 problem with mpatrol that
> GC> I haven't yet looked into.
> 
>  I didn't have any other problems.

Are you running 'make unit_tests' as well as building the wx target?
I see 37 out of 40 unit-test targets passing with gcc-3.4.2, and 38
out of 40 with gcc-3.4.4 . The difference is 'numeric_io_test', which
exposes a defect with MinGW gcc-3.4.2 . The other two tests that fail
are:

 - 'comma_punct_test', which exposes a libstdc++ defect
     http://gcc.gnu.org/bugzilla/show_bug.cgi?id=20914
   marked "Fixed for 4.0.1"

 - 'snprintf_test', which demonstrates defects in snprintf(); msvcrt
   (used by MinGW) is defective, and so is borland's standard library,
   but I'd hope that gnu libc gets it right

If you're using at least gcc-4.0.1, you might see no unit-test errors
at all. BTW, 'make -f fancy.make' compiles and runs all the unit tests
and filters out the details, showing just a summary.