Re: solaris patch

[Phil Edwards <pedwards at disaster dot jaj dot com>]

On Fri, Oct 27, 2000 at 05:53:41AM -0700, Benjamin Kosnik wrote:
> Right. I fixed this in a different way, and checked it in this morning. 

Testing against "broken" appeals to my sense of humor.  I like it.


> I'd appreciate knowing current solaris and aix status as of today's CVS, 
> please. 

Builds on sol with 1) --disable-long-long and 2) a tweak to config.cache.

In reverse order:  the tweak has to do with 'modff'.  That function
exists, and is declared in the system math.h, and is correctly discovered
by configure.  Yet when building one of the valarray sources, the compiler
reports that ::modff is undeclared.

If I tweak config.cache (and/or bits/c++config.h) to say that modff isn't
available, then the valarray code uses a workaround, and all is well.

The issue with 1) seems similar; 'strtoull' is available and detected as
such, but locale_facets.tcc suddenly can't find it.  I don't know what may
be wrong after that point; there's no way to work around this problem as
with modff, so The Build Stops Here.


I have not investigated this yet.  We're in C++ mode when testing for modff,
but in C mode when testing for strtoull.  (Last month I experimented with
putting configure into C++ mode by default, and switching to C only when
needed -- currently we do the opposite -- and found that at least one of
the AC_CHECK_FUNC tests just fails miserably, some of the math tests give
different answers, and the ctype tests aren't actually valid.  A patch
was worked up but got overtaken by events.)


The odd bus errors at runtime were going to be my next task.  At one
point I thought the gc-sections error was causing it, and reported as
such to binutils, but the problem has reappeared even without gc-sections.
Time to fire up the debuggers.  (And I'm still fuzzy on how we can correctly
GC before actually creating an executable, but I'll take the experts at
their word.)


Phil

-- 
pedwards at disaster dot jaj dot com  |  pme at sources dot redhat dot com
devphil at several other less interesting addresses in various dot domains
The gods do not protect fools.  Fools are protected by more capable fools.