Re: [tree-ssa-20020619-branch] current status report on i386-*-freebsd4*

[law at redhat dot com]

In message <200402180025.17189.stevenb@suse.de>, Steven Bosscher writes:
 >On Wednesday 18 February 2004 00:13, Gerald Pfeifer wrote:
 >> On Tue, 17 Feb 2004, Loren James Rittle wrote:
 >> > 2. For reasons I'm still investigating, many libstdc++-v3 execution
 >> > tests are failing with "terminate called after throwing an instance of
 >> > 'std::runtime_error' [junk]".  Many reports from the system malloc
 >> > implementation that "X is already free" are also seen (sometimes
 >> > without causing the test to fail).  Since that is under my purview, I
 >> > will take responsibility for fixing it, or posting whatever I find if
 >> > it escape.
 >>
 >> I've also been seeing this for my application!  If you build for i686
 >> instead of i386, it goes away.
 >
 >This looks like the compiler miscompiling itself on i386.
Probably the first thing I would do is compile a single stage with the
system compiler, then run the testsuite and see if anything pops up.

That would avoid a level of indirection (and potentially a ton of work) if
indeed the problem is that the compiler is mis-compiling itself.  

Of course, sometimes that procedure produces 100% good results and you
have to debug things the hard way....  But it's helped enough that it's
usually worth trying.

jeff