This is the mail archive of the mailing list for the binutils project.

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]
Other format: [Raw text]

Re: Binutils 2.13 branch created

On Sun, Jul 14, 2002 at 10:46:09AM -0400, Ben Elliston wrote:
> > There is one test failure on my i386-linux host (gasp flonums) which I
> > believe has been there for a while; I'd like to have either the problem
> > or the test fixed in the next week or so, if anyone wants to step up.  
> > As usual, feel free to fix problems on the branch until we freeze it;
> > I'd appreciate being CC'd on any branch patches.
> I wrote this test earlier in the year and it should definitely pass.  I 
> can take a look at it, if you like?

My log is full of things like:

<       .float -09.0
>       .float -9.0
<       .float 0.09
>       .float 0.9


<       .float -2.0900
>       .float -2.900

The host is Debian unstable (glibc 2.2.5 + some patches) and the system
compiler is gcc 2.95 + some patches.  It shows up when configured for
--enable-targets=all --enable-64-bit-bfd and using the system compiler.

It does _not_ show up when using GCC 3.1 and a simple default
(../src/configure) toolchain.  The two tests in cdtest fail because of
certain inconsistencies in the way the testsuite selects a compiler,
but that's not a real failure.  It also does not show up for a GCC 3.1
--enable-targets=all --enable-64-bit-bfd build, or for a GCC 2.95 single-target build.

I suspect we are triggering a bug in 2.95.x's long long handling :(  If
so a MIPS target build ought to also show the failure, but that doesn't
appear to happen.  I'll try to narrow it down further.

(An interesting note: ld-bootstrap.exp takes a ridiculously long time
on GCC 3.1, about six times as long - meaningless since I had
optimization turned off.  Just another symptom of the growth of useless
code generated by 3.1 and cleaned up later (slower linker and bigger
objects), I expect, and the larger debug info associated with DWARF-2
(much bigger objects).  Primarily slower LD since they are comparable
at -O2.)

Daniel Jacobowitz                           Carnegie Mellon University
MontaVista Software                         Debian GNU/Linux Developer

Index Nav: [Date Index] [Subject Index] [Author Index] [Thread Index]
Message Nav: [Date Prev] [Date Next] [Thread Prev] [Thread Next]