Re: *-rtems and binutils 2.14.92

[Daniel Jacobowitz <drow at false dot org>]

On Tue, May 11, 2004 at 12:48:42PM -0500, Joel Sherrill wrote:
> 
> Hi,
> 
> I have managed to build the following targets with 2.14.92
> on a Linux host.
> 
>   arm-rtems4.7		  h8300-rtems4.7
>   i386-rtems4.7		  i960-rtems4.7
>   m68k-rtems4.7		  mips64-rtems4.7
>   mips-rtems4.7		  powerpc-rtems4.7
>   sh-rtems4.7		  sh-rtemself4.7
>   sparc-rtems4.7	  tic4x-rtems4.7
> 
> I could not successfully build or32-rtems.  I do not know if
> any or32 target is expected to build.

or32-elf had only 3 LD testsuite failures, according to Nick's posted
results, so this may be a configury issue.

> I am in the process of building all RTEMS configurations with
> those binutils and gcc 3.3.3/newlib 1.11.0.  Those were built
> using binutils 2.14 if that matters to anyone.  I am midway
> in the m68k in building RTEMS but wanted to pass this much
> along:
> 
> h8300 - bad expression error on file compiled by gcc with
>   "-mh -mint32" is that helps.
> 
> I trimmed the assembly language down to this:
> 
>        .h8300h
>         .section .text
>         cmp.l   #--2147483648,er2
> 
> gcc invoked gas with only --traditional-format as an
> argument -- there were no CPU flag options if that matters.
> 
> Hmmm... that "--" shouldn't be there.  Is this a gcc 3.3.3 bug
> that only a newer binutils has caught?  Does anyone recognize it?

I believe that the -- used to be parsed as - ( - 2147483648 ), and
Nathan added a patch to prevent this because most people expected --
and ++ to mean some sort of increment/decrement.  Certainly the
expression above doesn't make much sense; that's -MIN_INT ....

-- 
Daniel Jacobowitz