This is the mail archive of the gcc@gcc.gnu.org mailing list for the GCC 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: i960-elf




I acctually didn't do a reply-all and missed replying to the list.
So Jim ignore the private one.

Jim Wilson wrote:
> 
> I figured out why the i960-elf build doesn't work.  It is because there is no
> i960-elf support in gcc.  There is i960-elf support in binutils and gdb, but
> none in gcc.

<sigh>  That explains a lot.

> i960-elf appears to work in gcc because there is a i960-*-* target, but that
> is really a misnamed i960-bout target.  You get ELF object files because
> binutils knows about i960-elf, but the code that gcc is emitting is not right
> for an ELF target.  It is using the wrong debug info format.  It is using
> the wrong method for static constructors/destructors.  It is using the wrong
> EH mechanism.  Etc.  And of course the problem that stops the build, it
> doesn't support named sections which is needed for section attributes.
> Fixing this means adding a real i960-elf target, and I think that is beyond
> what can be reasonably expected of me.

It is certainly way beyond what I thought I was reporting.  I thought I
was 
reporting problems in a target that existed and was supposed to work. 
I had no intention of someone implementing a target.

> There are a few things I'd be willing to do.  I could add an i960-elf target
> that prints an error message at configure or build time.  I could rename
> the misnamed i960-*-* target so that we get a configure error for an
> unsupported target.  I could add a i960-elfbroken target that works like
> the current one but has named section support so that the build will complete.
> This should still work with binutils and gdb since they check for i960-elf*.

I think getting a configure error for unsupported target is the
appropriate thing to do.  
 
I have no explicit need for i960-elf.  I started building it because 
the RTEMS community wanted to generally move to ELF across all targets
possible.  If the target appeared to be mature (m68k-elf i386-elf) we
dropped using the coff target.  Otherwise we continued to use it.
The sh and i960 were the last targets I saw hope in moving to elf.
The sh looks like it will eventually happen but now I know the i960 
just won't any time soon.

> Or we could just stop trying to build i960-elf.

This is reasonable and implementing it should be documented on the 
gcc projects list as well. 

I go with Jim's proposal to ...

  + give a configure error for i960-elf
  + add implementing i960-elf to the gcc projects list
  + close all gcc PRs that are only for i960-elf

How does that sound Jim?

> Jim

-- 
Joel Sherrill, Ph.D.             Director of Research & Development
joel@OARcorp.com                 On-Line Applications Research
Ask me about RTEMS: a free RTOS  Huntsville AL 35805
   Support Available             (256) 722-9985


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