darwin-powerpc compiler/linker problem

darwin-powerpc compiler/linker problem

[Laurent GUERBY]

Hi, I'm trying to compile a large Ada application with macada.org GCC
3.3-apple based GNAT and with 3.5. In both cases I managed to compile
the software, but at link time I get piles of linker message like
the following:

ld: xxx.o relocation overflow for relocation entry 587 in section
(__TEXT,__text) (displacement too large)

I tried -fno-common, -mminimal-toc, -fPIC (all found via google) and
various other combinations but without success. Small programs
or ACATS have no problem compiling, linking and running (-fno-common
needed with 3.5 though).

Any idea of what to try?

Thank you in advance,

Laurent

Re: darwin-powerpc compiler/linker problem

[Andrew Pinski]

On Mar 23, 2004, at 11:48, Laurent GUERBY wrote:

> Hi, I'm trying to compile a large Ada application with macada.org GCC
> 3.3-apple based GNAT and with 3.5. In both cases I managed to compile
> the software, but at link time I get piles of linker message like
> the following:
>
> ld: xxx.o relocation overflow for relocation entry 587 in section
> (__TEXT,__text) (displacement too large)
>
> I tried -fno-common, -mminimal-toc, -fPIC (all found via google) and
> various other combinations but without success. Small programs
> or ACATS have no problem compiling, linking and running (-fno-common
> needed with 3.5 though).
>
> Any idea of what to try?

Yes this is a known linker bug for Darwin's linker, there is nothing
you can do about it except to complain to Apple or fix the linker
yourself.

Thanks,
Andrew Pinski

Re: darwin-powerpc compiler/linker problem

[Mike Stump]

On Tuesday, March 23, 2004, at 11:48 AM, Laurent GUERBY wrote:
> Hi, I'm trying to compile a large Ada application with macada.org GCC
> 3.3-apple based GNAT and with 3.5. In both cases I managed to compile
> the software, but at link time I get piles of linker message like
> the following:
>
> ld: xxx.o relocation overflow for relocation entry 587 in section
> (__TEXT,__text) (displacement too large)
>
> I tried -fno-common, -mminimal-toc, -fPIC (all found via google) and
> various other combinations but without success. Small programs
> or ACATS have no problem compiling, linking and running (-fno-common
> needed with 3.5 though).
>
> Any idea of what to try?

Source code mods to avoid large arrays is the usual work around as I 
recall.

[ I hope I'm not confusing with another problem that looks like this 
one... ]

Re: darwin-powerpc compiler/linker problem

[Laurent GUERBY]

On Tue, 2004-03-23 at 21:49, Mike Stump wrote:
> > Any idea of what to try?
> 
> Source code mods to avoid large arrays is the usual work around as I 
> recall.

You mean static/global arrays? Otherwise our software is full
of large heap allocated arrays so it's hopeless :). We have
some dynamic sized on the stack too (but not that much
since it doesn't work when stack size is limited).

I don't believe we have large static/global arrays
(our software is routinely used as a thread safe library
by multi threaded programs on various platforms and we do
no locking).

I'll double check, but since we got linker complaints
about neatly all of our object files I doubt I'll
find something in the area (at most one or two source files
could have some constant static arrays).

May be GNAT is generating some things in our back though.

Any idea on how to get more precise information about
what the linker calls an "entry"?

Laurent

Re: darwin-powerpc compiler/linker problem

[Mike Stump]

On Tuesday, March 23, 2004, at 01:00 PM, Laurent GUERBY wrote:
> On Tue, 2004-03-23 at 21:49, Mike Stump wrote:
>>> Any idea of what to try?
>>
>> Source code mods to avoid large arrays is the usual work around as I
>> recall.
>
> You mean static/global arrays?

Yes.

> Any idea on how to get more precise information about
> what the linker calls an "entry"?

otool -r xxx.o and then count down 587 lines (first line is 1 I think), 
and then take that address and find the code in the otool -d xxx.o 
stuff?  You can confirm it, by grabbing the .s for it, editing it down 
to not overflow and then relink I think.