Regression caused by checkin.

classic Classic list List threaded Threaded
4 messages Options
Reply | Threaded
Open this post in threaded view
|

Regression caused by checkin.

Khem Raj
Hi Nick

While building glibc with latest binutils from CVS, gas reported an
error which was not there before. It is caused by this checkin

http://sourceware.org/ml/binutils-cvs/2005-10/msg00047.html

Here is a short test case from glibc to demonstrate the problem. This
assembles fine with assembler before this check in.


.globl foo
 foo:
 .long 0
.weak foo
foo = foo

The error is

error "symbol `foo' is already defined"
--
Khem Raj <[hidden email]>
MontaVista Software Inc.

Reply | Threaded
Open this post in threaded view
|

Re: Regression caused by checkin.

Paul Brook
On Friday 21 October 2005 20:51, Khem Raj wrote:

> Hi Nick
>
> While building glibc with latest binutils from CVS, gas reported an
> error which was not there before. It is caused by this checkin
>
> http://sourceware.org/ml/binutils-cvs/2005-10/msg00047.html
>
> Here is a short test case from glibc to demonstrate the problem. This
> assembles fine with assembler before this check in.
>
>
> .globl foo
>  foo:
>  .long 0
> .weak foo
> foo = foo

This has been fixed in recent glibc on most targets.
Daniel Jacobowitz has patches for Arm.

Paul
Reply | Threaded
Open this post in threaded view
|

Re: Regression caused by checkin.

Khem Raj


Paul Brook wrote:

>On Friday 21 October 2005 20:51, Khem Raj wrote:
>  
>
>>Hi Nick
>>
>>While building glibc with latest binutils from CVS, gas reported an
>>error which was not there before. It is caused by this checkin
>>
>>http://sourceware.org/ml/binutils-cvs/2005-10/msg00047.html
>>
>>Here is a short test case from glibc to demonstrate the problem. This
>>assembles fine with assembler before this check in.
>>
>>
>>.globl foo
>> foo:
>> .long 0
>>.weak foo
>>foo = foo
>>    
>>
>
>This has been fixed in recent glibc on most targets.
>Daniel Jacobowitz has patches for Arm.
>  
>
I believe he has not posted the patches yet.

The gcc 3.4 generated

    .text
.Ltext0:
    .weak    mcount
mcount = _mcount
......
        b mcount
       
for

extern __typeof (_mcount) mcount __attribute__ ((weak, alias ("_mcount")));

and assembler gave this error
Error: Local symbol `mcount' can't be equated to undefined symbol `_mcount'

This might be related to another patch from H. J. Lu.

>Paul
>  
>

--
Khem Raj <[hidden email]>
MontaVista Software, Inc.
www.mvista.com

Reply | Threaded
Open this post in threaded view
|

Re: Regression caused by checkin.

Daniel Jacobowitz-2
On Mon, Oct 24, 2005 at 05:35:31PM -0700, Khem Raj wrote:
> Paul Brook wrote:
> >This has been fixed in recent glibc on most targets.
> >Daniel Jacobowitz has patches for Arm.
> >
> >
> I believe he has not posted the patches yet.

Sorry for the delay; I have now.  glibc HEAD should work.

--
Daniel Jacobowitz
CodeSourcery, LLC