binutils 2.20 regressions (20091002)

[Tristan Gingold]

On Oct 2, 2009, at 5:52 PM, Dave Korn wrote:

> H.J. Lu wrote:
>> On Fri, Oct 2, 2009 at 8:35 AM, Dave Korn wrote:
>>> Tristan Gingold wrote:
>>>
>>>> Should we include Jakub's patch on .cfi_* directives ?
>>> That's obviously highly desirable, since there won't be another  
>>> official
>>> binutils release before gcc-4.5.0 comes out, but of course it's  
>>> risky.  It
>>> looks safe-ish because a lot of it is semi-mechanically wrapping  
>>> conditionals
>>> around the existing code, but of course just one typo'd/wrong/ 
>>> missing
>>> condition amongst them could end up generating bad debug info.
>>>
>>
>> I'd like to see testcases for those new .cfi_* directives first. We
>> have no ideas how they should work without testcases.
>
>  Good point.  A good set of testcases (including checks that old  
> behaviour is
> still correct) could go a long way toward mitigating the risk.

Agreed.

Note that it is possible to make a 2.20.1 release around January that  
includes this patch.  I think it is
safer than including this patch first and maybe create in a hurry a  
new release if a bug is found.

Tristan.