Here is the fix! (was Re: LFS build method flawed?)

Bruce Dubbs bdubbs at swbell.net
Thu Jul 11 17:34:04 PDT 2002


Nicholas Dille wrote:

>On Thu, Jul 11, 2002 at 10:02:59AM -0500, Bruce Dubbs wrote:
>  
>
>>I don't think thats a problem for most of us.  :)
>>    
>>
>
>i disagree. if you've implemented a mathematical algorithm and provided
>a test which is meant to check the correctness of the impementation (in
>this case the resulting binary from the implementation) any error is
>entirely unacceptable.
>
>i think you are underestimating the problems which might arise from
>error propagation. an error of 10^-6 might unexpectedly grow into by
>several magnitudes in a disadvantageous context.
>
>so, in my opinion we should not fix the test but rather that which
>causes the implementation to fail!
>(assuming that the results which glibc's calculation is tested against
>are correct)
>
I am not underestimating the propagation problems.  If you are using 
those functions, the problem is serious.  The key word here is 'IF'.  I 
can't think of any mainstream application that would use the function 
that failed.  Specialized users who need that type of function can and 
do check them (as in the glibc check).

Besides, I'm fairly confident that the error is in hardware, not glibc. 
 If the test passes on Intel wouldn't you draw the same conclusion?  

  -- Bruce  

-- 
Unsubscribe: send email to listar at linuxfromscratch.org
and put 'unsubscribe lfs-dev' in the subject header of the message



More information about the lfs-dev mailing list