Re: [PATCH] head,tail: consistently diagnose write errors

[Bernhard Voelker]

On 01/30/2014 12:02 AM, Bernhard Voelker wrote:
> On 01/29/2014 07:43 PM, Pádraig Brady wrote:
>> On 01/29/2014 12:06 PM, Pádraig Brady wrote:
>>
>>> --- a/NEWS
>>> +++ b/NEWS
>>
>>> +  tail now diagnoses all failures when writing to stdout.  Previously write
>>> +  errors could have been silently ignored if some data was output.
>>> +  [This bug was present in "the beginning".]
>>
>> Actually this is not the case. The atexit handler would have caught
>> partial write failures, albeit with a less detailed error message.
>> So I'll remove the NEWS but leave the code changes.
> 
> Thanks.
> 
> It seems your patch didn't make it thru the mail systems:
> 
>   error: patch failed: src/head.c:504
>   error: src/head.c: patch does not apply
>   Patch failed at 0001 head,tail: consistently diagnose write errors
> 
> Can you resend as attachment, please?

Sorry, I missed that this patch was based on the other patch for
bug http://bugs.gnu.org/16329. After that, it applies cleanly.

> However, if head is exiting earlier on the first write error,
> then there should be a user visible change.
> 
>   $ seq -f'%1000g' 1000 | { src/head --lines=-1 > /dev/full ; wc -l ; }
>   src/head: write error
>   0
> 
> With the patch, I'd expect wc to see more than 0 lines in the above
> case, right?

Indeed, now with the patch:

  $ seq -f'%1000g' 1000 | { src/head --lines=-1 > /dev/full ; wc -l ; }
  src/head: write error: No space left on device
  src/head: write error
  984

Unfortunately, the atexit code now produces a second error diagnostic.
It doesn't hurt, but it looks a bit ugly.

Have a nice day,
Berny