Re: [Bug-tar] spurious warning in short_read

[Mads Kiilerich]

On 05/22/2009 02:09 PM, Samuel Thibault wrote:
> Hello,
>
> Since at least tar 1.22, there is an additional warning in short_read:
>
>    if (left&&  left % BLOCKSIZE == 0
>        &&  verbose_option
>        &&  record_start_block == 0&&  status != 0)
>      {
>        unsigned long rsize = status / BLOCKSIZE;
>        WARN ((0, 0,
>               ngettext ("Record size = %lu block",
>                         "Record size = %lu blocks",
>                         rsize),
>               rsize));
>      }
>
> Why is that done?  Reads on pipes typically return short reads, and
> these are typically aligned on page size, and thus record_size (10240
> here) % BLOCKSIZE is typically 0...
>    

It has been discussed before, for example at 
http://www.nabble.com/tar:-Record-size-%3D-8-blocks-td22265061.html and 
http://www.nabble.com/record-size-%3D-x-blocks--td22477515.html

The current behaviour was introduced because it sometimes provides 
useful information on tape / block devices. FWIW I still don't see how 
any kind of semantics in read chunk sizes can ever be reliable anywhere.

/Mads