bug#7357: csplit: memory exhausted when using stdout / pipe instead of a file

[Jim Meyering]

Bob Proulx wrote:
> Jim Meyering wrote:
>>   http://git.sv.gnu.org/cgit/coreutils.git/commit/?id=cf0fa2d930de1609
>>
>>     Use seq, not `yes' to generate link target.
>>     Otherwise, on systems (DJGPP) that emulate pipes using files,
>>     this test would never complete, waiting for `yes' to terminate.
>>
>> diff --git a/tests/du/slink b/tests/du/slink
>> -  name=`yes|tr '\n' y|head -c$len`
>> +  name=`seq 1 $len|tr -c x y |head -c$len`
>>
>> Is DJGPP still relevant?
>> The last update of http://www.delorie.com/djgpp/ was in 2003.
>> Now, I'm tempted to leave your seq|head pipe there,
>> just to see if anyone notices.
>
> Is using files to emulate pipes even a valid operating model?  I
> wouldn't have thought so.  In which case any system like that is just
> insufficient to the task.  Avoiding pipes just for such systems
> doesn't seem very productive to me.
>
> I would be inclined to have a test not for the program but for the
> operating system to ensure that the OS is sufficiently capable.

Precisely.  That commit was over 7 years ago (mea culpa ;-).
Our testing practices/policies have evolved since then,
and the approach you and Pádraig suggest is definitely what
I prefer now.  There are many examples of skipping tests on
less-capable systems.