bug#11424: coreutils: split tests hang on /dev/zero on GNU/Hurd

[Jim Meyering]

Paul Eggert wrote:
> On 05/07/2012 12:46 AM, Jim Meyering wrote:
>> +
>> +  /* stat.st_size is valid only for regular files.  For others, use 0.  */
>> +  file_size = S_ISREG (stat_buf.st_mode) ? stat_buf.st_size : 0;
>
> Is it right to use 0 there, for non-regular files?
> Won't later code compute incorrect sizes in that case?

Hi Paul,

I agree that more change is required and do prefer the direction your
patches suggest.  However, to fix the Hurd/infloop with minimal
impact elsewhere, I have a slight preference for my small change.
I.e. continuing to operate on non-regular files with --number we
don't have to change the split --number tests that operate on /dev/zero.
Then, introducing the behavior change (with your follow-on patch) can
be independent of the bug fix commit.

I do admit that without being able to determine a size up front, there's
little point in using that option, so your patch (reject files with
unusable stat.st_size) is required.

With or without my patch on Linux/GNU, if you split /dev/zero,
it sets file_size = 0, so at least for the tested cases
I don't think that patch introduces a regression.

> Also, as a nit, stat.st_size is also valid for
> SHM and TMO files (this was in the patch I just sent).

Good point.
Do you feel like adding something like this to system.h
and completing your patch?

    /* Return a boolean indicating whether sb->st_size is defined.  */
    static inline bool
    usable_st_size (struct stat const *sb)
    {
      return S_ISREG (sb->st_mode) || S_TYPEISSHM (sb) || S_TYPEISTMO (sb);
    }