[PATCH] tst-fopen-threaded: Only check EOF for failing read

[Florian Weimer]

* Siddhesh Poyarekar:

>> You don't mention anywhere why you remove the clearerr call?  I think
>> that's necessary because the test assumes that feof is sticky.
>
> Would it change the outcome though?  If there are multiple failing
> readers, they would reliably result in the test failing regardless of
> whether the EOF/error flag is cleared.

I think you get another race if you leave in the clearerr because a
reader that sees EOF in fread (0 return) might not observe ot in feof
because it got reverted by a concurrent clearerr (which clears the EOF
indicator, too).

(By the way, it's not clear from the patch context that ret == 0 is
correct here, that's because sizeof (read_buffer) is 1 is as well.  So
the short read scenario cannot really happen.)

Thanks,
Florian