[PATCH v2] Add new tests for fopen

[Adhemerval Zanella Netto]

On 02/02/25 10:47, Florian Weimer wrote:
> * Martin Coufal:
> 
>> +static void *
>> +threadReadRoutine (void *argv)
>> +{
>> +  struct thread_data *my_data;
>> +  unsigned char read_buffer;
>> +  int ret = 0;
>> +  my_data = (struct thread_data *) argv;
>> +  /* Wait for all threads to be ready to read.  */
>> +  xpthread_barrier_wait (&barrier);
>> +
>> +  ret =
>> +    fread (&read_buffer, sizeof (char), sizeof (read_buffer), my_data->fd);
>> +  if (feof (my_data->fd) != 0)
>> +    {
>> +      clearerr (my_data->fd);
>> +      my_data->eof = true;
>> +    }
>> +  else
>> +    {
>> +      TEST_COMPARE (ret, 1);
>> +      /* Save the read value.  */
>> +      my_data->value = read_buffer;
>> +    }
> 
>> +      for (int i = 0; i < NUM_THREADS; i++)
>> +        {
>> +          if (thread_data_array[i].eof)
>> +            {
>> +              /* EOF was read.  */
>> +              present_values[NUM_THREADS - 1] = true;
>> +              eof_cnt++;
>> +            }
>> +          else
>> +            {
>> +              /* The same value shouldn't be read twice.  */
>> +              TEST_VERIFY (!present_values[thread_data_array[i].value]);
>> +              present_values[thread_data_array[i].value] = true;
>> +            }
>> +        }
>> +      /* EOF is read exactly once.  */
>> +      TEST_COMPARE (eof_cnt, 1);
> 
> This test has a race condition: more than one thread can report
> end-of-file because the check for this condition is separate from the
> fread call.  It is possible that all threads return from their fread
> calls before feof is called on multiple threads.  This test bug is quite
> visible on larger AArch64 systems, for example.

It has failed on unrelated patch as well [1]. Should we revert this?

[1] https://patchwork.sourceware.org/project/glibc/patch/20250306183418.3086902-1-aurelien@aurel32.net/