Qfyilyi

The manual of drand48() says:

The drand48() and erand48() functions return non-negative,
double-precision, floating-point values, uniformly distributed over
the interval [0.0 , 1.0].

(source)

So just for clarification: Does this mean that 1.0 can be returned by drand48()? Or does it mean that the return value will be < 1.0?

I'm asking because in a brute force test drand48() never seems to return 1.0. It often returns 0, but never 1.0. However, for me the manual of drand48() seems to say that 1.0 is indeed a result that could come up, although I haven't been able to make drand48() return 1.0 in practice.

Anybody here who can shed some light onto this? Will the result always be less than 1.0 or can it also be equal to 1.0?

"over the interval [0.0 , 1.0]." It is safe to assume mathematical notation of interval. Parenthesis interval ends are exclusive and square brackets interval ends are inclusive. So according to the notation the returned value can be 1.0.

However this is a documentation of an old version and the current ones specify [0.0, 1.0) as the interval (which means it can't be 1.0):

• 
  

  The Single UNIX ® Specification, Version 2 http://pubs.opengroup.org/onlinepubs/7908799/xsh/drand48.html

  
  
  

  lists [0.0, 1.0]

  
  


• 
  

  while

  
  
  
  
  

  • The Open Group Base Specifications Issue 7, 2018 edition
    http://pubs.opengroup.org/onlinepubs/9699919799/functions/drand48.html

  
  

  • Linux Man Pages - POSIX.1-2001
    https://www.systutorials.com/docs/linux/man/3-drand48/

  
  

  • Linux man page - POSIX.1-2001.
    https://linux.die.net/man/3/erand48

  
  
  
  

  all list [0.0, 1.0)

  
  

So I would be inclined to believe [0.0, 1.0) is the correct specification. (it could be that the definition changed since SUS2 which is from 1997 or that it is a typo on the page, but I have no expertise on this so I won't weigh in)

in a brute force test drand48() never seems to return 1.0. It often
returns 0, but never 1.0.

This seems highly suspicious. 0.0 should be have the same probability to appear as any other number in the interval so "It often returns 0" doesn't look right. Check that you print with full precision.

As a side note be aware that dran48 and friends are obsolete

You are looking at a very old POSIX documentation. The POSIX 2017 for drand48 says:

The drand48() and erand48() functions shall return non-negative,
double-precision, floating-point values, uniformly distributed over
the interval [0.0,1.0).

This is also consistent with Linux man pages (which typically represents glibc implementation).

The same appears in POSIX 2001 too.