[metapost] quality of randomness

Larry Siebenmann laurent at math.toronto.edu
Sun Mar 13 06:14:30 CET 2005

Hi Taco,

 > All web2c-based distros, as well as most commercial ones I know of,
 > already use gettimeofday() if and when that is avaible. And any
 > system that does not have that, is very unlikely to have something
 > even newer like /dev/random.

Hans and Jacko and I don't quite see it that way. Where exactly is the
"gettimeofday()" recommendation and related stuff? If it is not in the
"mp.web" or other mainline doc, how do you expect implementors for all
platforms to stay in step?  They are mostly NOT mp users.
And they are not all Web2C desciples.

 > As I said before, my advise to Larry is to get a current distribution.

I have one.

Incidentally the "gettimeofday()" is a dead issue for me since
I provided a perfectly portable workaround.  The live issue is
the one Karl raised, namely the quality of the random numbers
generated for "uniformdeviate" and "normaldeviate".
 > (There might be better things
 > than /dev/random these days, for that matter; a lot of work has gone
 > into this topic, since ssh and other crypto applications need
 > randomness.)

I have a vague impression that my ProfileBezier.mp turns up
too many runs of the same bezier type.  Maybe a randomness
quality failure ?

Laurent S.

More information about the metapost mailing list