bug#6683: mktemp foo.XXXXXXXXXXX is not sufficiently random

[Paul Eggert]

On 07/20/10 10:41, Eric Blake wrote:
> Meanwhile, glibc's mkstemp() only replaces the last 6 X, regardless of
> how many additional X are present in the template.  Do we even need the
> extra randomness if the template contains more X?

Well, I did say that it was a _theoretical_ bug.  You need the extra
randomness if you run mktemp about 18e18 times (or more, of course).

Limiting it to the randomness needed for 6 Xs would give about 57 million
possibilities, which is fine for most applications, but the arbitrary limit
does rankle a bit given that one of GNU's tenets is no arbitrary limits.
The current coreutils code limits it to the randomness needed for about 10.7 Xs,
but that also is arbitrary, and it's easy to remove the arbitrary limit.