Unintuitive + undocumented behavior when NaN used in for-loop

["Sebastian" <sebastian@...>]

I found some strange behavior when NaN is used in a numeric for-loop:

	-- only executes once
	for i = 0/0, 10 do
		print(i) -- -nan
	end

	-- only executes once
	for i = 0/0, 0/0 do
		print(i) -- -nan
	end

	-- never executes
	for i = 1, 10, 0/0 do
		print(i)
	end

	-- executes once
	for i = 10, 1, 0/0 do
		print(i) -- 10
	end

In the first two examples, the crux of the problem is the initial check
in FORPREP checks that limit < init, but subsequent checks in FORLOOP
check that idx <= limit. So NaN passes the first time, but not the
second time.

The latter two examples show that NaN is always treated as negative when
used as the step, even if you do math.abs(0/0). The check for whether
the step is negative is 0 < step, which is of course false for NaN.

Maybe using NaN within a numeric for-loop should be disallowed (just as
using 0 as the step is disallowed)?