Integer to double check fails when return object contains run time

[helske]

Currently, the check where the integer is replaced with double can fail if the result contains information about the run time. I thought about fixing this by first repeating the original call, then comparing whether the double version differs more from the original call than its duplicate (in terms of length of setdiff), i.e.

no_change <- length (setdiff (out1, out2)) == length (setdiff (out1, outb))

But I then I noticed this line:

autotest/R/input-int.R

Line 367 in dbdf264

Sys.sleep (0.5) # in case Sys.time is used

So now I wonder if this fails in purpose or is that related to something else?

[mpadge]

@helske Sorry it's taken so long to get back to these issues. Is this one still important? You're proposing to replace this line:

autotest/R/input-int.R

Line 378 in 7bde4ba

no_change <- length (setdiff (out1, out2)) == 0

... but i'm unsure what you intend outb to be in your above suggestion. At that point there are only the two objects for comparison, out1 and out2.

More generally, any information about run time should generally be some kind of time-specific object, and so should not trigger these integer tests, unless I'm misunderstanding?

[helske]

If I remember correctly, the issue with the current version was that if out1 and out2 are for example lists with one list element containing the runtime such as output from system.time, then no_change will be FALSE. In my not very elegant suggestion outb is a repeat of the original call, so the idea was to test whether two repeated calls with integers differ less than the call with double conversion.