Question about item's length and source length

[dsyrock]

Here is a simple reaper project that has only an item. I already set it to media source start and end. So the item's length and source length of it are suppose to be equal, right?

But I ran this script to check:

it=reaper.GetMediaItem(0, 0)

tk=reaper.GetActiveTake(it)

length=reaper.GetMediaItemInfo_Value(it, "D_LENGTH")

source=reaper.GetMediaItemTake_Source(tk)

orilen=reaper.GetMediaSourceLength(source)

msg("length="..length)

msg("orilen="..orilen)

msg("length-orilen="..length-orilen)

and I got this result:

length=3.5526530612245
orilen=3.5526530612245
length-orilen=4.4408920985006e-016

So why doesn't length-orilen equal to zero, but a number that is clse to zero?

[X-Raym]

Seems there is some roubding issue, maybe from a sample or sub sample value. Not sure why.

[juliansader]

Check out this thread: GetMediaItemInfo_Value API does not return the real values.

* REAPER's IDE and ShowConsoleMsg don't display floating point numbers to their full precision, so numbers may seem equal even if they are not.

* 4.4408920985006e-016 is close to the limits of the previous two numbers' 64-bit float precision, so this is probably just normal floating point errors, not sample rounding.

* Floating point numbers should never be compared directly for equality. Instead, compare their difference to a tiny cutoff value such as 1e-14 (or whatever is about 14 decimals smaller than your expected value range).

[_Stevie_]

Could it be, that this issue happens, because the take properties (in ticks) get converted to project time and therefore have a different precision?

[schwa]

4e-16 is less than one-billionth of a sample at 48k

[X-Raym]

@schwa
This can be enough to break some equality check :P

[_Stevie_]

I found another solution for this and wanted to share it.
I'm using the following round function in order to round both values, 5 digits after the comma:

local function Round(num, idp)
  local mult = 10^(idp or 0)
  return math.floor(num * mult + 0.5) / mult
end

Thanks to LBX for the function!

[_Stevie_]

Okay, strike that, it works with some values, but not all.

[Meo-Ada Mespotine]

The craziest thing I once did was to convert these values to string and round them character by character. Then convert back into number.
I didn't know anything about float-precision at that time o_O

[_Stevie_]

Been there, done that, got the T-Shirt...