Convert to and from FLAC on floating point signals

[vascofsantos]

(Audacity 2.3.1 on macOS 10.14.5)

I was testing the losslessness of the FLAC coder/decoder in Audacity using a 32-bit WAV created with random noise generated within Audacity itself, and I got some tiny roundoff errors when comparing the original WAV (32-bit float) with the decoded FLAC signal.

I read about it here: https://xiph.org/flac/faq.html#general__samples_fp
It seems that FLAC doesn't (and probably never will) support floating-point signals, so these errors must be happening because there are float-to-int and/or int-to-float conversions going on before/after the codec.

So, shouldn't Audacity warn about this to the user upon exporting floating-point audio to a lossless format?
I mean, some people are probably using this without realising that they are actually getting some (albeit tiny) losses on exporting to a format they might perceive as being completely lossless.

Cheers,
Vasco.

[steve]

When you export from Audacity to FLAC, there is a choice of 16-bit or 24-bit. If you are converting from 32-bit float to 16 or 24 bit, then I think it should be pretty obvious that there has to be some sort of rounding. If we warn about this for FLAC, then logically we would have to warn users for every format other than 32-bit float. I don't think most users would want that.