Normalization of rfft~/rifft~ different from Pd application

[grrrr]

Hi, using spectral processing in a Camomile patch, it turns out that the signal resulting from an
rfft~/rifft~ passthrough is higher by a factor n (= fft size) compared to the vanilla Pd application.
I guess this is likely a libpd problem, but i lack the insight to prove this.

Attached, you will find the source to the camomile plugin
spectral_pass.zip
with a two-channel spectral pass-through

and a simplistic reaper project illustrating the issue:
spectral_pass_test.RPP.zip

The platform is MacOS 10.14, building a VST3 plugin using Camomile version 1.0.7