I came across this paper just now by Nevin Liber:

    "We shouldn't rush to require either CHAR_BIT==8 or (u)intNN_t"

which is a rebuttal to the following paper by JF Bastien:

    p3477r3 - There are exactly 8 bit in a byte

I want to add the point that I program microscopes 5 days a week that
have a Texas Instruments microcontroller inside them that has CHAR_BIT
== 16.

There are compilers alive and kicking today, getting used to produce
saleable products, that have CHAR_BIT