Requiring multiplication to be parenthesised against addition because
"precedence people often get confused about" is beyond words.

Still, easier to comply than to silence.

The added parentheses decrease readability of the code, thus split that
part out.

MacOS woes continue...

While technically it is available on MacOS, there's some terrible fuckery
wrt XCode versions not matching libraries on the rest of the system, various
versions not working with each others, etc.  Reportedly it has been fixed
but at least whatever is installed on CirrusCI doesn't work.

According to a 2017 commit in libressl getentropy() is not functional on
at least iOS, and I'm not inclined to find out more.

Thus, let's not bother, it's not like KBtin uses randomness for anything
important.  That 1980's-style getpid()^time() trick is no regression.

No regex-specific stuff for now.

Not bothering with configure checks, the function should be everywhere by
now (last major OS to get it was Solaris in 2015).  Would bother only if
proven otherwise.

runtime error: signed integer overflow: 53989 * 65537 cannot be represented in type 'int'

That's perfectly fine here (except for the 1980-style randomization...).

Colors proper preserve background and flags, flags preserve everything else.
That is, eg. "underline" doesn't change color to gray-on-black anymore.

Emulated flags (dim and inverse) don't work as expected yet, but that's
no regression.

By using ~::~ with empty fields, you can preserve an aspect of the existing
color; the fields being ~foreground:background:flags~.  Ie, :1 means:
 * retain foreground color
 * set background to dark blue
 * drop any flags

Can't (yet?) set/preserve an individual flag.

It seems the highcolor was supposed to be sticky somehow, yet the code
didn't actually implement that.  And the moron who wrote it didn't bother
to comment more than it's "an ugly kludge", thus the intent is lost with
my decayed brain cells.

Not very useful currently, but let's add non-destructive patterns later.

Seriously, all the string handling needs to be rewritten...

Same as #strlen except for ignoring color codes.

Needs cmake to be configured with -coverage for CFLAGS.

The behaviour did differ based on contents of unrelated memory...