bug#41544: 26.3; Possible incorrect results from color-distance

[Eli Zaretskii]

> From: Simen Heggestøyl <simenheg@runbox.com>
> Cc: Mattias Engdegård <mattiase@acm.org>,  Richard Copley
>  <rcopley@gmail.com>, 41544@debbugs.gnu.org
> Date: Sun, 07 Jun 2020 11:04:30 +0200
> 
> I can't comment on the patch overall, but this part at least seems to
> address Richard Copley's complaints in bug#30295. A dark foreground is
> now used for #0f0/rgba(0,255,0,0.5) and #5e5 as Richard requested, which
> indeed looks more readable to me too.
> 
> Maybe the pendulum has swung too far however. For instance, a dark
> foreground is now used for #ef716e, which I think was easier to read
> with the light foreground used before. Could that be fixed by tweaking
> the cut-off values in color-dark-p, perhaps?

We can (and did) tweak the various constants involved in this, more
than once.  The lesson I took from that is that we could never produce
something that will fit all the needs, let alone satisfy all the
users.  There are two main factors here on which we have no control:
the subjective differences between color perception by different
people, and variations in how different terminals and displays show
the same colors.

That is why I object to making changes in this low-level functionality
with the motivation of "fixing them all": I think it's simply
impossible.  We will make it slightly better in some situations and
slightly worse in others.  There's no net win here.