Re: The Chthulhoid horror that is keyboard handling in Unix

[Magnus Eriksson <magetoo%fastmail.fm@localhost>]

On Thu, 1 Apr 2010, Johan van Zanten wrote:

> That's a very sweet and wholesome concept, and i'm really amused by the
> notion that you think we've passed out of the Stone Age.

:-)

> But you are
> ignoring the reality that we are still typing on mechanical keyboards
> ...

> I'm just as ready as the next humanoid for the socket at the base of the
> skull that translates my intent into pure, divine manifestation, but the
> fact of the matter is that it just doesn't yet exist.

That would perhaps be nice, but I think you overestimate the magnitude of 
what I'm getting at here.  (I love my Model M, and I plan on using it for 
at least another decade.)


> > What I mean is this: The up arrow key doesn't send a "move cursor up
> > character"[*]

> Really?  I think it does. Ahh, well maybe it's several bytes of key 
> codes (characters), but you see what i mean, perhaps.

See, this is why I put the asterisk in there.  :-)

Right, so the keyboard wiggles the data line for a while to send a stream 
of bits to the keyboard controller.  I don't really care whether it sends 
PS/2 or USB scancodes, or whether the PS/2 scancodes are set 1, 2, or 3, 
as long as they communicate my intent to the kernel.  (I would bet real 
money that you don't particularly care about this either, as long as it 
works.)

In any case, since I am running X, the scan codes that X gets are Set 1, 
of Original IBM PC vintage; they are converted by the kernel if the 
keyboard controller hasn't already done the job.  It's a stupid system, 
but seems to be what X expects.  The X server translates it to an X 
keycode.  I still don't care.

Somewhere in there though, another part of X converts the keycode, 
representing the physical key, into a keysym, representing a symbolic 
name.  If all of the above worked as it's supposed to (and amazingly 
enough it did) the keysym that gets delivered to the program is "Up".

*This* I do care about.  Not the details of how it got there, but "Up" is 
what I want.


The short version: Hit key, irrelevant implementation details, "Up" 
keysym.  I hope you (all) will agree with me that the short version is 
easier to understand.


But we're not done yet, of course.  The program in question is often a 
terminal emulator, and now we get into the perhaps more familiar territory 
of terminal emulations.  (Although to me, it's a strange and frightening 
land, haunted by the ghosts of yesteryear...)

You know the deal; "Up" gets turned into some escaped sequence; shell / 
text editor receives escaped bytes and decodes it, and does what the user 
wants.

More can be said, but again, these are implementation details.  Why should 
the user have to care?  It should just work.  I certainly didn't care 
before switching to Unix and had to fight against the system to get it 
to release my "national" characters.


Are there users who care strongly about what the terminal emulator sends? 
I guess there are quite a few, especially here, but xterm/rxvt aren't 
going away.  However for a large number of users, the exact details are 
irrelevant.  I would certainly trade VTnn compatibility for unambigious 
character encoding (let's hear it for UTF-32!) and having all special keys 
working out of the box, in normal daily use.  That can't be so hard, can 
it?


> >  Asking users to set environment variables and mucking around with
> > dot files is the wrong way to go about things, unless the goal is to annoy
> > people.

> Don't fear the command line.  It's where the power is.

Oh please, not the "WIMPy GUI user" argument.  I don't fear the command 
line.  But I also know that as things stand today, it's lousy for some 
things; does it really make sense to *require* users to got intimate with 
obscure parts of the system before being able to speak to the system in 
their own alphabet?

(Would you find it acceptable if by default, 'q', 'w', and 'z' didn't 
work?  "Oh yeah, you have to 'set input-meta on' to get those 
international characters."  I assume you would at least be pretty annoyed. 
But that is the reality for a lot of us today.  We rely on The Google, and 
passed-on but poorly understood fragments of configuration to get our 
systems into something resembling a usable state, and we still avoid using 
"international" characters if we have any sense.  And that's even before 
getting into "properly" international things like UTF-8.)


But the short version: it could be better.  So why not figure out the 
right way to do it?


MAgnus