Gentoo Forums

Hi, Gentoo forums! I have a "why is that?" question I hope someone would be kind enough to answer.

I was thinking... when setting a keymap, the mappings between keysyms and char codes are determined by the charset you choose for this purpose. So, in essence, the charset selects what character encoding the keyboard driver uses when talking to the rest of the system. Now, it seems to me that if this is different from the character encoding you've told the rest of the system to use, you have a potential problem. So, one would think that the set of available charsets would be the same as the set of available character encodings in general. In fact, one might even expect that they would be rigged to automatically be identical in all cases.

Unfortunately, this appears to be totally different from reality. I'm amazed to learn that no Unicode charset exists. So, if you use Unicode, your keyboard driver will always give you "wrong" values in some cases, right? Why is there no Unicode charset? Surely there is some technical reason that my tiny, shriveled brain has not yet clued in on?

That was my main question. But I also wonder, why would you ever want the charset for the keyboard driver to be different from the rest of the system? And in fact, why can't you just make your own charsets? I mean if it's just a mapping between keysyms and char codes, that sounds like a simple text file.

Excuse my ignorance on the subject. I've searched around a fair bit and I just feel like there must be something I'm missing.

Thanks!

OK, so nobody knows that one. Well then, let me try a (hopefully) easier one:

Why does the comment in /etc/conf.d/keymaps say that DUMPKEYS_CHARSET only matters if you set UNICODE="yes" in /etc/rc.conf? I would think the reverse would be true.

Any takers?

Yikes. After lots and lots of painful experimentation and kernel source code reading, I've determined that, if I'm understanding everything correctly, Linux's keymap support is really, really, REALLY broken. Is that even possible?

It looks like Gentoo's DUMPKEYS_CHARSET is a mistake on someone's part. It is being used to basically tell loadkeys what encoding an earlier invocation of loadkeys autodetected. This doesn't make sense to me as-is, but it makes even less sense when you consider that loadkeys autodetects these things on a key-by-key basis, so there's pretty much no right answer for this setting unless you're using iso-8859-1. That certainly would explain the poor documentation for what to put there! The fix (again, assuming I'm understanding everything right) should be to remove the lines: from /etc/init.d/keymaps, and instead add an if-statement to add "--unicode" on the /bin/loadkeys command line a few lines above that whenever UNICODE="yes". Is it possible to tell a developer about this, assuming I'm right (which I'd rather not assume without some confirmation)?

But there are more problems. Unless I misunderstand, "compose" definitions (which are the coolest thing ever, when they work) are only allowed for single-byte encodings, and must always be specifically iso-8859-1 when Unicode is in use. Great. So as long as we're all Western Europeans there's no problem. I consider this a bug in the keymap file format, and therefore a bug in loadkeys and dumpkeys. Not Gentoo's fault, I guess.

But wait, there's more. I'm less sure about this one, but it really does appear that the kernel cannot handle Unicode characters above 0xEFF. Since that leaves out all the astral planes, as well as a few categories of special characters in plane 0, it doesn't seem like full Unicode support to me.

Actually, though, I still can't understand why the decision was made to try and support Unicode as this whole other category of life instead of just another charset (as mentioned in my original post). I don't think the keyboard driver should know what UTF-8 is at all. loadkeys and dumpkeys should know about it, and just map the keysyms to the appropriate "macro" of several successive action codes to make the kernel output the correct multibyte sequence when a key is pressed. This, combined with a fix to make compose definitions work based on keysyms instead of literal single-byte values, would completely fix everything and end the scourge of console-mode keyboard localization problems for all time. More importantly, I would finally be able to set up the compose definitions I've been aching for

I would be deliriously happy to fix all of these myself, except that I doubt I would know how or where to post it to get anyone else to use the fixed versions, and I get the distinct feeling nobody really cares much about any of this console Unicode junk anyway, so I'd probably be the only user of it, which would make me feel like a pathetic loser, so barring a sudden surge of interest in this monologue thread, I'm probably not going to do it.

Heh, well in case anyone was interested, there's the final conclusion. I'll leave you alone now.

Actually.. I've noticed this before as well. While annoying, I just shrugged it off and did without euro symbol support. Charmap bug fixes can go to bugs.gentoo.org. If you want to fix it, great.

Xezlec wrote: I would be deliriously happy to fix all of these myself, except that I doubt I would know how or where to post it to get anyone else to use the fixed versions, and I get the distinct feeling nobody really cares much about any of this console Unicode junk anyway, so I'd probably be the only user of it, which would make me feel like a pathetic loser, so barring a sudden surge of interest in this monologue thread, I'm probably not going to do it.

NOOOOO!

You should definitely post your findings here:

http://bugzilla.kernel.org/

They usually answer quite promptly.

Oh, ok, thanks!

Do you happen to know who is responsible for loadkeys/dumpkeys? I haven't been able to figure that out. Of course, I could just write my own versions of these programs, since I don't really like the current versions overall. I guess if I ebuild it, they will come.

But again, you can lodge bug reports with bugs.gentoo.org, and it should get pushed upstream.

Awesome. That's good to know.

As far as the kernel is concerned, the unicode.txt file in /usr/src/linux/Documents
gives F000-F8FF as the Commercial (Linux) Zone, and E000-EFFF as the End User
Zone, with F000-F0FF hard coded to map directly to a loaded font, without going
through translation. But I don't know if the information is still current.

Will