AllowEmptyInput and HAL (SOLVED!!!)
dbn.lists at gmail.com
Fri May 8 16:35:36 PDT 2009
On Fri, May 8, 2009 at 10:46 AM, Phil Endecott
<spam_from_xorg at chezphil.org> wrote:
> Finally solved, after something like 60 hours of hacking.
> My libX11.so was old. This caused xkbcomp to fail to parse the keymap
> files - it didn't recognise ISO_Level5 stuff. It looks like xkbcomp
> generated a keymap with some sort of Any+Any definition that caused
> every key to toggle the modifiers.
> This morning I attempted to purge and re-install all of my X-related
> Debian packages. I got rid of the server stuff but I couldn't get rid
> of the client libraries because that would cause huge numbers of client
> programs to be removed too. But I didn't worry about that because the
> problem was "clearly" on the server side. I eventually latched on to
> the "level 5" warnings and started looking for them with strings. Then
> ldd on xkbcomp pointed at the guilty library.
> I'm surprised that it didn't work with the xserver that I built using
> khbuild. Presumably this is because that was still using libraries
> from /usr/lib, not the ones that it had just built and put somewhere in
> $HOME. Is that an rpath issue with khbuild?
It's because xkbcomp links with libX11.so, not the server. However, if
you built xkbcomp and Xlib from jhbuild, it should pick up the new
Xlib during the xkbcomp build. And libtool usually adds a rpath if it
detects one of the libraries is outside of the standard linker path.
You'd have to inspect what jhbuild is doing if that's not the case.
> What can we learn?
> - Debian needs a more strongly-versioned dependency between some of its
> packages. I'll take that up with them (if the appropriate people are
> already reading this, please let me know).
> - xkbcomp, when called by the server, claims that "xkbcomp errors are
> not fatal to the server". And only some of its messages appear in the
> X log file. I feel that this particular error should be considered
> fatal. And ideally, the consequence of an unrecognised keysym should
> not be the Any Key effect that I got. (Is the person responsible for
> xkbcomp reading this?).
No one is really responsible for xkbcomp currently. The entire
interaction with xkbcomp by the server is a complete eye sore. I've
been working to make this go away, but it took a lot of surgery and
hasn't been reviewed yet.
More information about the xorg