This question pertains to

        M-x set-input-method RET tamil-trans RET

I am on Debian (Squeeze) running XFCE.  


The attached screenshot shows the difference in behaviour between Emacs
launched from Bzr (trunk) and Emacs23 that is distributed with Debian
itself.  The character that is shown is the one you get by typing "ki"
in tamil-itrans.

You will note following differences:

1. Emacs23 uses Lohit font while EmacsBzr picks up gnu-unifont.
2. Emacs23 "composes" the glyphs correctly while EmacsBzr messes up the
   display.

I have the following in .emacs.

    (set-face-font 'default "fontset-default")
    (set-fontset-font "fontset-default" 'tamil "Lohit Tamil")

and xterm reports the following.

    ,----
    | kjambunathan@debian-6:~$ fc-list :lang="ta"
    | TSCu_Paranar:style=Bold
    | TSCu_Times:style=Normal
    | TAMu_Maduram:style=Normal
    | FreeSerif:style=Medium,Mittel,µεσαία,Normal,Gemiddeld,navadno
    | TSCu_Comic:style=Normal
    | TAMu_Kadambri:style=Regular
    | TAMu_Kalyani:style=Regular
    | Lohit Tamil:style=Regular
    | unifont:style=Medium
    | TSCu_Paranar:style=Regular
    `----

I would like to get Emacs23 behaviour in EmacsBzr.  Why do the two
Emacs'es behave differently?  What settings do I need to tweak?

I am not much familiar with X Desktop environment and I have built my
Emacs in no special way (i.e., in a way any newcomer would).