Re: sawfish 1.6.3 can't compile sawfish.wm.util.prompt

[Harald van Dijk <harald gigawatt nl>]

On 23/05/10 16:41, Jeremy Hankins wrote:
> Harald van Dijk<harald gigawatt nl>  writes:
>    
> > [...]
> > (compiling sawfish/wm/util/ping.jl)
> > (compiling sawfish/wm/util/prompt.jl)
> > error-->  (void-value prompt-completion-fun)
> > make[1]: *** [lisp] Error 10
> > make[1]: Leaving directory `/home/harald/sawfish/lisp'
> > make: *** [all] Error 1
> >
> > This file is unchanged from 1.6.2, but as of commit
> > 2277ff2ca26ea48fd6ce96f45faa4971a52bc81a
> > <http://git.gnome.org/browse/sawfish/commit/?id=2277ff2ca26ea48fd6ce96f45faa4971a52bc81a>
> > ,sawfish.wm.util.prompt is no longer implicitly included when compiling
> > itself. Note that you will not see this error if you are working from
> > a previous build directory where the .jlc files already exist. I
> > noticed the build completes without errors or warnings if I simply
> > remove the "Compilation hack" block; is it still necessary? If it is
> > necessary, another way to fix the problem is by having
> > sawfish.wm.util.prompt explicitly (open) itself.
> >      
> I haven't been able to replicate this, and it doesn't make a whole lot
> of sense to me.  A file shouldn't have to be opened from itself in order
> to compile it, and I can't see where a (setq foo bar) would generate a
> void-value error on foo.
>    
As far as I could tell, if the file is not opened from itself, the 
defvar is only evaluated after the setq, because the setq is part of an 
eval-when-compile. (open)ing prompt.jl from within itself ensures the 
defvar is processed earlier. I am not sure why you are not seeing this, 
I will try to track this down further. For reference, I started the 
build with librep 0.90.5, rep-gtk 0.90.2, and sawfish 1.6.2 installed, 
no git snapshots. And I do not think this should matter, but it's 
happening on a fairly standard x86_64 linux system.