[pulseaudio-discuss] /.libs/.libs/ gotcha

[Peter Meerwald]

Hello,

> > > I tried specifying "-p $(pwd)/src" instead, but that fails in another way:
> > > the
> > > ".ifexists" command in our CLI does not add a ".libs" the way ltdl does,
> > > so
> > > the result is fewer errors, but then, it only loads the required modules,
> > > and
> > > prefers module-detect over module-udev-detect, e g, which is quite broken.
> > 
> > -p $(pwd)/src worked for me, but maybe I did not look close enough
> 
> Could you look a bit closer and see if you get a warning message saying that
> you use module-detect instead of module-udev-detect?

yes, same here, it is using module-detect

> > > Any thoughts on the topic? Should we add ".libs" too, when looking for
> > > files
> > > in ".ifexists"?
> > > Or should we lower the severity of not finding modules? Perhaps write with
> > > debug severity instead, and do that regardless of if we loaded a module or
> > > not?

> I would probably go for lowering severity + reverting the changes that suggest
> using "src" instead of "src/.libs", because there does not seem to be a way we
> can get the ".libs" string out of libtool for our own search (for the
> ".ifexists" command).

I think that the idea of ifexits is broken; the rules of lt_dlopenext() 
loading a module are complex; .ifexists only checks if a file is present

probably
  .nofail
  load-module
  .fail
should be used instead of
  .ifexits module-xy
  load-module module-xy
  .endif

maybe tryload-module could be added (with the semantics of ".nofail / 
load-module / .fail")?

but that's probably not a short-term solution

I think ".libs/" is a libtool internal and should not appear in PA; 
lt_dlopenext() should be able to figure out the .so file from an .la file 
(which lives in src/) -- have not tried


a quick hack would be to 
(i) revert the README change,
(ii) remove (or demote) the logging from bind_now_open() -- 
bind_now_open() is called multiple times by ltdl


regards, p.

-- 

Peter Meerwald
+43-664-2444418 (mobile)