Re: deplibs.test broken.

[Ralf Wildenhues]

Hi Kurt,

Sorry for the late reply.  I overlooked your mail the first time.
(Feel free to send reminders if answers take time.)

* Kurt Roeckx wrote on Sun, Jul 31, 2005 at 02:06:22PM CEST:
> 
> I'm having a few problem with the deplibs.test.  It's broken for 
> several reasons:
> - It's linking a shared lib against a static lib.  This is not
>   portable, and we even give a warning about this.

Yes.  It should only be run when deplibs_check_method=pass_all.
Unfortunately, that is set wrongly in some cases (e.g., linux/x86_64).

> - The shared lib is not using any symbols from the static lib.
> - The test program is linked against the shared lib and using
>   symbols from the static lib.

Yep.

> The only reason the test is succesful is that the shared lib
> isn't using any symbols from the static lib.  If it were using
> symbols from the static lib, the shared lib would have (tried)
> including whatever is needed to have them, and then linking the
> shared lib would have failed on some platforms.

Yep.  I believe the idea was that on the platforms without pass_all,
the idea was not to link against the static lib, but to record the
dependency in the .la file, so that it would be picked up by later
program links (and libs which link against it).  Obviously the
-no-undefined flag then hurts.

To tell you the truth, I don't even know whether that works at all.
Most likely it doesn't.

> Using symbols from a library you're not linking to, and is only
> pulled in by the dependencies really isn't what you want to do
> either.

Well, some people have wished for this.  Regard the .la file as
portable record of dependencies.

No, I don't like this much either.

> I have no idea what the intention of this test really is that
> isn't covered by the other test, so I don't have a suggestion on
> how to fix it.

Did this help?

BTW, will you be maintaining the Debian Libtool package?

Cheers,
Ralf