[PATCH 1/5] Only call dlerror() when dlsym() fails.

[Thomas Klausner]

On Wed, Nov 07, 2012 at 03:55:30PM +0100, Werner Koch wrote:
> On Wed,  7 Nov 2012 13:34, tk at giga.or.at said:
> 
> > a few times, and I don't see it defining what dlerror() returns if it
> > was never called before and no error by dynamic linking occurred yet.
> 
> Right, the proper way to use it would be call dlerror() before calling
> dlsym.  This makes sure that we see the right error.  However, this
> won't help with NetBSD's wrong implementation.

I think I don't understand what you see as the problem in the NetBSD
implementation. According to its man page, it should behave the same
as specified above. What behaves differently?

> > So since POSIX doesn't specify that dlerror() on NetBSD is behaving
> > incorrectly and the patch is trivial, I ask you to use the patch
> 
> Recall that we also use dlopen et al. as interface to the respective
> Windows functions.  Thus this change is not trivial.

I'll take your word on that.

> > instead of adding quite a bit of overhead in configury.
> 
> Agreed.
> 
> However, a better way to fix this is to get rid of dlopen.  There is no
> more need for idea-stub.c and we may add include IDEA support directly.
> (That algorithm will probably never die - like Arcfour)

That solution is fine with me, of course.

Thanks,
 Thomas