On Feb 5, 2008, at 2:34 AM, Ralf S. Engelschall wrote:

> On Mon, Feb 04, 2008, Jeff Johnson wrote:
>
>> There's a failure mode where the compilation test for
>> popt fails because popt is mandatory, yet the test failed
>> because of an incompletely specified neon -> gnutls linkage.
>>
>> Does it make sense to test for mandatory (and mostly
>> standalone) libraries like libpopt earlier in configure.ac?
>>
>> Here's what I would do (but I'll leave the AutoFu to Ralf) below.
>>
>> There are perhaps a few other reorganizations, like keyutils, that
>> perhaps should be done earlier as well.
>
> The reordering for POPT should be fine. I see no side-effects.
> Just commit it, please. But I do not understand why this fixes any
> NEON->GNUTLS issues. Neither NEON nor GNUTLS use POPT AFAIK...
>

POPT immediately after neon detection just happened to be where
the problem was detected, leading to a misleading error message
./configure full stop because POPT is mandatory.

The real fix needs to be in NEON detection, or by adding
a GNUTLS prerequisite test.

The problem was triggered by F9 switching to gnutls
from openssl under neon. There have been some weird hacks
in F9 neon packaging to handle no *.la files policiies, and
gssapi dependencies (from memory, still diagnosing).

BTW, all this started because the autoconf test for uuid_import failed
against F9 uuid-1.6.0 packaging.

73 de Jeff