Re: GnuTLS regression test failing on OpenBSD/sparc64

[Simon Josefsson]

Landry Breuil <address@hidden> writes:

> On Wed, Oct 22, 2008 at 05:04:48PM +0200, Simon Josefsson wrote:
>> Landry Breuil <address@hidden> writes:
>> 
>> > Hi,
>> >
>> > after working on an update to latest libgcrypt, i'm in the process of
>> > updating GnuTLS in OpenBSD's ports-tree. It's currently building fine on
>> > sparc64, but i got an error in regression suite.
>> 
>> Hi.  Thanks for working on this.
>> 
>> > Bus error (core dumped)
>> > FAIL: rsa-md5-collision
>> > ===================================
>> > 1 of 1 tests failed
>> > Please report to address@hidden
>> > ===================================
>> >
>> > certtool is crashing badly due to symbol conflicts:
>> > WARNING: symbol(dig) size mismatch, relink your program
>> > /usr/obj/ports/gnutls-2.6.0/gnutls-2.6.0/src/.libs/certtool : WARNING:
>> > symbol(dig) size mismatch, relink your program
>> >
>> > The attached patch fixes the issue by avoiding name conflicts between
>> > two different variables of the same name and different types.
>> 
>> This has been fixed on the trunk in a better way, see:
>> 
>> http://git.savannah.gnu.org/gitweb/?p=gnutls.git;a=commitdiff;h=8080538ff60efe605727401bf37f270112703f46
>
> Hm, sorry but it seems not fixed for me.. i cherry picked you patch and
> still get the failing test here on amd64:
>
> /usr/obj/ports/gnutls-2.6.0/gnutls-2.6.0/src/.libs/certtool : WARNING:
> symbol(dig) size mismatch, relink your program
> Segmentation fault (core dumped) 
> FAIL: rsa-md5-collision

Did you recompile fipsmd5.c and re-link libgnutls-extra?

>> > I also noticed that two regress tests were failing to build when
>> > using included libtasn1, there's a missing -I$(top_srcdir)/lib/minitasn1/
>> > in tests/pkcs12-decode/Makefile.am and tests/Makefile.am
>> 
>> Also already fixed:
>> 
>> http://git.savannah.gnu.org/gitweb/?p=gnutls.git;a=commitdiff;h=3cbc220926c1490cbd726e63bb30efcd542c383e
>> http://git.savannah.gnu.org/gitweb/?p=gnutls.git;a=commitdiff;h=fbdc6734027077b1f3bf9864f3f3efc265cd91d0
>
> Cool, thanks. Anyway i'm now planning to use the updated libtasn1
> version we have in ports-tree.

That's the preferred approach, yes.

/Simon