t-constant fail i686-darwin10

Jack Howarth howarth at bromo.med.uc.edu
Thu Jan 14 02:12:31 CET 2010


On Wed, Jan 13, 2010 at 11:35:57AM +0100, Torbjorn Granlund wrote:
> David Fang <fang at csl.cornell.edu> writes:
> 
>   > This looks like non-clean builds.
>   > Please make sure to start with a clean directory (or at least do a 'make
>   > distclean' between builds).
>   
>   Torbjorn,
>   	This was indeed a fresh, clean build.
>   
> OK, then I am at loss explaining this.
> 
> Apple changes every OS version or Xcode version to introduce new quirks
> that break builds.  I would need 500 Mac OS test machines of all
> possible combinations to keep up,..
> 
> If you want GMP to work on your system, then I am afraid that you need
> to "port" it, or wait for somebody to port it.  I change that makes it
> work on your system but breaks builds on all other systems might give me
> some hint, but then I need to figre out for which OS/Xcode combinations
> your change is needed, and for which OS/Xcode combinations is breaks
> things.
> 
> -- 
> Torbjörn
> _______________________________________________
> gmp-bugs mailing list
> gmp-bugs at gmplib.org
> https://gmplib.org/mailman/listinfo/gmp-bugs

Torbjörn,
    I also see this failure on 32-bit intel darwin9. The failing section of
t-constants.c appears to be the line...

CHECK_LIMB (PP_INVERTED, pp_inverted_calc);

since commenting out that line eliminates the t-constants testsuite failure.
When it is uncommented, the output from t-constants contains the line...

PP_INVERTED == 53e5645c, but pp_inverted_calc == 538ce830

Is that expected? Also, once the failure of the t-constants testcase
is suppressed, a couple new failures are exposed on 32-bit intel darwin9...

ERROR in test 0, n = 12
8001FFFFFFFFFF0000000000000000000003FFFFFFFFFC0000000000000000000FFFFFFFF003FFFC0000000000000000
F30C95F59FAC8AF7E1520C5725B8097277DFDC4B4B39179B68AB2B3CBCD13F1A5B7D61347F36A0E75528470A76B7D861
/bin/sh: line 1: 66469 Abort trap              ${dir}$tst
FAIL: t-invert
*******************************************************************************
mpn_dcpi1_div_qr failed test 1: q too small
N=    000fffff ffffffff ffffffff ffffffff ... 00000000 00000000 00000000 00000000
D=    80000fff ffffffff ffffffff ffffffff ... ffffffff ffffffff ffffffff ffffffff
Q=    00000000 001e625c 311825ba ee9896aa ... 19adfa2f 6bbefc7b d1e35026 c3d46ed0
R=    4e8426b7 5ac8bdd2 a3b0f79a a885c8cd ... 19adfa2f 6bbefc7b d1e35026 c3d46ed0
N-Q*D=00000000 0000ced0 014e2a11 065805c1 ... 19adfa2f 6bbefc7b d1e35026 c3d46ed0
nn = 5045, dn = 939, qn = 4107
/bin/sh: line 1: 66490 Abort trap              ${dir}$tst
FAIL: t-div

I assume these testcases are decoupled and don't rely on each other to function
properly.
          Jack
ps I have found that on intel darwin9, I have use the Apple gcc-4.2 compiler
to avoid linkage issues (which I suspect are due to the use of __gnu_inline__
which the older gcc-4.0 doesn't understand). Otherwise, the linkage of
libgmp.3.dylib fails with the error...

ld: duplicate symbol ___gmpz_abs in .libs/compat.o and .libs/assert.o

Which MacOS X and Xcode release did you test gmp 5.0 on?


More information about the gmp-bugs mailing list