Re: RFS: mia/2.4.7-14 [RC]

Andreas Tille <[email protected]> writes:

> You are kindly invited to revert my admittedly drastic means.  I simply
> was weighthing my personal time I could either spent for some quite
> rarely used software which is most probably not used on 32bit any more
> against fixing a heap of RC bugs in packages that are actually used.

FTR, I took a quick stab at conditionally building with -ffloat-store,
without which some floating-point calculations can have excess internal
precision on i386.  The results were a mixed bag.  As I'd hoped, it fixed
the 2dimage-cost-lsd discrepancy; however, it unsurprisingly did nothing
for 3d-vectorfield's NULL pointer dereference, and moreover broke
3d-rigidregister:

  189: ./mia/3d/test_rigidregister.cc(105): error: in "test_rigidreg_translate_gd": difference{0.00876236} between params[i]{1.0087623596191406} and orig_params[i]{1} exceeds 0.8%

As such, a proper fix won't be so easy; I don't have time to come up
with one either. :-/