(let's continue this as Bug#815712 instead) Steven Chamberlain wrote: > [...] I'm not sure how that could have regressed since > the -8 package... I wonder if it has to do with the patch that was > only meant to disable the FPU features on hurd and kfreebsd.
I'm fairly sure Samuel's patch didn't cause that regression on Linux, but someone who has a linux-i386 build environment could perhaps try building -9 without it; or even better, try a fresh build of -8 to see if the regression is rather in some build dependency. Possibly it is just due to the newer gcc-5 package used, maybe doing floating point operations in a different order and getting a slightly different result? | $ LD_LIBRARY_PATH=./Imath/.libs/:./Iex/.libs/:./Half/.libs/ gdb ImathTest/.libs/ImathTest | (gdb) run | [...] | Program received signal SIGABRT, Aborted. | 0x28433fc7 in __syscall_thr_kill () from /lib/i386-kfreebsd-gnu/i686/cmov/libc.so.0.1 | | ImathTest: testBoxAlgo.cpp:892: void {anonymous}::boxMatrixTransform(): Assertion `b21 == b2' failed. | (gdb) bt full | [...] | #5 0x0806e879 in (anonymous namespace)::boxMatrixTransform () at testBoxAlgo.cpp:892 | b21 = {min = {x = 11.6798019, y = 8.62185574, z = -24.5730953}, max = {x = 16.5198555, y = 12.7859335, z = -20.120142}} | b2 = {min = {x = 11.6798019, y = 8.62185574, z = -24.5730953}, max = {x = 16.5198536, y = 12.7859335, z = -20.120142}} | #6 0x08070dae in testBoxAlgo () at testBoxAlgo.cpp:1023 max.x only differs by 0.0000019 The attached patch fixes it, although I've found another testsuite failure after this. Regards, -- Steven Chamberlain ste...@pyro.eu.org
--- a/ImathTest/testBoxAlgo.cpp +++ b/ImathTest/testBoxAlgo.cpp @@ -889,7 +889,8 @@ assert (approximatelyEqual (b3.max, b4.max, e)); assert (approximatelyEqual (b3.max, b4.max, e)); - assert (b21 == b2); + assert (approximatelyEqual (b2.min, b21.min, e)); + assert (approximatelyEqual (b2.max, b21.max, e)); assert (b31 == b3); M[0][3] = 1;
signature.asc
Description: Digital signature