perf@efd.lth.se (03/31/89)
Environment: Sun 3/280, 3/50 SunOS 4.0.1 g++ 1.34.1 libg++ 1.34.0 Test 8,10 and 23 fails with the following output. (Test23 works ok in a window with 24 rows.) Compiling without optimization, as someone suggested in gnu.g++.lib.bug, doesn't change anything. ------- test8 ------------------------------------------------- BitSet tests: a = 0* b = 0000000000000000000000000010* Failed assertion b[10] == 1 at line 65 of `test8.cc'. IOT trap (core dumped) --------------------------------------------------------------- ------- test10 ------------------------------------------------ five random ACG integers: 1525072166 1954057046 3406008937 226879594 4107049426 five random MLCG integers: 1341853672 83475514 936613571 888739672 2097844081 Binomial r1( 100, 0.50, &gen1) ... five samples: Failed assertion result.d < 1.0 && result.d >= 0 at line 77 of `../g++-include/RNG.h'. IOT trap (core dumped) --------------------------------------------------------------- ------- test23 ------------------------------------------------ CursesWindow Error: Cannot construct subwindow IOT trap (core dumped) --------------------------------------------------------------- ____________________________________________________________________________ | Per Foreby | Email: perf@efd.lth.se | | System manager at EFD | Snail: Tekniska Hogskolan i Lund | | Lund Institute of Technology | Box 118, S-221 00 LUND, Sweden | | Sweden | Phone: int + 46 46-10 75 98 | |________________________________|___________________________________________|
dl@ROCKY.OSWEGO.EDU (Doug Lea) (03/31/89)
About test8:
There was a typo in a preliminary version of libg++-1.34.0
src/BitSet.cc:longtoBitSet that caused it to fail on `big-endian'
machines, and that you might have if you ftp'ed it from rocky. The
correct version is:
BitSetTmp longtoBitSet(unsigned long w)
{
unsigned short u[2];
u[0] = w & ((unsigned short)(~(0)));
u[1] = w >> BITSETBITS;
BitSetRep* r = BitSetalloc(0, &u[0], 2, 0, 3);
trim(r);
return r;
}
If this is not your problem, try compiling *without* the -fsave-memoized
option, which is buggy in g++-1.34.1
About test10:
As was previously posted, you must, on a sun3, compile src/RNG.cc with
the -ffloat-store option.
About test23:
No one has been able to tell me how/why this sometimes fails on Sun3s
and Sun4s. The test program does require running on a screen of at
least 24X80, but it seems that it sometimes fails in calling
libcurses:subwin even in such cases, but never when run from a remote
terminal. Any further diagnosis (e.g., a gdb run after compiling your
local libcurses library -g) would be very much apreciated.
-Doug