Alpha computers have problems with floats.
Most of these are due to precision loss in encode_value().
Presumably due to the "Reducing float" code, which always truncates the mantissa to 32 bits. Is this code really needed?
/ Marcus Comstedt (ACROSS) (Hail Ilpalazzo!)
Previous text:
2004-03-20 14:10: Subject: 7.5/7.7
As always there are no absolutes. A run through of the failing systems on the pikefarm gives:
Alpha computers have problems with floats.
Most of these are due to precision loss in encode_value().
dmalloc currently finds a leak.
This should be fixad ASAP.
Many non-gcc compilers fails. This is really not a priority IMHO.
I'm of the opposite opinion; there are already too many projects that only support compilation with gcc.
Solaris <9 fails the grubba process trace test.
This is not a big issue; the test can be disabled for the release.
- sasserver2.emsphone.com, AIX 5.2
dumpes core when building post modules.
This is probably a machine-code generation problem.
- earl-grey, Power Mac, Darwin 6.8
Problem with recognizing Gmp.mpz objects. Problem with Stdio.File->lock and ->trylock.
OS bug. Can be worked around by creating the lockfile on a local filesystem (eg /tmp).
- bimac, Power Mac, Darwin 7.3
Crash in testsuite.
Dies of SIGBUS in the resolver test. No core.
- E-T.local, Power Mac, Darwin 7.2.0
Error in process-trace test.
- dan.emsphone.com, i386, FreeBSD 5.2
Error in grubbas process-trace test. Crash in the Nettle module. (looks to me like strlen("$1$") crashes)
Probably due to using a crypt_md5 from some other place than crypt_md5.c.
- alpha.home.oav.net, alpha, FreeBSD 5.2
Inf/Nan appears to be broken.
Probably broken float support in the OS.
A Process test fails ("Probable operating system bug.")
As the message says.
Error in the grubba process trace test. Crash in the Nettle module. (again looks like strlen to me)
As above.
- fonda.roxen.com, HP-UX
Unresolved symbol int64_from_bignum in Shuffler
Probably due to int64_from_bignum being a variable, and Shuffler.cmod not including bignum.h.
Unresolved symbol __eprintf in Image.Tiff
Probably similar error, but in libtiff.a.
Poll related core dump.
The system reaper thread is in poll(2), yes. Looks like it hung in the process trace test.
- cbs-server.dmz.cbc, i686, Linux 2.4
Thread-related crash.
Looks like it hung in the process trace test.
- dbsrv1.emsphone.com, alpha, OSF1
Float problem (I guess).
encode_value() problem.
Crash in what appears to be in type generation code.
Crash when initializing the Unicode module.
Looks like pike_type_hash contains broken pointers. Might be due to some name collision in the Unicode module (cf FreeBSD above).
- kali.roxen.com, i686, Windows NT
Nonstaitc fails to build the image module. Kludge by adding the lib objs to the module makefile?
Probably a pntld-related problem.
It is enough for me for a release if Solaris 8 and Windows gets green.
/ Henrik Grubbström (Lysator)