10.19.2 Bugs Fixed
- The Windows installer would not install the Emacs setup file.
- Under Windows and x86 Linux, a number of problems related to floating point
processing
caused by non-standard C compiler floating point optimizations.
Symptoms include memory access errors and incorrect printing of some
numbers, e.g.
3.1E11
.
- Ill-formatted output in response to the debugger's g (print
ancestor goals) command.
-
Memory allocation bug in the Visual Basic interface.
-
Due to a change in Linux kernels newer than 2.4.9 it was not possible to
start more than one SICStus run-time in the same process.
Linux distributions affected include Red Hat 7.3 and later (but not
Red Hat 7.2).
- Under UNIX, a blocking
socket:socket_accept/[2,3]
is now
interruptible with ^C (`SIGINT').
Other signal handlers installed with SP_signal()
will also run
immediately if the signal arrives during a blocking socket accept
operation.
-
Work around a problem with the system call
inet_ntoa
that affected several predicates in
library(sockets)
. The problem affected IRIX and on 64-bit AIX.
-
format/[2,3]
could intercept resource errors and report them as consistency errors;
the “print float” formats were poorly documented.
save_program/[1,2]
had a memory leak, which could cause crashes.
library(clpfd)
:
-
Arithmetic constraints could cause infinite loops.
-
in/2
and in_set/2
could silently fail or succeed when they
should raise a type or representation error.
fdbg_start_labeling/1
was missing.
-
Bad Prolog term could arise from very large domains.
library(chr)
:
-
Attempted removal of attribute or constraint that is already gone
would cause failure.
- Jasper: The implementation of
newObjectTerm(Object obj)
in
multi-threaded Jasper caused a NullPointerException: Creating or
accessing SPTerm in wrong thread. (IllegalCallerException in disguise.)
- Jasper: Exceptions are now propagated properly from Prolog to Java.
This has never worked as intended before.
-
The Gauge profiling tool was broken.
library(fastrw)
: fast_read/2
after peek_byte/2
and
friends on the same stream did not work.
-
Mac OS X
load_foreign_resource/1
would report run-time linker
errors incorrectly.