Updated JPL .class files to Java 8 as the latest OpenJDK no longer
handles version 7.
Several performance enhancements to arithmetic.
Fixed a memory leak in PL_put_term_from_chars() for numbers.
Enjoy — Jan
SWI-Prolog Changelog since version 9.1.15
PORT: Partial work to allow MSVC build under Conda. This patch works
around issues for CMake FindThreads.cmake. It also explicitly enables
MSVC /MD to compile the packages if threading is enabled.
BUILD: Runtime configuration during boot for Windows As, on Windows,
we install all DLLs in the same directory we do not need to add
the package bin dirs to the foreign search path. But, we must use
prolog_to_os_filename/2 to ensure proper handling of the search based
on the executable.
FIXED: #1197 function cmpr failing for negative values.
FIXED: #1097 ctrl-C in read causes exit from REPL Fix consists of
two parts. (1) if something (libedit) already set the stream error
condition, make S__fillbuf() return with an error immediately and
(2) do not abort the main query loop if we get an abort request.
DOC: additions and cleanup wrap_predicate/4, pack_install/1, handling
blobs and streams from foreign code, returning from foreign predicates.
ENHANCED: Allow interrupts during evalExpression()
PORT: Make new stuff work again using MSVC
ENHANCED: dark theme arithmetic used blue on black. [no ci]
PERFORMANCE: Add shortcut to PL_unify_number() Add a shortcut for
the typical case where the number is a small integer, the target is
a variable and we are not short on stack space.
PERFORMANCE: Simplify clearNumber() Notably improves performance for
float arithmetic.
FIXED: Memory leak in PL_put_term_from_chars() fast route. This
function implements a fast route for integers and rationals and leaked
the GMP/LibBF object for rationals and large integers.
ENHANCED: PL_get_float() to raise a float overflow on too large
rational.
BUILD: Added scripts/configure This script is used for a long time
to maintain multiple versions under the same source tree. See script
for details. [no ci]
PORT: Work around broken MacOS versions lacking <term.h> Reported
by Matthias Gondan after failing build for CRAN
FIXED: PL_get_nchars() and friends CVT_INTEGER handling Broken
in 2ac1553770ac1a9d6753b8b6e72ac2028815a028
ADDED: PL_get_nchars(): CVT_XINTEGER Allow converting numbers to
hexadecimal strings.
DOC: library(occurs) [no ci]
Package cpp
ENHANCEMENT: added PlAcquireStream
ENHANCED: AtomMap utility
DOC: blobs and exceptions
ENHANCED: std::string instead of char* in some calls (no code change
needed) - Can add char* interface if needed for performance (unlikely)
PlTerm(term_t) constructor is public, for consistency with PlAtom
Package jpl
UPDATED: Java standards from 7 to 8. Support for 7 is dropped from
latest OpenJDK releases and 7 is officially EOL for a while.
Package swipy
DOC: Handling Undefined
ADDED: Class Undefined to deal with WFS undefined results.
MODIFIED: Renamed status and truth_vals to truth As decided in
Janus meeting, September 20.
TEST: Install XSB test data if installation includes the tests
BUILD: Starting a pyproject.toml
DOC: Document handling virtual environments.
ADDED: Support Python virtual environments
ADDED: Pass instances of class Enum as name (atom).
FIXED: Avoid recursive errors crashing while janus.py is being loaded.
It seems we cannot call certain APIs while processing an error.
ADDED: Map Prolog rational numbers to Python fractions.Fraction().
ENHANCED: Use hexadecimal conversion for bigints.
DOC: Included references
COMPAT: Sync py_func/3,4 and py_dot/4,5. Updated docs.
MODIFIED: Sync XSB view with current agreement. - pyfunc/3-5,
pydot/4,5 renamed to py_func/3-5 and py_dot/4,5. - Deleted
free_python_object/1 (=py_free/1)
MODIFIED: Rules when py_object(true) is effective
MODIFIED: Use eval(Term) to evaluate Python expressions inline.
MODIFIED: Prolog representation of Python sets to use py_set(List)
Was pySet(List).
$ cmake ..
$ ninja
ninja: error: 'C:/msys64/ucrt64/lib/libpython3.dll.a', needed by 'src/janus.dll', missing and no known rule to make it
There is a /ucrt64/bin/libpython3.10.dll, as well as a /ucrt64/lib/libpython3.10.dll.a, which is correctly reported by message("${Python_LIBRARIES}") in the CMakeFile.
Workaround is cmake .. -DSWIPL_PACKAGES_PYTHON=OFF.
$ PATH=/x86_64-w64-mingw32.static.posix/bin:$PATH cmake .. -DSWIPL_PACKAGES_PYTHON=OFF -DSWIPL_PACKAGES_X=OFF -DINSTALL_DOCUMENTATION=OFF
$ PATH=/x86_64-w64-mingw32.static.posix/bin:$PATH make
...
[ 81%] Building C object packages/ssl/CMakeFiles/plugin_crypto4pl.dir/crypto4pl.c.obj
In file included from C:/Users/matthias/Documents/swipl-devel/packages/ssl/crypto4pl.c:55:
C:/Users/matthias/Documents/swipl-devel/packages/ssl/cryptolib.c:54:1: error: static declaration of 'EVP_MD_CTX_free' follows non-static declaration
54 | EVP_MD_CTX_free(EVP_MD_CTX *ctx)
Ahh, very vaguely, I remember this from months or even years ago. What was it…? I’ll search for a fix.
I think the problem is that on Windows we use the generic python3.dll, but we do so without testing it exists. It is part of the Python binary installation, but apparently not of the MSYS2 version. Added a check.
That sounds weird. Looks like it is picking up a wrong version of the headers? MINGW_ROOT points at a directory that has all the MinGW depedencies installed. That is supposed not to be a bin dir, but the dir above bin, lib and include.
actually does, but if it does the same as check_symbol_exists(...), then it compiles and links a little program with a pointer to, e.g., EVP_MD_CTX_free. And the linker needs zlib.h under MinGW, which is only given if MINGW_ROOT has been defined.
But it works for the wrong reason Seems the condition is part of the MSVC porting work where zlib.lib is used from vcpkg and, called zlib1.lib and apparently not required to link OpenSSL. I think we must make this unconditional again and use ${ZLIB_LIBRARIES}
Finding the right conditions for differences in the configuration is quite a challenge
edit: Pushed 86bf1a111390f7457e77f930fd5241999c2de472 to address this. Works for me using MinGW as well as MSVC. This code is only used on Windows, so other platforms are not affected.
When one installs swi-prolog via macports python’s latest release is also one of the dependencies, right? So one installs both swipl and python311 plus a bunch of other things. I was surprised to see that you don’t have to install it manually if you run “port install swi-prolog-devel”
PS: maybe the other thread with the announcement was more appropriate. This one is for arithmetics
Macports is fairly relaxed about big dependencies. As most Macports installations already have Python installed anyway, it is not that big. This is quit different from Debian that tries to keep packages small, such that you can configure a minimal system with exactly the stuff you need and as little as possible extras.
So, yes, Macports gives you a pretty much complete SWI-Prolog, possibly with some stuff you might not have wanted …
Thanx a lot for the explanation. Regarding stuff one doesn’t want…well, you know if you put yourself in someone else’s hands or judgements…you have to accept what’s included in the package so to say. Of course whether this is fair or not remains an open issue. At least for me. But I’m more interested in what comes next than in brooding about the past (although it is meaningful to do so). Best things
PS: maybe “reflecting” is better than “brooding” which at least to my ears seems to have a negative connotation