swi-prolog 8.4.1 wouldn't compile (amd64/Centos7/gcc-10.2)

NevemTeve · 12-08-2021, 08:59 AM

Oh well, why would it compile?
Here is the build script I tried:

Code:

#!/bin/bash

rm -rf build/

mkdir -p man/archive

mkdir -p build; cd build

set -e
set -o pipefail

export CFLAGS="$COMMON_CFLAGS64 -std=c17 $COMMON_CPPFLAGS"
export LDFLAGS="$COMMON_LDFLAGS64 -L/usr/local/lib64 -Wl,-R/usr/local/lib64"

cmake -DCMAKE_INSTALL_PREFIX=/usr/local .. 2>&1 | tee ../log.cmake

make VERBOSE=yes 2>&1 | tee ../log.make.all

ctest -j 4 2>&1 | tee ../log.ctest

make VERBOSE=yes install 2>&1 | tee ../log.make.install

NevemTeve · 12-08-2021, 09:02 AM

And here is the error-message:

Code:

make[2]: Entering directory `/usr/local/src/swipl-8.4.1/build'
[ 56%] Building C object src/CMakeFiles/swiplobjs.dir/pl-atom.c.o
cd /usr/local/src/swipl-8.4.1/build/src && /usr/local/bin/cc  -I/usr/local/src/swipl-8.4.1/build/src -I/usr/local/src/swipl-8.4.1/src -m
64 -std=c99 -std=c17 -I/usr/local/include -D__UNIX__ -D__unix__ -D_GNU_SOURCE -D_XOPEN_SOURCE=700 -D_ALL_SOURCE -D_LARGEFILE_SOURCE -D_L
ARGEFILE64_SOURCE -D_FILE_OFFSET_BITS=64 -D_LARGE_FILES -D_THREAD_SAFE -O2 -fPIC -fvisibility=hidden -Wall -fno-strict-aliasing -o CMake
Files/swiplobjs.dir/pl-atom.c.o -c /usr/local/src/swipl-8.4.1/src/pl-atom.c
In file included from /usr/local/src/swipl-8.4.1/src/pl-incl.h:135,
                 from /usr/local/src/swipl-8.4.1/src/pl-atom.h:36,
                 from /usr/local/src/swipl-8.4.1/src/pl-atom.c:38:
/usr/local/src/swipl-8.4.1/src/pl-builtin.h:295:5: error: expected declaration specifiers or '...' before '(' token
  295 |     ( sizeof(*__PL_ld) > 1 ? (PL_local_data_t *)(__PL_ld) \
      |     ^
/usr/local/src/swipl-8.4.1/src/pl-macros.h:98:24: note: in definition of macro '_M_IF0b'
   98 | #define _M_IF0b(...)   __VA_ARGS__
      |                        ^~~~~~~~~~~
/usr/local/src/swipl-8.4.1/src/pl-macros.h:63:26: note: in expansion of macro 'A_ECHO'
   63 | #define A_UNWRAP(list)   A_ECHO list

NevemTeve · 12-08-2021, 09:05 AM

pl-builtin.h:

Code:

    291 /* These will never be compiled, they are here simply for scope detection */
    292 extern char *__PL_ld;
    293 extern struct {char *engine;} *PL__ctx;
    294 #define _LD_WITH_FALLBACK(f) \
    295                   ( sizeof(*__PL_ld) > 1 ? (PL_local_data_t *)(__PL_ld) \
    296                   : sizeof(*PL__ctx->engine) > 1 ? (PL_local_data_t *)(PL__ctx->engine) \
    297                   : (PL_local_data_t *)(f) )
    298 #define LOCAL_LD  _LD_WITH_FALLBACK(_ASSERT_LD(sizeof(*__PL_ld) > 1 || sizeof(*PL__ctx->engine) > 1))
    299 #define ANY_LD    _LD_WITH_FALLBACK(GLOBAL_LD)

shruggy · 12-08-2021, 09:10 AM

Before compiling from source, I'd try rebuilding Fedora Rawhide source RPM package.

This would require some adjustments to the spec file. E.g. readline-devel in CentOS 7 doesn't provide a pkg-config .pc file, so you'll have to BR readline-devel rather than pkgconfig(readline). For some builds, I had to redefine %__cmake_configure as well.

NevemTeve · 12-08-2021, 09:14 AM

Calling only cpp doesn't help much either:

Code:

In file included from /usr/local/src/swipl-8.4.1/src/pl-incl.h:576,
                 from /usr/local/src/swipl-8.4.1/src/pl-atom.h:36,
                 from /usr/local/src/swipl-8.4.1/src/pl-atom.c:38:
/usr/local/src/swipl-8.4.1/src/pl-vmi.h:86:1: error: macro "_M_TEMPLATE1" requires 4 arguments, but only 2 given
   86 |   ),
      | ^ ~~

vmi-metadata.h:

Code:

      7     Note: this file is generated by ./mkvmi from pl-vmi.c.  DO NOT EDIT....

     15 #define FOREACH_VMI_CALL(sep,f,...) f(D_BREAK, ## __VA_ARGS__) sep() <another 7915 bytes>

pl-vmi.h:

Code:

     71 #define FOREACH_VMI(sep,...) FOREACH_VMI_CALL(sep, M_TEMPLATE, __VA_ARGS__)

     83 typedef enum
     84 { FOREACH_VMI(T_COMMA,
     85     ,VMI_NAME,
     86   ),
     87   VMI_END_LIST
     88 } vmi;

pl-macros.h:

Code:

     62 #define A_ECHO(...)              __VA_ARGS__

    105 #define M_TEMPLATE(ctxarg, ...)  A_ECHO(_M_TEMPLATE0(ctxarg, __VA_ARGS__))
    106 #define _M_TEMPLATE0(ctx, lit, f, ...)  lit _M_TPLEXPAND(ctx, f, 1, ## __VA_ARGS__) (ctx, ## __VA_ARGS__)

    118 #define _M_TPLEXPAND(ctx, f, n, ...)  M_IFPAREN(f) (A_UNWRAP(f)) (_M_TPLCALL(f)(ctx)) M_IFEMPTY(__VA_ARGS__) (A_IGNORE) (_M_TPLNEXT(ctx, n, ## __VA_ARGS__))

shruggy · 12-08-2021, 01:57 PM

I've built it and uploaded here. If you want to rebuild it from source, you'll additionally need

in order to build pl-doc: texlive-a4wide and texlive-tabulary (can be taken from Fedora 19)
in order to build pl-odbc: unixODBC-devel (can be installed from Microsoft Prod repo for RHEL 7, NB: not from their repo for CentOS 7)
default GCC 4.8.5 in CentOS 7 is too old, I used devtoolset-8 from the SCLO RH repo

ntubski · 12-08-2021, 03:28 PM

Quote:

Originally Posted by NevemTeve

Calling only cpp doesn't help much either:
vmi-metadata.h:

Code:

      7     Note: this file is generated by ./mkvmi from pl-vmi.c.  DO NOT EDIT....

     15 #define FOREACH_VMI_CALL(sep,f,...) f(D_BREAK, ## __VA_ARGS__) sep() <another 7915 bytes>

## __VA_ARGS__ is a GNU specific extension and you passed -std=c17. Maybe -std=gnu17 or similar would work?

https://gcc.gnu.org/onlinedocs/cpp/Variadic-Macros.html

NevemTeve · 12-09-2021, 12:15 AM

Thank you guys for your help. Still, I'm about going down to version 7.6.4 -- it compiles out of the box, also it doesn't use cmake. [Don't get me wrong, I know cmake is a great software. Or perhaps `large` is the right term.]

Edit: -std=gnu17 did help though, the compilation stopped elsewhere.

shruggy · 12-09-2021, 05:23 AM

Does the source RPM I linked to above not compile on your system?

NevemTeve · 12-09-2021, 06:33 AM

I'll admit, I didn't try it

I prefer to compile programs for source without distribution modifications... Also I need SWI-prolog only for one purpose, which is running this: https://github.com/fnogatz/xsd2json -- and version 7.6.4 is perfectly able to run it.

shruggy · 12-09-2021, 06:46 AM

Quote:

Originally Posted by NevemTeve

I prefer to compile programs for source without distribution modifications...

Interesting. My preferences are the opposite. I create a DEB / an RPM whenever I can, so that all installed software could be managed with distribution package management tools.