exact_arith: implement it

Promote on overflow to bigint/num, and not to NV.
This is a new lexical user-pragma to use exact
arithmetic without loosing precision on all builtin arithmetic ops.
As in perl6.

It is of course a bit slower than without.

Closes #21.

Author: Reini Urban

Configure

@@ -1424,6 +1424,7 @@ CONFIG=''
 usecperl=''
 fake_signatures=''
 usenamedanoncv=''
+useexactarith=''
 
 : Detect odd OSs
 define='define'
@@ -7584,6 +7585,32 @@ $define)
 	;;
 esac
 
+case "$useexactarith" in
+$define|true|[yY]*)	dflt='y' ;;
+*)			dflt='n' ;;
+esac
+cat <<EOM
+
+Would you like to build Perl so that every builtin arithmetic overflow
+does not promote to internal fast double/long double numbers but
+instead to slow and precise bigint? So that exact_arith is enabled by default.
+
+If this doesn't make any sense to you, just accept the default '$dflt'.
+EOM
+rp='Use exact arith, overflow to bigint as in perl6?'
+. ./myread
+case "$ans" in
+y|Y)	val="$define" ;;
+*)	val="$undef"  ;;
+esac
+set useexactarith
+eval $setvar
+case "$useexactarith" in
+$define)
+	echo "exact_arith selected." >&4
+	;;
+esac
+
 case "$usecperl" in
 $define) usecperl='define'
          echo "cperl variant selected." >&4
@@ -25913,6 +25940,7 @@ usecrosscompile='$usecrosscompile'
 usedevel='$usedevel'
 usedl='$usedl'
 usedtrace='$usedtrace'
+useexactarith='$useexactarith'
 usefaststdio='$usefaststdio'
 useffi='$useffi'
 useithreads='$useithreads'

MANIFEST

@@ -5686,6 +5686,8 @@ lib/dumpvar.pl			A variable dumper
 lib/dumpvar.t			A variable dumper tester
 lib/English.pm			Readable aliases for short variables
 lib/English.t			See if English works
+lib/exact_arith.pm		Pragma to set exact arithmetic as in perl6
+lib/exact_arith.t		See if use exact_arith works
 lib/ExtUtils/Embed.pm		Utilities for embedding Perl in C programs
 lib/ExtUtils/t/Embed.t		See if ExtUtils::Embed and embedding works
 lib/ExtUtils/typemap		Extension interface types

Porting/Maintainers.pl

@@ -1732,6 +1732,7 @@ package Maintainers;
                 lib/DBM_Filter/
                 lib/DirHandle.{pm,t}
                 lib/English.{pm,t}
+                lib/exact_arith.{pm,t}
                 lib/ExtUtils/Embed.pm
                 lib/ExtUtils/XSSymSet.pm
                 lib/ExtUtils/t/Embed.t

config_h.SH

@@ -5488,6 +5488,14 @@ sed <<!GROK!THIS! >$CONFIG_H -e 's!^#undef\(.*/\)\*!/\*#define\1 \*!' -e 's!^#un
 #$usesafehashiter	USE_SAFE_HASHITER		/**/
 #endif
 
+/* USE_EXACT_ARITH:
+ *	This symbol, if defined, indicates that Perl uses exact_arith as default.
+ */
+#define PERL_EXACT_ARITH
+#ifndef USE_EXACT_ARITH
+#$useexactarith	USE_EXACT_ARITH		/**/
+#endif
+
 /* PERL_HASH_FUNC_*:
  *	This symbol defines the used perl hash function variant.
  *      It is set in Configure or via -Dhash_func=, but can be left blank.

embed.fnc

@@ -528,6 +528,7 @@ pR	|int	|PerlSock_accept_cloexec|int listenfd \
 pR	|int	|PerlSock_socketpair_cloexec|int domain|int type|int protocol \
 				|NN int *pairfd
 #endif
+AMp	|void	|bigint_arith	|NN const char *op|NN SV *const left|NN SV *const right
 #if defined(PERL_IN_DOIO_C)
 s	|IO *	|openn_setup    |NN GV *gv|NN char *mode|NN PerlIO **saveifp \
 				|NN PerlIO **saveofp|NN int *savefd \

embed.h

@@ -67,6 +67,7 @@
 #define av_top_index(a)		S_av_top_index(aTHX_ a)
 #define av_undef(a)		Perl_av_undef(aTHX_ a)
 #define av_unshift(a,b)		Perl_av_unshift(aTHX_ a,b)
+#define bigint_arith(a,b,c)	Perl_bigint_arith(aTHX_ a,b,c)
 #define block_end(a,b)		Perl_block_end(aTHX_ a,b)
 #define block_gimme()		Perl_block_gimme(aTHX)
 #define block_start(a)		Perl_block_start(aTHX_ a)

ext/Config/Config_xs.in

@@ -1496,6 +1496,7 @@ usecrosscompile,	T_INV,0,ALN64I"@@usecrosscompile@@"
 usedevel,		T_INV,0,ALN64I"@@usedevel@@"
 usedl,			T_INV,0,ALN64I"@@usedl@@"
 usedtrace,		T_INV,0,ALN64I"@@usedtrace@@"
+useexactarith,		T_INV,0,ALN64I"@@useexactarith@@"
 usefaststdio,		T_INV,0,ALN64I"@@usefaststdio@@"
 useffi,			T_INV,0,ALN64I"@@useffi@@"
 useithreads,		T_INV,0,ALN64I"@@useithreads@@"

lib/exact_arith.pm

@@ -0,0 +1,30 @@
+package exact_arith;
+our $VERSION = '0.01';
+sub unimport { delete $^H{exact_arith}; }
+sub import { $^H{exact_arith} = 1; }
+
+1;
+__END__
+
+=head1 NAME
+
+exact_arith - promote on overflow to bigint/num
+
+=head1 SYNOPSIS
+
+    use exact_arith;
+    print 18446744073709551614 * 2; # => 36893488147419103228, a bigint object
+
+    { no exact_arith;
+      print 18446744073709551614 * 2; # => 3.68934881474191e+19
+    }
+
+=head1 DESCRIPTION
+
+This is a new lexical user-pragma since cperl5.24 to use exact
+arithmetic, without loosing precision on all builtin arithmetic ops.
+As in perl6.
+
+It is of course a bit slower, than without.
+
+=cut

lib/exact_arith.t

@@ -0,0 +1,24 @@
+#!./perl -- -*- mode: cperl; cperl-indent-level: 4 -*-
+
+BEGIN {
+    chdir 't' if -d 't';
+    @INC = ( '.', '../lib' );
+}
+
+use strict;
+require '../t/test.pl';
+plan(4);
+
+$|=1;
+
+use exact_arith;
+my $n = 18446744073709551614 * 2; # => 36893488147419103228, a bigint object
+is(ref $n, 'bigint');
+is($n, 36893488147419103228);
+
+{
+    no exact_arith;
+    my $m = 18446744073709551614 * 2;
+    is(ref $n, '');
+    is($n, 3.68934881474191e+19);
+}

perl.c

@@ -1932,6 +1932,9 @@ S_Internals_V(pTHX_ CV *cv)
 #  ifdef PERL_DONT_CREATE_GVSV
 			     " PERL_DONT_CREATE_GVSV"
 #  endif
+#  ifdef PERL_EXACT_ARITH
+			     " PERL_EXACT_ARITH"
+#  endif
 #  ifdef PERL_EXTERNAL_GLOB
 			     " PERL_EXTERNAL_GLOB"
 #  endif
@@ -2034,6 +2037,9 @@ S_Internals_V(pTHX_ CV *cv)
 #  ifdef USE_CPERL
 			     " USE_CPERL"
 #  endif
+#  ifdef USE_EXACT_ARITH
+			     " USE_EXACT_ARITH"
+#  endif
 #  ifdef USE_FAST_STDIO
 			     " USE_FAST_STDIO"
 #  endif

perl.h

@@ -317,9 +317,11 @@
     RX_ENGINE(rx_sv)->dupe(aTHX_ (rx_sv),(param))
 #endif
 
-
-
-
+#ifdef PERL_EXACT_ARITH
+#define IS_EXACT_ARITH   cop_hints_fetch_pvs(PL_curcop, "exact_arith", REFCOUNTED_HE_EXISTS)
+#else
+#define IS_EXACT_ARITH   0
+#endif
 
 /*
  * Because of backward compatibility reasons the PERL_UNUSED_DECL

pod/perlcdelta.pod

@@ -547,6 +547,12 @@ release manager will have to investigate the situation carefully.)
 
 =over 4
 
+=item L<exact_arith> 0.01
+
+Promote on overflow to bigint/bignum as in perl6, do not loose precision
+with all builtin arithmetic operators.
+L<[cperl #21]|https://github.com/perl11/cperl/issues/21>
+
 =item L<ffi> 0.01c
 
 ffi helpers and ffi types.