On Sat, 18 Jun 2005 18:51:40 +0100, Nicholas Clark <[EMAIL PROTECTED]> wrote:
> Is this a suitable patch to apply to Configure (and backport to the
> metaconfig units), for the purpose of testing whether 0.0 is stored as all
> bits zero in memory? It's a useful thing to know, because if it's true
> (which I think is generally the case), then memset of a block of memory to
> 0 is sufficient to correctly initialise any NVs that it contains. It's
> quite safe for it to default to #undef on non-Configure platforms.
Looks complete and correct. I wonder if the debug messages in the test
program should go to stderr instead.
Feel free to commit. I have to backport to a new meta-unit and check deps,
and I don't know how long it would take.
IF you commit, don't hesitate to also USE the define somewhere in the core
code, since I must otherwise (ab)use handy.h again to make the new unit to be
included in Configure, since metaconfig will skit the unit if the things it
defines are not used anywhere
> In theory we should also be checking that NULL pointers are stored as all
> bits zero, but I think that that portability battle is already lost - we
> already Zero() SV bodies that contain pointers and expect things to work.
>
> Also, I think that by adding another symbol to config_h.SH this breaks VMS,
> but I can't remember the correct fix, or where it goes. Can someone help?
>
> Nicholas Clark
>
>
> --- Configure.orig Tue May 31 07:09:02 2005
> +++ Configure Sat Jun 18 17:55:48 2005
> @@ -1002,6 +1002,7 @@ perl5=''
> perladmin=''
> perlpath=''
> d_nv_preserves_uv=''
> +d_nv_zero_is_allbits_zero=''
> i16size=''
> i16type=''
> i32size=''
> @@ -14833,6 +14834,128 @@ esac
>
> $rm -f try.* try
>
> +$echo "Checking whether NV 0.0 is all bits zero in memory..." >&4
> +: volatile so that the compiler has to store it out to memory.
> +if test X"$d_volatile" = X"$define"; then
> + volatile=volatile
> +fi
> +$cat <<EOP >try.c
> +#include <stdio.h>
> +#$i_stdlib I_STDLIB
> +#ifdef I_STDLIB
> +#include <stdlib.h>
> +#endif
> +#$i_string I_STRING
> +#ifdef I_STRING
> +# include <string.h>
> +#else
> +# include <strings.h>
> +#endif
> +#include <sys/types.h>
> +#include <signal.h>
> +#ifdef SIGFPE
> +$volatile int bletched = 0;
> +$signal_t blech(s) int s; { bletched = 1; }
> +#endif
> +
> +int checkit($nvtype d, char *where) {
> + unsigned char *p = (char *)&d;
> + unsigned char *end = p + sizeof(d);
> + int fail = 0;
> +
> + while (p < end)
> + fail += *p++;
> +
> + if (!fail)
> + return 0;
> +
> + p = (char *)&d;
> + printf("No - %s: 0x", where);
> + while (p < end)
> + printf ("%02X", *p++);
> + printf("\n");
> + return 1;
> +}
> +
> +int main(int argc, char **argv) {
> + $nvtype d = 0.0;
> + int fail = 0;
> + fail += checkit(d, "0.0");
> +
> + /* The compiler shouldn't be assuming that bletched is 0 */
> + d = bletched;
> +
> + fail += checkit(d, "bleched");
> +
> +#ifdef SIGFPE
> + signal(SIGFPE, blech);
> +#endif
> +
> + /* Paranoia - the compiler should have no way of knowing that ANSI says
> + that argv[argc] will always be NULL. Actually, if it did assume
> this it
> + would be buggy, as this is C and main() can be called from
> elsewhere in
> + the program. */
> + d = argv[argc] ? 1 : 0;
> +
> + if (d) {
> + printf("Odd argv[argc]=%p, d=%g\n", argv[argc], d);
> + }
> +
> + fail += checkit(d, "ternary");
> +
> + memset(&d, sizeof(d), argv[argc] ? 1 : 0);
> +
> + if (d != 0.0) {
> + printf("No - memset doesn't give 0.0\n");
> + /* This might just blow up: */
> + printf("(gives %g)\n", d);
> + return 1;
> + }
> +
> +#ifdef SIGFPE
> + if (bletched) {
> + printf("No - something bleched\n");
> + return 1;
> + }
> +#endif
> + if (fail) {
> + printf("No - %d fail(s)\n", fail);
> + return 1;
> + }
> + printf("Yes\n");
> + return 0;
> +}
> +EOP
> +set try
> +
> +d_nv_zero_is_allbits_zero="$undef"
> +if eval $compile; then
> + xxx="`$run ./try`"
> + case "$?" in
> + 0)
> + case "$xxx" in
> + Yes) cat >&4 <<EOM
> +0.0 is represented as all bits zero in memory
> +EOM
> + d_nv_zero_is_allbits_zero="$define"
> + ;;
> + *) cat >&4 <<EOM
> +0.0 is not represented as all bits zero in memory
> +EOM
> + d_nv_zero_is_allbits_zero="$undef"
> + ;;
> + esac
> + ;;
> + *) cat >&4 <<EOM
> +0.0 is not represented as all bits zero in memory
> +EOM
> + d_nv_zero_is_allbits_zero="$undef"
> + ;;
> + esac
> +fi
> +
> +$rm -f try.* try
> +
>
> : check for off64_t
> echo " "
> @@ -21012,6 +21135,7 @@ d_mymalloc='$d_mymalloc'
> d_nice='$d_nice'
> d_nl_langinfo='$d_nl_langinfo'
> d_nv_preserves_uv='$d_nv_preserves_uv'
> +d_nv_zero_is_allbits_zero='$d_nv_zero_is_allbits_zero'
> d_off64_t='$d_off64_t'
> d_old_pthread_create_joinable='$d_old_pthread_create_joinable'
> d_oldpthreads='$d_oldpthreads'
> --- config_h.SH.orig Thu May 26 12:56:06 2005
> +++ config_h.SH Sat Jun 18 17:31:21 2005
> @@ -3235,6 +3235,10 @@ sed <<!GROK!THIS! >$CONFIG_H -e 's!^#und
> * This symbol contains the number of bits a variable of type NVTYPE
> * can preserve of a variable of type UVTYPE.
> */
> +/* NV_ZERO_IS_ALLBITS_ZERO
> + * This symbol, if defined, indicates that a variable of type NVTYPE
> + * stores 0.0 in memory as all bits zero.
> + */
> #define IVTYPE $ivtype /**/
> #define UVTYPE $uvtype /**/
> #define I8TYPE $i8type /**/
> @@ -3263,6 +3267,7 @@ sed <<!GROK!THIS! >$CONFIG_H -e 's!^#und
> #define NVSIZE $nvsize /**/
> #$d_nv_preserves_uv NV_PRESERVES_UV
> #define NV_PRESERVES_UV_BITS $nv_preserves_uv_bits
> +#$d_nv_zero_is_allbits_zero NV_ZERO_IS_ALLBITS_ZERO
> #if UVSIZE == 8
> # ifdef BYTEORDER
> # if BYTEORDER == 0x1234
> --- Porting/Glossary.orig Fri May 20 09:23:14 2005
> +++ Porting/Glossary Sat Jun 18 17:32:22 2005
> @@ -1383,6 +1383,10 @@ d_nv_preserves_uv (perlxv.U):
> This variable indicates whether a variable of type nvtype
> can preserve all the bits a variable of type uvtype.
>
> +d_nv_zero_is_allbits_zero (perlxv.U):
> + This variable indicates whether a variable of type nvtype
> + stores 0.0 in memory as all bits zero.
> +
> d_off64_t (d_off64_t.U):
> This symbol will be defined if the C compiler supports off64_t.
>
>
--
H.Merijn Brand Amsterdam Perl Mongers (http://amsterdam.pm.org/)
using Perl 5.6.2, 5.8.0, 5.8.5, & 5.9.2 on HP-UX 10.20, 11.00 & 11.11,
AIX 4.3 & 5.2, SuSE 9.2 & 9.3, and Cygwin. http://www.cmve.net/~merijn
Smoking perl: http://www.test-smoke.org, perl QA: http://qa.perl.org
reports to: [EMAIL PROTECTED], perl-qa@perl.org
