From: Tony Cook Date: Tue, 19 Jan 2010 14:07:21 +0000 (+1100) Subject: Fix for #71506: work around possible gcc bug X-Git-Url: http://git.shadowcat.co.uk/gitweb/gitweb.cgi?a=commitdiff_plain;h=275663fa076b91276d12d914795b38d0f554c9e4;p=p5sagit%2Fp5-mst-13.2.git Fix for #71506: work around possible gcc bug When memcpy() is used on a long double pointer with gcc 4.4 in some cases it seems to treat it as a long double assignment, copying only the first 12 bytes. Use unions so the types we're copying into the pack output or from the unpack input are unsigned char[], to avoid the apparent bug. --- diff --git a/pp_pack.c b/pp_pack.c index 03eaf7f..0670548 100644 --- a/pp_pack.c +++ b/pp_pack.c @@ -70,6 +70,18 @@ typedef struct tempsym { (symptr)->previous = NULL; \ } STMT_END +typedef union { + NV nv; + U8 bytes[sizeof(NV)]; +} NV_bytes; + +#if defined(HAS_LONG_DOUBLE) && defined(USE_LONG_DOUBLE) +typedef union { + long double ld; + U8 bytes[sizeof(long double)]; +} ld_bytes; +#endif + #if PERL_VERSION >= 9 # define PERL_PACK_CAN_BYTEORDER # define PERL_PACK_CAN_SHRIEKSIGN @@ -146,17 +158,20 @@ typedef struct tempsym { #define PUSH32(utf8, cur, p) PUSH_BYTES(utf8, cur, OFF32(p), SIZE32) /* Only to be used inside a loop (see the break) */ -#define SHIFT_VAR(utf8, s, strend, var, datumtype) \ +#define SHIFT_BYTES(utf8, s, strend, buf, len, datumtype) \ STMT_START { \ if (utf8) { \ if (!uni_to_bytes(aTHX_ &s, strend, \ - (char *) &var, sizeof(var), datumtype)) break;\ + (char *) (buf), len, datumtype)) break; \ } else { \ - Copy(s, (char *) &var, sizeof(var), char); \ - s += sizeof(var); \ + Copy(s, (char *) (buf), len, char); \ + s += len; \ } \ } STMT_END +#define SHIFT_VAR(utf8, s, strend, var, datumtype) \ + SHIFT_BYTES(utf8, s, strend, &(var), sizeof(var), datumtype) + #define PUSH_VAR(utf8, aptr, var) \ PUSH_BYTES(utf8, aptr, &(var), sizeof(var)) @@ -2085,25 +2100,25 @@ S_unpack_rec(pTHX_ tempsym_t* symptr, const char *s, const char *strbeg, const c break; case 'F': while (len-- > 0) { - NV anv; - SHIFT_VAR(utf8, s, strend, anv, datumtype); - DO_BO_UNPACK_N(anv, NV); + NV_bytes anv; + SHIFT_BYTES(utf8, s, strend, anv.bytes, sizeof(anv.bytes), datumtype); + DO_BO_UNPACK_N(anv.nv, NV); if (!checksum) - mPUSHn(anv); + mPUSHn(anv.nv); else - cdouble += anv; + cdouble += anv.nv; } break; #if defined(HAS_LONG_DOUBLE) && defined(USE_LONG_DOUBLE) case 'D': while (len-- > 0) { - long double aldouble; - SHIFT_VAR(utf8, s, strend, aldouble, datumtype); - DO_BO_UNPACK_N(aldouble, long double); + ld_bytes aldouble; + SHIFT_BYTES(utf8, s, strend, aldouble.bytes, sizeof(aldouble.bytes), datumtype); + DO_BO_UNPACK_N(aldouble.ld, long double); if (!checksum) - mPUSHn(aldouble); + mPUSHn(aldouble.ld); else - cdouble += aldouble; + cdouble += aldouble.ld; } break; #endif @@ -3156,26 +3171,26 @@ extern const double _double_constants[]; } break; case 'F': { - NV anv; + NV_bytes anv; Zero(&anv, 1, NV); /* can be long double with unused bits */ while (len-- > 0) { fromstr = NEXTFROM; - anv = SvNV(fromstr); + anv.nv = SvNV(fromstr); DO_BO_PACK_N(anv, NV); - PUSH_VAR(utf8, cur, anv); + PUSH_BYTES(utf8, cur, anv.bytes, sizeof(anv.bytes)); } break; } #if defined(HAS_LONG_DOUBLE) && defined(USE_LONG_DOUBLE) case 'D': { - long double aldouble; + ld_bytes aldouble; /* long doubles can have unused bits, which may be nonzero */ Zero(&aldouble, 1, long double); while (len-- > 0) { fromstr = NEXTFROM; - aldouble = (long double)SvNV(fromstr); + aldouble.ld = (long double)SvNV(fromstr); DO_BO_PACK_N(aldouble, long double); - PUSH_VAR(utf8, cur, aldouble); + PUSH_BYTES(utf8, cur, aldouble.bytes, sizeof(aldouble.bytes)); } break; }