build: remove various uses of C undefined behavior

If you thought gcc -ansi -pedantic was pedantic, just wait
until you meet clang -fsanitize=undefined.

I think this addresses all the reported "errors", but we'll
need another run to be sure.

all.bash still passes.

Update #5764

Dave, can you please try again?

R=golang-dev, bradfitz
CC=golang-dev
https://golang.org/cl/13334049

Author: rsc

include/bio.h

@@ -79,7 +79,7 @@ struct	Biobuf
 #define	BGETLE2(bp)\
 	((bp)->icount<=-2?((bp)->icount+=2,((bp)->ebuf[(bp)->icount-2])|((bp)->ebuf[(bp)->icount-1]<<8)):Bgetle2((bp)))
 #define	BGETLE4(bp)\
-	((bp)->icount<=-4?((bp)->icount+=4,((bp)->ebuf[(bp)->icount-4])|((bp)->ebuf[(bp)->icount-3]<<8)|((bp)->ebuf[(bp)->icount-2]<<16)|((bp)->ebuf[(bp)->icount-1]<<24)):Bgetle4((bp)))
+	(int)((bp)->icount<=-4?((bp)->icount+=4,((bp)->ebuf[(bp)->icount-4])|((bp)->ebuf[(bp)->icount-3]<<8)|((bp)->ebuf[(bp)->icount-2]<<16)|((uint32)(bp)->ebuf[(bp)->icount-1]<<24)):Bgetle4((bp)))
 
 /*
  * These macros put 1-, 2-, and 4-byte integer values by writing the

src/cmd/cc/com.c

@@ -1325,10 +1325,10 @@ compar(Node *n, int reverse)
 		if(lt->width == 8)
 			hi = big(0, ~0ULL);
 		else
-			hi = big(0, (1LL<<(l->type->width*8))-1);
+			hi = big(0, (1ULL<<(l->type->width*8))-1);
 	}else{
-		lo = big(~0ULL, -(1LL<<(l->type->width*8-1)));
-		hi = big(0, (1LL<<(l->type->width*8-1))-1);
+		lo = big(~0ULL, -(1ULL<<(l->type->width*8-1)));
+		hi = big(0, (1ULL<<(l->type->width*8-1))-1);
 	}
 
 	switch(op){

src/cmd/cc/lex.c

@@ -1019,7 +1019,7 @@ mpatov(char *s, vlong *v)
 			c += 10-'A';
 		else
 			goto bad;
-		nn = n*16 + c;
+		nn = (uvlong)n*16 + c;
 		if(n < 0 && nn >= 0)
 			goto bad;
 		n = nn;

src/cmd/cc/lexbody

@@ -224,7 +224,7 @@ Sym*
 lookup(void)
 {
 	Sym *s;
-	int32 h;
+	uint32 h;
 	char *p;
 	int c, l;
 	char *r, *w;
@@ -400,7 +400,7 @@ l1:
 			if(c >= '0' && c <= '9') {
 				if(c > '7' && c1 == 3)
 					break;
-				yylval.lval <<= c1;
+				yylval.lval = (uvlong)yylval.lval << c1;
 				yylval.lval += c - '0';
 				c = GETC();
 				continue;
@@ -410,7 +410,7 @@ l1:
 			if(c >= 'A' && c <= 'F')
 				c += 'a' - 'A';
 			if(c >= 'a' && c <= 'f') {
-				yylval.lval <<= c1;
+				yylval.lval = (uvlong)yylval.lval << c1;
 				yylval.lval += c - 'a' + 10;
 				c = GETC();
 				continue;
@@ -770,6 +770,6 @@ ieeedtod(Ieee *ieee, double native)
 	f = 65536L;
 	fr = modf(fr*f, &ho);
 	ieee->l = ho;
-	ieee->l <<= 16;
+	ieee->l = (uint32)ieee->l << 16;
 	ieee->l |= (int32)(fr*f);
 }

src/cmd/gc/bv.c

@@ -41,7 +41,7 @@ bvset(Bvec *bv, int32 i)
 
 	if(i < 0 || i >= bv->n)
 		fatal("bvset: index %d is out of bounds with length %d\n", i, bv->n);
-	mask = 1 << (i % WORDBITS);
+	mask = 1U << (i % WORDBITS);
 	bv->b[i / WORDBITS] |= mask;
 }
 

src/cmd/gc/go.h

@@ -78,7 +78,7 @@ typedef	struct	Strlit	Strlit;
 struct	Strlit
 {
 	int32	len;
-	char	s[3];	// variable
+	char	s[1]; // variable
 };
 
 enum

src/cmd/gc/md5.c

@@ -196,7 +196,7 @@ md5block(MD5 *dig, uchar *p, int nn)
 
 		for(i=0; i<16; i++) {
 			j = i*4;
-			X[i] = p[j] | (p[j+1]<<8) | (p[j+2]<<16) | (p[j+3]<<24);
+			X[i] = p[j] | (p[j+1]<<8) | (p[j+2]<<16) | ((uint32)p[j+3]<<24);
 		}
 
 		// Round 1.

src/cmd/gc/mparith2.c

@@ -565,11 +565,11 @@ mpgetfix(Mpint *a)
 		return 0;
 	}
 
-	v = (vlong)a->a[0];
-	v |= (vlong)a->a[1] << Mpscale;
-	v |= (vlong)a->a[2] << (Mpscale+Mpscale);
+	v = (uvlong)a->a[0];
+	v |= (uvlong)a->a[1] << Mpscale;
+	v |= (uvlong)a->a[2] << (Mpscale+Mpscale);
 	if(a->neg)
-		v = -v;
+		v = -(uvlong)v;
 	return v;
 }
 
@@ -586,7 +586,7 @@ mpmovecfix(Mpint *a, vlong c)
 	x = c;
 	if(x < 0) {
 		a->neg = 1;
-		x = -x;
+		x = -(uvlong)x;
 	}
 
 	a1 = &a->a[0];

src/cmd/gc/subr.c

@@ -322,7 +322,7 @@ setlineno(Node *n)
 uint32
 stringhash(char *p)
 {
-	int32 h;
+	uint32 h;
 	int c;
 
 	h = 0;
@@ -333,9 +333,9 @@ stringhash(char *p)
 		h = h*PRIME1 + c;
 	}
 
-	if(h < 0) {
+	if((int32)h < 0) {
 		h = -h;
-		if(h < 0)
+		if((int32)h < 0)
 			h = 0;
 	}
 	return h;

src/cmd/ld/go.c

@@ -37,13 +37,12 @@ static void imported(char *pkg, char *import);
 static int
 hashstr(char *name)
 {
-	int h;
+	uint32 h;
 	char *cp;
 
 	h = 0;
 	for(cp = name; *cp; h += *cp++)
 		h *= 1119;
-	// not if(h < 0) h = ~h, because gcc 4.3 -O2 miscompiles it.
 	h &= 0xffffff;
 	return h;
 }

src/cmd/ld/lib.c

@@ -951,7 +951,7 @@ _lookup(char *symb, int v, int creat)
 {
 	Sym *s;
 	char *p;
-	int32 h;
+	uint32 h;
 	int c;
 
 	h = v;
@@ -1613,7 +1613,7 @@ le16(uchar *b)
 uint32
 le32(uchar *b)
 {
-	return b[0] | b[1]<<8 | b[2]<<16 | b[3]<<24;
+	return b[0] | b[1]<<8 | b[2]<<16 | (uint32)b[3]<<24;
 }
 
 uint64
@@ -1631,7 +1631,7 @@ be16(uchar *b)
 uint32
 be32(uchar *b)
 {
-	return b[0]<<24 | b[1]<<16 | b[2]<<8 | b[3];
+	return (uint32)b[0]<<24 | b[1]<<16 | b[2]<<8 | b[3];
 }
 
 uint64

src/cmd/pack/ar.c

@@ -937,21 +937,12 @@ objsym(Sym *s, void *p)
 int
 hashstr(char *name)
 {
-	int h;
+	uint32 h;
 	char *cp;
 
 	h = 0;
 	for(cp = name; *cp; h += *cp++)
 		h *= 1119;
-	
-	// the code used to say
-	//	if(h < 0)
-	//		h = ~h;
-	// but on gcc 4.3 with -O2 on some systems,
-	// the if(h < 0) gets compiled away as not possible.
-	// use a mask instead, leaving plenty of bits but
-	// definitely not the sign bit.
-
 	return h & 0xfffffff;
 }
 

src/libmach/5obj.c

@@ -130,7 +130,7 @@ addr(Biobuf *bp)
 		BGETC(bp);
 		break;
 	case D_CONST2:
-		BGETLE4(bp);	// fall through
+		Bgetle4(bp); // fall through
 	case D_OREG:
 	case D_CONST:
 	case D_BRANCH:

src/libmach/6obj.c

@@ -134,7 +134,7 @@ addr(Biobuf *bp)
 			off = ((vlong)l << 32) | (off & 0xFFFFFFFF);
 		}
 		if(off < 0)
-			off = -off;
+			off = -(uvlong)off;
 	}
 	if(a.flags & T_SYM)
 		a.sym = BGETC(bp);

src/libmach/8obj.c

@@ -131,7 +131,7 @@ addr(Biobuf *bp)
 			off = -off;
 	}
 	if(a.flags & T_OFFSET2){
-		BGETLE4(bp);
+		Bgetle4(bp);
 	}
 	if(a.flags & T_SYM)
 		a.sym = BGETC(bp);

src/libmach/obj.c

@@ -244,7 +244,7 @@ processprog(Prog *p, int doautos)
 static void
 objlookup(int id, char *name, int type, uint sig)
 {
-	int32 h;
+	uint32 h;
 	char *cp;
 	Sym *s;
 	Symtab *sp;